kernel/trace/bpf_trace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   3  * Copyright (c) 2016 Facebook
   4  */
   5 #include <linux/kernel.h>
   6 #include <linux/types.h>
   7 #include <linux/slab.h>
   8 #include <linux/bpf.h>
   9 #include <linux/bpf_perf_event.h>
  10 #include <linux/btf.h>
  11 #include <linux/filter.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/ctype.h>
  14 #include <linux/kprobes.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/btf_ids.h>
  19 #include <linux/bpf_lsm.h>
  20
  21 #include <net/bpf_sk_storage.h>
  22
  23 #include <uapi/linux/bpf.h>
  24 #include <uapi/linux/btf.h>
  25
  26 #include <asm/tlb.h>
  27
  28 #include "trace_probe.h"
  29 #include "trace.h"
  30
  31 #define CREATE_TRACE_POINTS
  32 #include "bpf_trace.h"
  33
  34 #define bpf_event_rcu_dereference(p)                                    \
  35         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  36
  37 #ifdef CONFIG_MODULES
  38 struct bpf_trace_module {
  39         struct module *module;
  40         struct list_head list;
  41 };
  42
  43 static LIST_HEAD(bpf_trace_modules);
  44 static DEFINE_MUTEX(bpf_module_mutex);
  45
  46 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  47 {
  48         struct bpf_raw_event_map *btp, *ret = NULL;
  49         struct bpf_trace_module *btm;
  50         unsigned int i;
  51
  52         mutex_lock(&bpf_module_mutex);
  53         list_for_each_entry(btm, &bpf_trace_modules, list) {
  54                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  55                         btp = &btm->module->bpf_raw_events[i];
  56                         if (!strcmp(btp->tp->name, name)) {
  57                                 if (try_module_get(btm->module))
  58                                         ret = btp;
  59                                 goto out;
  60                         }
  61                 }
  62         }
  63 out:
  64         mutex_unlock(&bpf_module_mutex);
  65         return ret;
  66 }
  67 #else
  68 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  69 {
  70         return NULL;
  71 }
  72 #endif /* CONFIG_MODULES */
  73
  74 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  75 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  76
  77 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
  78                                   u64 flags, const struct btf **btf,
  79                                   s32 *btf_id);
  80
  81 /**
  82  * trace_call_bpf - invoke BPF program
  83  * @call: tracepoint event
  84  * @ctx: opaque context pointer
  85  *
  86  * kprobe handlers execute BPF programs via this helper.
  87  * Can be used from static tracepoints in the future.
  88  *
  89  * Return: BPF programs always return an integer which is interpreted by
  90  * kprobe handler as:
  91  * 0 - return from kprobe (event is filtered out)
  92  * 1 - store kprobe event into ring buffer
  93  * Other values are reserved and currently alias to 1
  94  */
  95 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
  96 {
  97         unsigned int ret;
  98
  99         cant_sleep();
 100
 101         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 102                 /*
 103                  * since some bpf program is already running on this cpu,
 104                  * don't call into another bpf program (same or different)
 105                  * and don't send kprobe event into ring-buffer,
 106                  * so return zero here
 107                  */
 108                 ret = 0;
 109                 goto out;
 110         }
 111
 112         /*
 113          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 114          * to all call sites, we did a bpf_prog_array_valid() there to check
 115          * whether call->prog_array is empty or not, which is
 116          * a heuristic to speed up execution.
 117          *
 118          * If bpf_prog_array_valid() fetched prog_array was
 119          * non-NULL, we go into trace_call_bpf() and do the actual
 120          * proper rcu_dereference() under RCU lock.
 121          * If it turns out that prog_array is NULL then, we bail out.
 122          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 123          * was NULL, you'll skip the prog_array with the risk of missing
 124          * out of events when it was updated in between this and the
 125          * rcu_dereference() which is accepted risk.
 126          */
 127         ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
 128
 129  out:
 130         __this_cpu_dec(bpf_prog_active);
 131
 132         return ret;
 133 }
 134
 135 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 136 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 137 {
 138         regs_set_return_value(regs, rc);
 139         override_function_with_return(regs);
 140         return 0;
 141 }
 142
 143 static const struct bpf_func_proto bpf_override_return_proto = {
 144         .func           = bpf_override_return,
 145         .gpl_only       = true,
 146         .ret_type       = RET_INTEGER,
 147         .arg1_type      = ARG_PTR_TO_CTX,
 148         .arg2_type      = ARG_ANYTHING,
 149 };
 150 #endif
 151
 152 static __always_inline int
 153 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 154 {
 155         int ret;
 156
 157         ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 158         if (unlikely(ret < 0))
 159                 memset(dst, 0, size);
 160         return ret;
 161 }
 162
 163 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 164            const void __user *, unsafe_ptr)
 165 {
 166         return bpf_probe_read_user_common(dst, size, unsafe_ptr);
 167 }
 168
 169 const struct bpf_func_proto bpf_probe_read_user_proto = {
 170         .func           = bpf_probe_read_user,
 171         .gpl_only       = true,
 172         .ret_type       = RET_INTEGER,
 173         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 174         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 175         .arg3_type      = ARG_ANYTHING,
 176 };
 177
 178 static __always_inline int
 179 bpf_probe_read_user_str_common(void *dst, u32 size,
 180                                const void __user *unsafe_ptr)
 181 {
 182         int ret;
 183
 184         /*
 185          * NB: We rely on strncpy_from_user() not copying junk past the NUL
 186          * terminator into `dst`.
 187          *
 188          * strncpy_from_user() does long-sized strides in the fast path. If the
 189          * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
 190          * then there could be junk after the NUL in `dst`. If user takes `dst`
 191          * and keys a hash map with it, then semantically identical strings can
 192          * occupy multiple entries in the map.
 193          */
 194         ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 195         if (unlikely(ret < 0))
 196                 memset(dst, 0, size);
 197         return ret;
 198 }
 199
 200 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 201            const void __user *, unsafe_ptr)
 202 {
 203         return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
 204 }
 205
 206 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 207         .func           = bpf_probe_read_user_str,
 208         .gpl_only       = true,
 209         .ret_type       = RET_INTEGER,
 210         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 211         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 212         .arg3_type      = ARG_ANYTHING,
 213 };
 214
 215 static __always_inline int
 216 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
 217 {
 218         int ret = security_locked_down(LOCKDOWN_BPF_READ);
 219
 220         if (unlikely(ret < 0))
 221                 goto fail;
 222         ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
 223         if (unlikely(ret < 0))
 224                 goto fail;
 225         return ret;
 226 fail:
 227         memset(dst, 0, size);
 228         return ret;
 229 }
 230
 231 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 232            const void *, unsafe_ptr)
 233 {
 234         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 235 }
 236
 237 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 238         .func           = bpf_probe_read_kernel,
 239         .gpl_only       = true,
 240         .ret_type       = RET_INTEGER,
 241         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 242         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 243         .arg3_type      = ARG_ANYTHING,
 244 };
 245
 246 static __always_inline int
 247 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
 248 {
 249         int ret = security_locked_down(LOCKDOWN_BPF_READ);
 250
 251         if (unlikely(ret < 0))
 252                 goto fail;
 253
 254         /*
 255          * The strncpy_from_kernel_nofault() call will likely not fill the
 256          * entire buffer, but that's okay in this circumstance as we're probing
 257          * arbitrary memory anyway similar to bpf_probe_read_*() and might
 258          * as well probe the stack. Thus, memory is explicitly cleared
 259          * only in error case, so that improper users ignoring return
 260          * code altogether don't copy garbage; otherwise length of string
 261          * is returned that can be used for bpf_perf_event_output() et al.
 262          */
 263         ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
 264         if (unlikely(ret < 0))
 265                 goto fail;
 266
 267         return ret;
 268 fail:
 269         memset(dst, 0, size);
 270         return ret;
 271 }
 272
 273 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 274            const void *, unsafe_ptr)
 275 {
 276         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 277 }
 278
 279 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 280         .func           = bpf_probe_read_kernel_str,
 281         .gpl_only       = true,
 282         .ret_type       = RET_INTEGER,
 283         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 284         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 285         .arg3_type      = ARG_ANYTHING,
 286 };
 287
 288 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 289 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 290            const void *, unsafe_ptr)
 291 {
 292         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 293                 return bpf_probe_read_user_common(dst, size,
 294                                 (__force void __user *)unsafe_ptr);
 295         }
 296         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 297 }
 298
 299 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 300         .func           = bpf_probe_read_compat,
 301         .gpl_only       = true,
 302         .ret_type       = RET_INTEGER,
 303         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 304         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 305         .arg3_type      = ARG_ANYTHING,
 306 };
 307
 308 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 309            const void *, unsafe_ptr)
 310 {
 311         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 312                 return bpf_probe_read_user_str_common(dst, size,
 313                                 (__force void __user *)unsafe_ptr);
 314         }
 315         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 316 }
 317
 318 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 319         .func           = bpf_probe_read_compat_str,
 320         .gpl_only       = true,
 321         .ret_type       = RET_INTEGER,
 322         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 323         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 324         .arg3_type      = ARG_ANYTHING,
 325 };
 326 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 327
 328 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 329            u32, size)
 330 {
 331         /*
 332          * Ensure we're in user context which is safe for the helper to
 333          * run. This helper has no business in a kthread.
 334          *
 335          * access_ok() should prevent writing to non-user memory, but in
 336          * some situations (nommu, temporary switch, etc) access_ok() does
 337          * not provide enough validation, hence the check on KERNEL_DS.
 338          *
 339          * nmi_uaccess_okay() ensures the probe is not run in an interim
 340          * state, when the task or mm are switched. This is specifically
 341          * required to prevent the use of temporary mm.
 342          */
 343
 344         if (unlikely(in_interrupt() ||
 345                      current->flags & (PF_KTHREAD | PF_EXITING)))
 346                 return -EPERM;
 347         if (unlikely(uaccess_kernel()))
 348                 return -EPERM;
 349         if (unlikely(!nmi_uaccess_okay()))
 350                 return -EPERM;
 351
 352         return copy_to_user_nofault(unsafe_ptr, src, size);
 353 }
 354
 355 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 356         .func           = bpf_probe_write_user,
 357         .gpl_only       = true,
 358         .ret_type       = RET_INTEGER,
 359         .arg1_type      = ARG_ANYTHING,
 360         .arg2_type      = ARG_PTR_TO_MEM,
 361         .arg3_type      = ARG_CONST_SIZE,
 362 };
 363
 364 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 365 {
 366         if (!capable(CAP_SYS_ADMIN))
 367                 return NULL;
 368
 369         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 370                             current->comm, task_pid_nr(current));
 371
 372         return &bpf_probe_write_user_proto;
 373 }
 374
 375 static void bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
 376                 size_t bufsz)
 377 {
 378         void __user *user_ptr = (__force void __user *)unsafe_ptr;
 379
 380         buf[0] = 0;
 381
 382         switch (fmt_ptype) {
 383         case 's':
 384 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 385                 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 386                         strncpy_from_user_nofault(buf, user_ptr, bufsz);
 387                         break;
 388                 }
 389                 fallthrough;
 390 #endif
 391         case 'k':
 392                 strncpy_from_kernel_nofault(buf, unsafe_ptr, bufsz);
 393                 break;
 394         case 'u':
 395                 strncpy_from_user_nofault(buf, user_ptr, bufsz);
 396                 break;
 397         }
 398 }
 399
 400 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
 401
 402 #define BPF_TRACE_PRINTK_SIZE   1024
 403
 404 static __printf(1, 0) int bpf_do_trace_printk(const char *fmt, ...)
 405 {
 406         static char buf[BPF_TRACE_PRINTK_SIZE];
 407         unsigned long flags;
 408         va_list ap;
 409         int ret;
 410
 411         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 412         va_start(ap, fmt);
 413         ret = vsnprintf(buf, sizeof(buf), fmt, ap);
 414         va_end(ap);
 415         /* vsnprintf() will not append null for zero-length strings */
 416         if (ret == 0)
 417                 buf[0] = '\0';
 418         trace_bpf_trace_printk(buf);
 419         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 420
 421         return ret;
 422 }
 423
 424 /*
 425  * Only limited trace_printk() conversion specifiers allowed:
 426  * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %pB %pks %pus %s
 427  */
 428 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 429            u64, arg2, u64, arg3)
 430 {
 431         int i, mod[3] = {}, fmt_cnt = 0;
 432         char buf[64], fmt_ptype;
 433         void *unsafe_ptr = NULL;
 434         bool str_seen = false;
 435
 436         /*
 437          * bpf_check()->check_func_arg()->check_stack_boundary()
 438          * guarantees that fmt points to bpf program stack,
 439          * fmt_size bytes of it were initialized and fmt_size > 0
 440          */
 441         if (fmt[--fmt_size] != 0)
 442                 return -EINVAL;
 443
 444         /* check format string for allowed specifiers */
 445         for (i = 0; i < fmt_size; i++) {
 446                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 447                         return -EINVAL;
 448
 449                 if (fmt[i] != '%')
 450                         continue;
 451
 452                 if (fmt_cnt >= 3)
 453                         return -EINVAL;
 454
 455                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 456                 i++;
 457                 if (fmt[i] == 'l') {
 458                         mod[fmt_cnt]++;
 459                         i++;
 460                 } else if (fmt[i] == 'p') {
 461                         mod[fmt_cnt]++;
 462                         if ((fmt[i + 1] == 'k' ||
 463                              fmt[i + 1] == 'u') &&
 464                             fmt[i + 2] == 's') {
 465                                 fmt_ptype = fmt[i + 1];
 466                                 i += 2;
 467                                 goto fmt_str;
 468                         }
 469
 470                         if (fmt[i + 1] == 'B') {
 471                                 i++;
 472                                 goto fmt_next;
 473                         }
 474
 475                         /* disallow any further format extensions */
 476                         if (fmt[i + 1] != 0 &&
 477                             !isspace(fmt[i + 1]) &&
 478                             !ispunct(fmt[i + 1]))
 479                                 return -EINVAL;
 480
 481                         goto fmt_next;
 482                 } else if (fmt[i] == 's') {
 483                         mod[fmt_cnt]++;
 484                         fmt_ptype = fmt[i];
 485 fmt_str:
 486                         if (str_seen)
 487                                 /* allow only one '%s' per fmt string */
 488                                 return -EINVAL;
 489                         str_seen = true;
 490
 491                         if (fmt[i + 1] != 0 &&
 492                             !isspace(fmt[i + 1]) &&
 493                             !ispunct(fmt[i + 1]))
 494                                 return -EINVAL;
 495
 496                         switch (fmt_cnt) {
 497                         case 0:
 498                                 unsafe_ptr = (void *)(long)arg1;
 499                                 arg1 = (long)buf;
 500                                 break;
 501                         case 1:
 502                                 unsafe_ptr = (void *)(long)arg2;
 503                                 arg2 = (long)buf;
 504                                 break;
 505                         case 2:
 506                                 unsafe_ptr = (void *)(long)arg3;
 507                                 arg3 = (long)buf;
 508                                 break;
 509                         }
 510
 511                         bpf_trace_copy_string(buf, unsafe_ptr, fmt_ptype,
 512                                         sizeof(buf));
 513                         goto fmt_next;
 514                 }
 515
 516                 if (fmt[i] == 'l') {
 517                         mod[fmt_cnt]++;
 518                         i++;
 519                 }
 520
 521                 if (fmt[i] != 'i' && fmt[i] != 'd' &&
 522                     fmt[i] != 'u' && fmt[i] != 'x')
 523                         return -EINVAL;
 524 fmt_next:
 525                 fmt_cnt++;
 526         }
 527
 528 /* Horrid workaround for getting va_list handling working with different
 529  * argument type combinations generically for 32 and 64 bit archs.
 530  */
 531 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
 532 #define __BPF_TP(...)                                                   \
 533         bpf_do_trace_printk(fmt, ##__VA_ARGS__)
 534
 535 #define __BPF_ARG1_TP(...)                                              \
 536         ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64))        \
 537           ? __BPF_TP(arg1, ##__VA_ARGS__)                               \
 538           : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32))    \
 539               ? __BPF_TP((long)arg1, ##__VA_ARGS__)                     \
 540               : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
 541
 542 #define __BPF_ARG2_TP(...)                                              \
 543         ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64))        \
 544           ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__)                          \
 545           : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32))    \
 546               ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__)                \
 547               : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
 548
 549 #define __BPF_ARG3_TP(...)                                              \
 550         ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64))        \
 551           ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__)                          \
 552           : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32))    \
 553               ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__)                \
 554               : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
 555
 556         return __BPF_TP_EMIT();
 557 }
 558
 559 static const struct bpf_func_proto bpf_trace_printk_proto = {
 560         .func           = bpf_trace_printk,
 561         .gpl_only       = true,
 562         .ret_type       = RET_INTEGER,
 563         .arg1_type      = ARG_PTR_TO_MEM,
 564         .arg2_type      = ARG_CONST_SIZE,
 565 };
 566
 567 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 568 {
 569         /*
 570          * This program might be calling bpf_trace_printk,
 571          * so enable the associated bpf_trace/bpf_trace_printk event.
 572          * Repeat this each time as it is possible a user has
 573          * disabled bpf_trace_printk events.  By loading a program
 574          * calling bpf_trace_printk() however the user has expressed
 575          * the intent to see such events.
 576          */
 577         if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 578                 pr_warn_ratelimited("could not enable bpf_trace_printk events");
 579
 580         return &bpf_trace_printk_proto;
 581 }
 582
 583 #define MAX_SEQ_PRINTF_VARARGS          12
 584 #define MAX_SEQ_PRINTF_MAX_MEMCPY       6
 585 #define MAX_SEQ_PRINTF_STR_LEN          128
 586
 587 struct bpf_seq_printf_buf {
 588         char buf[MAX_SEQ_PRINTF_MAX_MEMCPY][MAX_SEQ_PRINTF_STR_LEN];
 589 };
 590 static DEFINE_PER_CPU(struct bpf_seq_printf_buf, bpf_seq_printf_buf);
 591 static DEFINE_PER_CPU(int, bpf_seq_printf_buf_used);
 592
 593 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 594            const void *, data, u32, data_len)
 595 {
 596         int err = -EINVAL, fmt_cnt = 0, memcpy_cnt = 0;
 597         int i, buf_used, copy_size, num_args;
 598         u64 params[MAX_SEQ_PRINTF_VARARGS];
 599         struct bpf_seq_printf_buf *bufs;
 600         const u64 *args = data;
 601
 602         buf_used = this_cpu_inc_return(bpf_seq_printf_buf_used);
 603         if (WARN_ON_ONCE(buf_used > 1)) {
 604                 err = -EBUSY;
 605                 goto out;
 606         }
 607
 608         bufs = this_cpu_ptr(&bpf_seq_printf_buf);
 609
 610         /*
 611          * bpf_check()->check_func_arg()->check_stack_boundary()
 612          * guarantees that fmt points to bpf program stack,
 613          * fmt_size bytes of it were initialized and fmt_size > 0
 614          */
 615         if (fmt[--fmt_size] != 0)
 616                 goto out;
 617
 618         if (data_len & 7)
 619                 goto out;
 620
 621         for (i = 0; i < fmt_size; i++) {
 622                 if (fmt[i] == '%') {
 623                         if (fmt[i + 1] == '%')
 624                                 i++;
 625                         else if (!data || !data_len)
 626                                 goto out;
 627                 }
 628         }
 629
 630         num_args = data_len / 8;
 631
 632         /* check format string for allowed specifiers */
 633         for (i = 0; i < fmt_size; i++) {
 634                 /* only printable ascii for now. */
 635                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) {
 636                         err = -EINVAL;
 637                         goto out;
 638                 }
 639
 640                 if (fmt[i] != '%')
 641                         continue;
 642
 643                 if (fmt[i + 1] == '%') {
 644                         i++;
 645                         continue;
 646                 }
 647
 648                 if (fmt_cnt >= MAX_SEQ_PRINTF_VARARGS) {
 649                         err = -E2BIG;
 650                         goto out;
 651                 }
 652
 653                 if (fmt_cnt >= num_args) {
 654                         err = -EINVAL;
 655                         goto out;
 656                 }
 657
 658                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 659                 i++;
 660
 661                 /* skip optional "[0 +-][num]" width formating field */
 662                 while (fmt[i] == '0' || fmt[i] == '+'  || fmt[i] == '-' ||
 663                        fmt[i] == ' ')
 664                         i++;
 665                 if (fmt[i] >= '1' && fmt[i] <= '9') {
 666                         i++;
 667                         while (fmt[i] >= '0' && fmt[i] <= '9')
 668                                 i++;
 669                 }
 670
 671                 if (fmt[i] == 's') {
 672                         void *unsafe_ptr;
 673
 674                         /* try our best to copy */
 675                         if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
 676                                 err = -E2BIG;
 677                                 goto out;
 678                         }
 679
 680                         unsafe_ptr = (void *)(long)args[fmt_cnt];
 681                         err = strncpy_from_kernel_nofault(bufs->buf[memcpy_cnt],
 682                                         unsafe_ptr, MAX_SEQ_PRINTF_STR_LEN);
 683                         if (err < 0)
 684                                 bufs->buf[memcpy_cnt][0] = '\0';
 685                         params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
 686
 687                         fmt_cnt++;
 688                         memcpy_cnt++;
 689                         continue;
 690                 }
 691
 692                 if (fmt[i] == 'p') {
 693                         if (fmt[i + 1] == 0 ||
 694                             fmt[i + 1] == 'K' ||
 695                             fmt[i + 1] == 'x' ||
 696                             fmt[i + 1] == 'B') {
 697                                 /* just kernel pointers */
 698                                 params[fmt_cnt] = args[fmt_cnt];
 699                                 fmt_cnt++;
 700                                 continue;
 701                         }
 702
 703                         /* only support "%pI4", "%pi4", "%pI6" and "%pi6". */
 704                         if (fmt[i + 1] != 'i' && fmt[i + 1] != 'I') {
 705                                 err = -EINVAL;
 706                                 goto out;
 707                         }
 708                         if (fmt[i + 2] != '4' && fmt[i + 2] != '6') {
 709                                 err = -EINVAL;
 710                                 goto out;
 711                         }
 712
 713                         if (memcpy_cnt >= MAX_SEQ_PRINTF_MAX_MEMCPY) {
 714                                 err = -E2BIG;
 715                                 goto out;
 716                         }
 717
 718
 719                         copy_size = (fmt[i + 2] == '4') ? 4 : 16;
 720
 721                         err = copy_from_kernel_nofault(bufs->buf[memcpy_cnt],
 722                                                 (void *) (long) args[fmt_cnt],
 723                                                 copy_size);
 724                         if (err < 0)
 725                                 memset(bufs->buf[memcpy_cnt], 0, copy_size);
 726                         params[fmt_cnt] = (u64)(long)bufs->buf[memcpy_cnt];
 727
 728                         i += 2;
 729                         fmt_cnt++;
 730                         memcpy_cnt++;
 731                         continue;
 732                 }
 733
 734                 if (fmt[i] == 'l') {
 735                         i++;
 736                         if (fmt[i] == 'l')
 737                                 i++;
 738                 }
 739
 740                 if (fmt[i] != 'i' && fmt[i] != 'd' &&
 741                     fmt[i] != 'u' && fmt[i] != 'x' &&
 742                     fmt[i] != 'X') {
 743                         err = -EINVAL;
 744                         goto out;
 745                 }
 746
 747                 params[fmt_cnt] = args[fmt_cnt];
 748                 fmt_cnt++;
 749         }
 750
 751         /* Maximumly we can have MAX_SEQ_PRINTF_VARARGS parameter, just give
 752          * all of them to seq_printf().
 753          */
 754         seq_printf(m, fmt, params[0], params[1], params[2], params[3],
 755                    params[4], params[5], params[6], params[7], params[8],
 756                    params[9], params[10], params[11]);
 757
 758         err = seq_has_overflowed(m) ? -EOVERFLOW : 0;
 759 out:
 760         this_cpu_dec(bpf_seq_printf_buf_used);
 761         return err;
 762 }
 763
 764 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
 765
 766 static const struct bpf_func_proto bpf_seq_printf_proto = {
 767         .func           = bpf_seq_printf,
 768         .gpl_only       = true,
 769         .ret_type       = RET_INTEGER,
 770         .arg1_type      = ARG_PTR_TO_BTF_ID,
 771         .arg1_btf_id    = &btf_seq_file_ids[0],
 772         .arg2_type      = ARG_PTR_TO_MEM,
 773         .arg3_type      = ARG_CONST_SIZE,
 774         .arg4_type      = ARG_PTR_TO_MEM_OR_NULL,
 775         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 776 };
 777
 778 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
 779 {
 780         return seq_write(m, data, len) ? -EOVERFLOW : 0;
 781 }
 782
 783 static const struct bpf_func_proto bpf_seq_write_proto = {
 784         .func           = bpf_seq_write,
 785         .gpl_only       = true,
 786         .ret_type       = RET_INTEGER,
 787         .arg1_type      = ARG_PTR_TO_BTF_ID,
 788         .arg1_btf_id    = &btf_seq_file_ids[0],
 789         .arg2_type      = ARG_PTR_TO_MEM,
 790         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 791 };
 792
 793 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
 794            u32, btf_ptr_size, u64, flags)
 795 {
 796         const struct btf *btf;
 797         s32 btf_id;
 798         int ret;
 799
 800         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 801         if (ret)
 802                 return ret;
 803
 804         return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
 805 }
 806
 807 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
 808         .func           = bpf_seq_printf_btf,
 809         .gpl_only       = true,
 810         .ret_type       = RET_INTEGER,
 811         .arg1_type      = ARG_PTR_TO_BTF_ID,
 812         .arg1_btf_id    = &btf_seq_file_ids[0],
 813         .arg2_type      = ARG_PTR_TO_MEM,
 814         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 815         .arg4_type      = ARG_ANYTHING,
 816 };
 817
 818 static __always_inline int
 819 get_map_perf_counter(struct bpf_map *map, u64 flags,
 820                      u64 *value, u64 *enabled, u64 *running)
 821 {
 822         struct bpf_array *array = container_of(map, struct bpf_array, map);
 823         unsigned int cpu = smp_processor_id();
 824         u64 index = flags & BPF_F_INDEX_MASK;
 825         struct bpf_event_entry *ee;
 826
 827         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 828                 return -EINVAL;
 829         if (index == BPF_F_CURRENT_CPU)
 830                 index = cpu;
 831         if (unlikely(index >= array->map.max_entries))
 832                 return -E2BIG;
 833
 834         ee = READ_ONCE(array->ptrs[index]);
 835         if (!ee)
 836                 return -ENOENT;
 837
 838         return perf_event_read_local(ee->event, value, enabled, running);
 839 }
 840
 841 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 842 {
 843         u64 value = 0;
 844         int err;
 845
 846         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 847         /*
 848          * this api is ugly since we miss [-22..-2] range of valid
 849          * counter values, but that's uapi
 850          */
 851         if (err)
 852                 return err;
 853         return value;
 854 }
 855
 856 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 857         .func           = bpf_perf_event_read,
 858         .gpl_only       = true,
 859         .ret_type       = RET_INTEGER,
 860         .arg1_type      = ARG_CONST_MAP_PTR,
 861         .arg2_type      = ARG_ANYTHING,
 862 };
 863
 864 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 865            struct bpf_perf_event_value *, buf, u32, size)
 866 {
 867         int err = -EINVAL;
 868
 869         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 870                 goto clear;
 871         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 872                                    &buf->running);
 873         if (unlikely(err))
 874                 goto clear;
 875         return 0;
 876 clear:
 877         memset(buf, 0, size);
 878         return err;
 879 }
 880
 881 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 882         .func           = bpf_perf_event_read_value,
 883         .gpl_only       = true,
 884         .ret_type       = RET_INTEGER,
 885         .arg1_type      = ARG_CONST_MAP_PTR,
 886         .arg2_type      = ARG_ANYTHING,
 887         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 888         .arg4_type      = ARG_CONST_SIZE,
 889 };
 890
 891 static __always_inline u64
 892 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 893                         u64 flags, struct perf_sample_data *sd)
 894 {
 895         struct bpf_array *array = container_of(map, struct bpf_array, map);
 896         unsigned int cpu = smp_processor_id();
 897         u64 index = flags & BPF_F_INDEX_MASK;
 898         struct bpf_event_entry *ee;
 899         struct perf_event *event;
 900
 901         if (index == BPF_F_CURRENT_CPU)
 902                 index = cpu;
 903         if (unlikely(index >= array->map.max_entries))
 904                 return -E2BIG;
 905
 906         ee = READ_ONCE(array->ptrs[index]);
 907         if (!ee)
 908                 return -ENOENT;
 909
 910         event = ee->event;
 911         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 912                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 913                 return -EINVAL;
 914
 915         if (unlikely(event->oncpu != cpu))
 916                 return -EOPNOTSUPP;
 917
 918         return perf_event_output(event, sd, regs);
 919 }
 920
 921 /*
 922  * Support executing tracepoints in normal, irq, and nmi context that each call
 923  * bpf_perf_event_output
 924  */
 925 struct bpf_trace_sample_data {
 926         struct perf_sample_data sds[3];
 927 };
 928
 929 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 930 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 931 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 932            u64, flags, void *, data, u64, size)
 933 {
 934         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 935         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 936         struct perf_raw_record raw = {
 937                 .frag = {
 938                         .size = size,
 939                         .data = data,
 940                 },
 941         };
 942         struct perf_sample_data *sd;
 943         int err;
 944
 945         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 946                 err = -EBUSY;
 947                 goto out;
 948         }
 949
 950         sd = &sds->sds[nest_level - 1];
 951
 952         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 953                 err = -EINVAL;
 954                 goto out;
 955         }
 956
 957         perf_sample_data_init(sd, 0, 0);
 958         sd->raw = &raw;
 959
 960         err = __bpf_perf_event_output(regs, map, flags, sd);
 961
 962 out:
 963         this_cpu_dec(bpf_trace_nest_level);
 964         return err;
 965 }
 966
 967 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 968         .func           = bpf_perf_event_output,
 969         .gpl_only       = true,
 970         .ret_type       = RET_INTEGER,
 971         .arg1_type      = ARG_PTR_TO_CTX,
 972         .arg2_type      = ARG_CONST_MAP_PTR,
 973         .arg3_type      = ARG_ANYTHING,
 974         .arg4_type      = ARG_PTR_TO_MEM,
 975         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 976 };
 977
 978 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 979 struct bpf_nested_pt_regs {
 980         struct pt_regs regs[3];
 981 };
 982 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 983 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 984
 985 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 986                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 987 {
 988         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 989         struct perf_raw_frag frag = {
 990                 .copy           = ctx_copy,
 991                 .size           = ctx_size,
 992                 .data           = ctx,
 993         };
 994         struct perf_raw_record raw = {
 995                 .frag = {
 996                         {
 997                                 .next   = ctx_size ? &frag : NULL,
 998                         },
 999                         .size   = meta_size,
1000                         .data   = meta,
1001                 },
1002         };
1003         struct perf_sample_data *sd;
1004         struct pt_regs *regs;
1005         u64 ret;
1006
1007         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
1008                 ret = -EBUSY;
1009                 goto out;
1010         }
1011         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
1012         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
1013
1014         perf_fetch_caller_regs(regs);
1015         perf_sample_data_init(sd, 0, 0);
1016         sd->raw = &raw;
1017
1018         ret = __bpf_perf_event_output(regs, map, flags, sd);
1019 out:
1020         this_cpu_dec(bpf_event_output_nest_level);
1021         return ret;
1022 }
1023
1024 BPF_CALL_0(bpf_get_current_task)
1025 {
1026         return (long) current;
1027 }
1028
1029 const struct bpf_func_proto bpf_get_current_task_proto = {
1030         .func           = bpf_get_current_task,
1031         .gpl_only       = true,
1032         .ret_type       = RET_INTEGER,
1033 };
1034
1035 BPF_CALL_0(bpf_get_current_task_btf)
1036 {
1037         return (unsigned long) current;
1038 }
1039
1040 BTF_ID_LIST_SINGLE(bpf_get_current_btf_ids, struct, task_struct)
1041
1042 static const struct bpf_func_proto bpf_get_current_task_btf_proto = {
1043         .func           = bpf_get_current_task_btf,
1044         .gpl_only       = true,
1045         .ret_type       = RET_PTR_TO_BTF_ID,
1046         .ret_btf_id     = &bpf_get_current_btf_ids[0],
1047 };
1048
1049 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
1050 {
1051         struct bpf_array *array = container_of(map, struct bpf_array, map);
1052         struct cgroup *cgrp;
1053
1054         if (unlikely(idx >= array->map.max_entries))
1055                 return -E2BIG;
1056
1057         cgrp = READ_ONCE(array->ptrs[idx]);
1058         if (unlikely(!cgrp))
1059                 return -EAGAIN;
1060
1061         return task_under_cgroup_hierarchy(current, cgrp);
1062 }
1063
1064 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
1065         .func           = bpf_current_task_under_cgroup,
1066         .gpl_only       = false,
1067         .ret_type       = RET_INTEGER,
1068         .arg1_type      = ARG_CONST_MAP_PTR,
1069         .arg2_type      = ARG_ANYTHING,
1070 };
1071
1072 struct send_signal_irq_work {
1073         struct irq_work irq_work;
1074         struct task_struct *task;
1075         u32 sig;
1076         enum pid_type type;
1077 };
1078
1079 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
1080
1081 static void do_bpf_send_signal(struct irq_work *entry)
1082 {
1083         struct send_signal_irq_work *work;
1084
1085         work = container_of(entry, struct send_signal_irq_work, irq_work);
1086         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
1087 }
1088
1089 static int bpf_send_signal_common(u32 sig, enum pid_type type)
1090 {
1091         struct send_signal_irq_work *work = NULL;
1092
1093         /* Similar to bpf_probe_write_user, task needs to be
1094          * in a sound condition and kernel memory access be
1095          * permitted in order to send signal to the current
1096          * task.
1097          */
1098         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
1099                 return -EPERM;
1100         if (unlikely(uaccess_kernel()))
1101                 return -EPERM;
1102         if (unlikely(!nmi_uaccess_okay()))
1103                 return -EPERM;
1104
1105         if (irqs_disabled()) {
1106                 /* Do an early check on signal validity. Otherwise,
1107                  * the error is lost in deferred irq_work.
1108                  */
1109                 if (unlikely(!valid_signal(sig)))
1110                         return -EINVAL;
1111
1112                 work = this_cpu_ptr(&send_signal_work);
1113                 if (irq_work_is_busy(&work->irq_work))
1114                         return -EBUSY;
1115
1116                 /* Add the current task, which is the target of sending signal,
1117                  * to the irq_work. The current task may change when queued
1118                  * irq works get executed.
1119                  */
1120                 work->task = current;
1121                 work->sig = sig;
1122                 work->type = type;
1123                 irq_work_queue(&work->irq_work);
1124                 return 0;
1125         }
1126
1127         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
1128 }
1129
1130 BPF_CALL_1(bpf_send_signal, u32, sig)
1131 {
1132         return bpf_send_signal_common(sig, PIDTYPE_TGID);
1133 }
1134
1135 static const struct bpf_func_proto bpf_send_signal_proto = {
1136         .func           = bpf_send_signal,
1137         .gpl_only       = false,
1138         .ret_type       = RET_INTEGER,
1139         .arg1_type      = ARG_ANYTHING,
1140 };
1141
1142 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
1143 {
1144         return bpf_send_signal_common(sig, PIDTYPE_PID);
1145 }
1146
1147 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
1148         .func           = bpf_send_signal_thread,
1149         .gpl_only       = false,
1150         .ret_type       = RET_INTEGER,
1151         .arg1_type      = ARG_ANYTHING,
1152 };
1153
1154 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
1155 {
1156         long len;
1157         char *p;
1158
1159         if (!sz)
1160                 return 0;
1161
1162         p = d_path(path, buf, sz);
1163         if (IS_ERR(p)) {
1164                 len = PTR_ERR(p);
1165         } else {
1166                 len = buf + sz - p;
1167                 memmove(buf, p, len);
1168         }
1169
1170         return len;
1171 }
1172
1173 BTF_SET_START(btf_allowlist_d_path)
1174 #ifdef CONFIG_SECURITY
1175 BTF_ID(func, security_file_permission)
1176 BTF_ID(func, security_inode_getattr)
1177 BTF_ID(func, security_file_open)
1178 #endif
1179 #ifdef CONFIG_SECURITY_PATH
1180 BTF_ID(func, security_path_truncate)
1181 #endif
1182 BTF_ID(func, vfs_truncate)
1183 BTF_ID(func, vfs_fallocate)
1184 BTF_ID(func, dentry_open)
1185 BTF_ID(func, vfs_getattr)
1186 BTF_ID(func, filp_close)
1187 BTF_SET_END(btf_allowlist_d_path)
1188
1189 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
1190 {
1191         if (prog->type == BPF_PROG_TYPE_TRACING &&
1192             prog->expected_attach_type == BPF_TRACE_ITER)
1193                 return true;
1194
1195         if (prog->type == BPF_PROG_TYPE_LSM)
1196                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
1197
1198         return btf_id_set_contains(&btf_allowlist_d_path,
1199                                    prog->aux->attach_btf_id);
1200 }
1201
1202 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
1203
1204 static const struct bpf_func_proto bpf_d_path_proto = {
1205         .func           = bpf_d_path,
1206         .gpl_only       = false,
1207         .ret_type       = RET_INTEGER,
1208         .arg1_type      = ARG_PTR_TO_BTF_ID,
1209         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
1210         .arg2_type      = ARG_PTR_TO_MEM,
1211         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1212         .allowed        = bpf_d_path_allowed,
1213 };
1214
1215 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
1216                          BTF_F_PTR_RAW | BTF_F_ZERO)
1217
1218 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
1219                                   u64 flags, const struct btf **btf,
1220                                   s32 *btf_id)
1221 {
1222         const struct btf_type *t;
1223
1224         if (unlikely(flags & ~(BTF_F_ALL)))
1225                 return -EINVAL;
1226
1227         if (btf_ptr_size != sizeof(struct btf_ptr))
1228                 return -EINVAL;
1229
1230         *btf = bpf_get_btf_vmlinux();
1231
1232         if (IS_ERR_OR_NULL(*btf))
1233                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
1234
1235         if (ptr->type_id > 0)
1236                 *btf_id = ptr->type_id;
1237         else
1238                 return -EINVAL;
1239
1240         if (*btf_id > 0)
1241                 t = btf_type_by_id(*btf, *btf_id);
1242         if (*btf_id <= 0 || !t)
1243                 return -ENOENT;
1244
1245         return 0;
1246 }
1247
1248 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1249            u32, btf_ptr_size, u64, flags)
1250 {
1251         const struct btf *btf;
1252         s32 btf_id;
1253         int ret;
1254
1255         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1256         if (ret)
1257                 return ret;
1258
1259         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1260                                       flags);
1261 }
1262
1263 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1264         .func           = bpf_snprintf_btf,
1265         .gpl_only       = false,
1266         .ret_type       = RET_INTEGER,
1267         .arg1_type      = ARG_PTR_TO_MEM,
1268         .arg2_type      = ARG_CONST_SIZE,
1269         .arg3_type      = ARG_PTR_TO_MEM,
1270         .arg4_type      = ARG_CONST_SIZE,
1271         .arg5_type      = ARG_ANYTHING,
1272 };
1273
1274 const struct bpf_func_proto *
1275 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1276 {
1277         switch (func_id) {
1278         case BPF_FUNC_map_lookup_elem:
1279                 return &bpf_map_lookup_elem_proto;
1280         case BPF_FUNC_map_update_elem:
1281                 return &bpf_map_update_elem_proto;
1282         case BPF_FUNC_map_delete_elem:
1283                 return &bpf_map_delete_elem_proto;
1284         case BPF_FUNC_map_push_elem:
1285                 return &bpf_map_push_elem_proto;
1286         case BPF_FUNC_map_pop_elem:
1287                 return &bpf_map_pop_elem_proto;
1288         case BPF_FUNC_map_peek_elem:
1289                 return &bpf_map_peek_elem_proto;
1290         case BPF_FUNC_ktime_get_ns:
1291                 return &bpf_ktime_get_ns_proto;
1292         case BPF_FUNC_ktime_get_boot_ns:
1293                 return &bpf_ktime_get_boot_ns_proto;
1294         case BPF_FUNC_ktime_get_coarse_ns:
1295                 return &bpf_ktime_get_coarse_ns_proto;
1296         case BPF_FUNC_tail_call:
1297                 return &bpf_tail_call_proto;
1298         case BPF_FUNC_get_current_pid_tgid:
1299                 return &bpf_get_current_pid_tgid_proto;
1300         case BPF_FUNC_get_current_task:
1301                 return &bpf_get_current_task_proto;
1302         case BPF_FUNC_get_current_task_btf:
1303                 return &bpf_get_current_task_btf_proto;
1304         case BPF_FUNC_get_current_uid_gid:
1305                 return &bpf_get_current_uid_gid_proto;
1306         case BPF_FUNC_get_current_comm:
1307                 return &bpf_get_current_comm_proto;
1308         case BPF_FUNC_trace_printk:
1309                 return bpf_get_trace_printk_proto();
1310         case BPF_FUNC_get_smp_processor_id:
1311                 return &bpf_get_smp_processor_id_proto;
1312         case BPF_FUNC_get_numa_node_id:
1313                 return &bpf_get_numa_node_id_proto;
1314         case BPF_FUNC_perf_event_read:
1315                 return &bpf_perf_event_read_proto;
1316         case BPF_FUNC_probe_write_user:
1317                 return bpf_get_probe_write_proto();
1318         case BPF_FUNC_current_task_under_cgroup:
1319                 return &bpf_current_task_under_cgroup_proto;
1320         case BPF_FUNC_get_prandom_u32:
1321                 return &bpf_get_prandom_u32_proto;
1322         case BPF_FUNC_probe_read_user:
1323                 return &bpf_probe_read_user_proto;
1324         case BPF_FUNC_probe_read_kernel:
1325                 return &bpf_probe_read_kernel_proto;
1326         case BPF_FUNC_probe_read_user_str:
1327                 return &bpf_probe_read_user_str_proto;
1328         case BPF_FUNC_probe_read_kernel_str:
1329                 return &bpf_probe_read_kernel_str_proto;
1330 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1331         case BPF_FUNC_probe_read:
1332                 return &bpf_probe_read_compat_proto;
1333         case BPF_FUNC_probe_read_str:
1334                 return &bpf_probe_read_compat_str_proto;
1335 #endif
1336 #ifdef CONFIG_CGROUPS
1337         case BPF_FUNC_get_current_cgroup_id:
1338                 return &bpf_get_current_cgroup_id_proto;
1339 #endif
1340         case BPF_FUNC_send_signal:
1341                 return &bpf_send_signal_proto;
1342         case BPF_FUNC_send_signal_thread:
1343                 return &bpf_send_signal_thread_proto;
1344         case BPF_FUNC_perf_event_read_value:
1345                 return &bpf_perf_event_read_value_proto;
1346         case BPF_FUNC_get_ns_current_pid_tgid:
1347                 return &bpf_get_ns_current_pid_tgid_proto;
1348         case BPF_FUNC_ringbuf_output:
1349                 return &bpf_ringbuf_output_proto;
1350         case BPF_FUNC_ringbuf_reserve:
1351                 return &bpf_ringbuf_reserve_proto;
1352         case BPF_FUNC_ringbuf_submit:
1353                 return &bpf_ringbuf_submit_proto;
1354         case BPF_FUNC_ringbuf_discard:
1355                 return &bpf_ringbuf_discard_proto;
1356         case BPF_FUNC_ringbuf_query:
1357                 return &bpf_ringbuf_query_proto;
1358         case BPF_FUNC_jiffies64:
1359                 return &bpf_jiffies64_proto;
1360         case BPF_FUNC_get_task_stack:
1361                 return &bpf_get_task_stack_proto;
1362         case BPF_FUNC_copy_from_user:
1363                 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1364         case BPF_FUNC_snprintf_btf:
1365                 return &bpf_snprintf_btf_proto;
1366         case BPF_FUNC_per_cpu_ptr:
1367                 return &bpf_per_cpu_ptr_proto;
1368         case BPF_FUNC_this_cpu_ptr:
1369                 return &bpf_this_cpu_ptr_proto;
1370         default:
1371                 return NULL;
1372         }
1373 }
1374
1375 static const struct bpf_func_proto *
1376 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1377 {
1378         switch (func_id) {
1379         case BPF_FUNC_perf_event_output:
1380                 return &bpf_perf_event_output_proto;
1381         case BPF_FUNC_get_stackid:
1382                 return &bpf_get_stackid_proto;
1383         case BPF_FUNC_get_stack:
1384                 return &bpf_get_stack_proto;
1385 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1386         case BPF_FUNC_override_return:
1387                 return &bpf_override_return_proto;
1388 #endif
1389         default:
1390                 return bpf_tracing_func_proto(func_id, prog);
1391         }
1392 }
1393
1394 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1395 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1396                                         const struct bpf_prog *prog,
1397                                         struct bpf_insn_access_aux *info)
1398 {
1399         if (off < 0 || off >= sizeof(struct pt_regs))
1400                 return false;
1401         if (type != BPF_READ)
1402                 return false;
1403         if (off % size != 0)
1404                 return false;
1405         /*
1406          * Assertion for 32 bit to make sure last 8 byte access
1407          * (BPF_DW) to the last 4 byte member is disallowed.
1408          */
1409         if (off + size > sizeof(struct pt_regs))
1410                 return false;
1411
1412         return true;
1413 }
1414
1415 const struct bpf_verifier_ops kprobe_verifier_ops = {
1416         .get_func_proto  = kprobe_prog_func_proto,
1417         .is_valid_access = kprobe_prog_is_valid_access,
1418 };
1419
1420 const struct bpf_prog_ops kprobe_prog_ops = {
1421 };
1422
1423 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1424            u64, flags, void *, data, u64, size)
1425 {
1426         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1427
1428         /*
1429          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1430          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1431          * from there and call the same bpf_perf_event_output() helper inline.
1432          */
1433         return ____bpf_perf_event_output(regs, map, flags, data, size);
1434 }
1435
1436 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1437         .func           = bpf_perf_event_output_tp,
1438         .gpl_only       = true,
1439         .ret_type       = RET_INTEGER,
1440         .arg1_type      = ARG_PTR_TO_CTX,
1441         .arg2_type      = ARG_CONST_MAP_PTR,
1442         .arg3_type      = ARG_ANYTHING,
1443         .arg4_type      = ARG_PTR_TO_MEM,
1444         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1445 };
1446
1447 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1448            u64, flags)
1449 {
1450         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1451
1452         /*
1453          * Same comment as in bpf_perf_event_output_tp(), only that this time
1454          * the other helper's function body cannot be inlined due to being
1455          * external, thus we need to call raw helper function.
1456          */
1457         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1458                                flags, 0, 0);
1459 }
1460
1461 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1462         .func           = bpf_get_stackid_tp,
1463         .gpl_only       = true,
1464         .ret_type       = RET_INTEGER,
1465         .arg1_type      = ARG_PTR_TO_CTX,
1466         .arg2_type      = ARG_CONST_MAP_PTR,
1467         .arg3_type      = ARG_ANYTHING,
1468 };
1469
1470 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1471            u64, flags)
1472 {
1473         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1474
1475         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1476                              (unsigned long) size, flags, 0);
1477 }
1478
1479 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1480         .func           = bpf_get_stack_tp,
1481         .gpl_only       = true,
1482         .ret_type       = RET_INTEGER,
1483         .arg1_type      = ARG_PTR_TO_CTX,
1484         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1485         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1486         .arg4_type      = ARG_ANYTHING,
1487 };
1488
1489 static const struct bpf_func_proto *
1490 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1491 {
1492         switch (func_id) {
1493         case BPF_FUNC_perf_event_output:
1494                 return &bpf_perf_event_output_proto_tp;
1495         case BPF_FUNC_get_stackid:
1496                 return &bpf_get_stackid_proto_tp;
1497         case BPF_FUNC_get_stack:
1498                 return &bpf_get_stack_proto_tp;
1499         default:
1500                 return bpf_tracing_func_proto(func_id, prog);
1501         }
1502 }
1503
1504 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1505                                     const struct bpf_prog *prog,
1506                                     struct bpf_insn_access_aux *info)
1507 {
1508         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1509                 return false;
1510         if (type != BPF_READ)
1511                 return false;
1512         if (off % size != 0)
1513                 return false;
1514
1515         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1516         return true;
1517 }
1518
1519 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1520         .get_func_proto  = tp_prog_func_proto,
1521         .is_valid_access = tp_prog_is_valid_access,
1522 };
1523
1524 const struct bpf_prog_ops tracepoint_prog_ops = {
1525 };
1526
1527 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1528            struct bpf_perf_event_value *, buf, u32, size)
1529 {
1530         int err = -EINVAL;
1531
1532         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1533                 goto clear;
1534         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1535                                     &buf->running);
1536         if (unlikely(err))
1537                 goto clear;
1538         return 0;
1539 clear:
1540         memset(buf, 0, size);
1541         return err;
1542 }
1543
1544 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1545          .func           = bpf_perf_prog_read_value,
1546          .gpl_only       = true,
1547          .ret_type       = RET_INTEGER,
1548          .arg1_type      = ARG_PTR_TO_CTX,
1549          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1550          .arg3_type      = ARG_CONST_SIZE,
1551 };
1552
1553 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1554            void *, buf, u32, size, u64, flags)
1555 {
1556 #ifndef CONFIG_X86
1557         return -ENOENT;
1558 #else
1559         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1560         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1561         u32 to_copy;
1562
1563         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1564                 return -EINVAL;
1565
1566         if (unlikely(!br_stack))
1567                 return -EINVAL;
1568
1569         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1570                 return br_stack->nr * br_entry_size;
1571
1572         if (!buf || (size % br_entry_size != 0))
1573                 return -EINVAL;
1574
1575         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1576         memcpy(buf, br_stack->entries, to_copy);
1577
1578         return to_copy;
1579 #endif
1580 }
1581
1582 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1583         .func           = bpf_read_branch_records,
1584         .gpl_only       = true,
1585         .ret_type       = RET_INTEGER,
1586         .arg1_type      = ARG_PTR_TO_CTX,
1587         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1588         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1589         .arg4_type      = ARG_ANYTHING,
1590 };
1591
1592 static const struct bpf_func_proto *
1593 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1594 {
1595         switch (func_id) {
1596         case BPF_FUNC_perf_event_output:
1597                 return &bpf_perf_event_output_proto_tp;
1598         case BPF_FUNC_get_stackid:
1599                 return &bpf_get_stackid_proto_pe;
1600         case BPF_FUNC_get_stack:
1601                 return &bpf_get_stack_proto_pe;
1602         case BPF_FUNC_perf_prog_read_value:
1603                 return &bpf_perf_prog_read_value_proto;
1604         case BPF_FUNC_read_branch_records:
1605                 return &bpf_read_branch_records_proto;
1606         default:
1607                 return bpf_tracing_func_proto(func_id, prog);
1608         }
1609 }
1610
1611 /*
1612  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1613  * to avoid potential recursive reuse issue when/if tracepoints are added
1614  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1615  *
1616  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1617  * in normal, irq, and nmi context.
1618  */
1619 struct bpf_raw_tp_regs {
1620         struct pt_regs regs[3];
1621 };
1622 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1623 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1624 static struct pt_regs *get_bpf_raw_tp_regs(void)
1625 {
1626         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1627         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1628
1629         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1630                 this_cpu_dec(bpf_raw_tp_nest_level);
1631                 return ERR_PTR(-EBUSY);
1632         }
1633
1634         return &tp_regs->regs[nest_level - 1];
1635 }
1636
1637 static void put_bpf_raw_tp_regs(void)
1638 {
1639         this_cpu_dec(bpf_raw_tp_nest_level);
1640 }
1641
1642 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1643            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1644 {
1645         struct pt_regs *regs = get_bpf_raw_tp_regs();
1646         int ret;
1647
1648         if (IS_ERR(regs))
1649                 return PTR_ERR(regs);
1650
1651         perf_fetch_caller_regs(regs);
1652         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1653
1654         put_bpf_raw_tp_regs();
1655         return ret;
1656 }
1657
1658 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1659         .func           = bpf_perf_event_output_raw_tp,
1660         .gpl_only       = true,
1661         .ret_type       = RET_INTEGER,
1662         .arg1_type      = ARG_PTR_TO_CTX,
1663         .arg2_type      = ARG_CONST_MAP_PTR,
1664         .arg3_type      = ARG_ANYTHING,
1665         .arg4_type      = ARG_PTR_TO_MEM,
1666         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1667 };
1668
1669 extern const struct bpf_func_proto bpf_skb_output_proto;
1670 extern const struct bpf_func_proto bpf_xdp_output_proto;
1671
1672 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1673            struct bpf_map *, map, u64, flags)
1674 {
1675         struct pt_regs *regs = get_bpf_raw_tp_regs();
1676         int ret;
1677
1678         if (IS_ERR(regs))
1679                 return PTR_ERR(regs);
1680
1681         perf_fetch_caller_regs(regs);
1682         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1683         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1684                               flags, 0, 0);
1685         put_bpf_raw_tp_regs();
1686         return ret;
1687 }
1688
1689 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1690         .func           = bpf_get_stackid_raw_tp,
1691         .gpl_only       = true,
1692         .ret_type       = RET_INTEGER,
1693         .arg1_type      = ARG_PTR_TO_CTX,
1694         .arg2_type      = ARG_CONST_MAP_PTR,
1695         .arg3_type      = ARG_ANYTHING,
1696 };
1697
1698 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1699            void *, buf, u32, size, u64, flags)
1700 {
1701         struct pt_regs *regs = get_bpf_raw_tp_regs();
1702         int ret;
1703
1704         if (IS_ERR(regs))
1705                 return PTR_ERR(regs);
1706
1707         perf_fetch_caller_regs(regs);
1708         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1709                             (unsigned long) size, flags, 0);
1710         put_bpf_raw_tp_regs();
1711         return ret;
1712 }
1713
1714 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1715         .func           = bpf_get_stack_raw_tp,
1716         .gpl_only       = true,
1717         .ret_type       = RET_INTEGER,
1718         .arg1_type      = ARG_PTR_TO_CTX,
1719         .arg2_type      = ARG_PTR_TO_MEM,
1720         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1721         .arg4_type      = ARG_ANYTHING,
1722 };
1723
1724 static const struct bpf_func_proto *
1725 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1726 {
1727         switch (func_id) {
1728         case BPF_FUNC_perf_event_output:
1729                 return &bpf_perf_event_output_proto_raw_tp;
1730         case BPF_FUNC_get_stackid:
1731                 return &bpf_get_stackid_proto_raw_tp;
1732         case BPF_FUNC_get_stack:
1733                 return &bpf_get_stack_proto_raw_tp;
1734         default:
1735                 return bpf_tracing_func_proto(func_id, prog);
1736         }
1737 }
1738
1739 const struct bpf_func_proto *
1740 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1741 {
1742         switch (func_id) {
1743 #ifdef CONFIG_NET
1744         case BPF_FUNC_skb_output:
1745                 return &bpf_skb_output_proto;
1746         case BPF_FUNC_xdp_output:
1747                 return &bpf_xdp_output_proto;
1748         case BPF_FUNC_skc_to_tcp6_sock:
1749                 return &bpf_skc_to_tcp6_sock_proto;
1750         case BPF_FUNC_skc_to_tcp_sock:
1751                 return &bpf_skc_to_tcp_sock_proto;
1752         case BPF_FUNC_skc_to_tcp_timewait_sock:
1753                 return &bpf_skc_to_tcp_timewait_sock_proto;
1754         case BPF_FUNC_skc_to_tcp_request_sock:
1755                 return &bpf_skc_to_tcp_request_sock_proto;
1756         case BPF_FUNC_skc_to_udp6_sock:
1757                 return &bpf_skc_to_udp6_sock_proto;
1758         case BPF_FUNC_sk_storage_get:
1759                 return &bpf_sk_storage_get_tracing_proto;
1760         case BPF_FUNC_sk_storage_delete:
1761                 return &bpf_sk_storage_delete_tracing_proto;
1762         case BPF_FUNC_sock_from_file:
1763                 return &bpf_sock_from_file_proto;
1764         case BPF_FUNC_get_socket_cookie:
1765                 return &bpf_get_socket_ptr_cookie_proto;
1766 #endif
1767         case BPF_FUNC_seq_printf:
1768                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1769                        &bpf_seq_printf_proto :
1770                        NULL;
1771         case BPF_FUNC_seq_write:
1772                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1773                        &bpf_seq_write_proto :
1774                        NULL;
1775         case BPF_FUNC_seq_printf_btf:
1776                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1777                        &bpf_seq_printf_btf_proto :
1778                        NULL;
1779         case BPF_FUNC_d_path:
1780                 return &bpf_d_path_proto;
1781         default:
1782                 return raw_tp_prog_func_proto(func_id, prog);
1783         }
1784 }
1785
1786 static bool raw_tp_prog_is_valid_access(int off, int size,
1787                                         enum bpf_access_type type,
1788                                         const struct bpf_prog *prog,
1789                                         struct bpf_insn_access_aux *info)
1790 {
1791         if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1792                 return false;
1793         if (type != BPF_READ)
1794                 return false;
1795         if (off % size != 0)
1796                 return false;
1797         return true;
1798 }
1799
1800 static bool tracing_prog_is_valid_access(int off, int size,
1801                                          enum bpf_access_type type,
1802                                          const struct bpf_prog *prog,
1803                                          struct bpf_insn_access_aux *info)
1804 {
1805         if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1806                 return false;
1807         if (type != BPF_READ)
1808                 return false;
1809         if (off % size != 0)
1810                 return false;
1811         return btf_ctx_access(off, size, type, prog, info);
1812 }
1813
1814 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1815                                      const union bpf_attr *kattr,
1816                                      union bpf_attr __user *uattr)
1817 {
1818         return -ENOTSUPP;
1819 }
1820
1821 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1822         .get_func_proto  = raw_tp_prog_func_proto,
1823         .is_valid_access = raw_tp_prog_is_valid_access,
1824 };
1825
1826 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1827 #ifdef CONFIG_NET
1828         .test_run = bpf_prog_test_run_raw_tp,
1829 #endif
1830 };
1831
1832 const struct bpf_verifier_ops tracing_verifier_ops = {
1833         .get_func_proto  = tracing_prog_func_proto,
1834         .is_valid_access = tracing_prog_is_valid_access,
1835 };
1836
1837 const struct bpf_prog_ops tracing_prog_ops = {
1838         .test_run = bpf_prog_test_run_tracing,
1839 };
1840
1841 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1842                                                  enum bpf_access_type type,
1843                                                  const struct bpf_prog *prog,
1844                                                  struct bpf_insn_access_aux *info)
1845 {
1846         if (off == 0) {
1847                 if (size != sizeof(u64) || type != BPF_READ)
1848                         return false;
1849                 info->reg_type = PTR_TO_TP_BUFFER;
1850         }
1851         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1852 }
1853
1854 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1855         .get_func_proto  = raw_tp_prog_func_proto,
1856         .is_valid_access = raw_tp_writable_prog_is_valid_access,
1857 };
1858
1859 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1860 };
1861
1862 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1863                                     const struct bpf_prog *prog,
1864                                     struct bpf_insn_access_aux *info)
1865 {
1866         const int size_u64 = sizeof(u64);
1867
1868         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1869                 return false;
1870         if (type != BPF_READ)
1871                 return false;
1872         if (off % size != 0) {
1873                 if (sizeof(unsigned long) != 4)
1874                         return false;
1875                 if (size != 8)
1876                         return false;
1877                 if (off % size != 4)
1878                         return false;
1879         }
1880
1881         switch (off) {
1882         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1883                 bpf_ctx_record_field_size(info, size_u64);
1884                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1885                         return false;
1886                 break;
1887         case bpf_ctx_range(struct bpf_perf_event_data, addr):
1888                 bpf_ctx_record_field_size(info, size_u64);
1889                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1890                         return false;
1891                 break;
1892         default:
1893                 if (size != sizeof(long))
1894                         return false;
1895         }
1896
1897         return true;
1898 }
1899
1900 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1901                                       const struct bpf_insn *si,
1902                                       struct bpf_insn *insn_buf,
1903                                       struct bpf_prog *prog, u32 *target_size)
1904 {
1905         struct bpf_insn *insn = insn_buf;
1906
1907         switch (si->off) {
1908         case offsetof(struct bpf_perf_event_data, sample_period):
1909                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1910                                                        data), si->dst_reg, si->src_reg,
1911                                       offsetof(struct bpf_perf_event_data_kern, data));
1912                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1913                                       bpf_target_off(struct perf_sample_data, period, 8,
1914                                                      target_size));
1915                 break;
1916         case offsetof(struct bpf_perf_event_data, addr):
1917                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1918                                                        data), si->dst_reg, si->src_reg,
1919                                       offsetof(struct bpf_perf_event_data_kern, data));
1920                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1921                                       bpf_target_off(struct perf_sample_data, addr, 8,
1922                                                      target_size));
1923                 break;
1924         default:
1925                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1926                                                        regs), si->dst_reg, si->src_reg,
1927                                       offsetof(struct bpf_perf_event_data_kern, regs));
1928                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1929                                       si->off);
1930                 break;
1931         }
1932
1933         return insn - insn_buf;
1934 }
1935
1936 const struct bpf_verifier_ops perf_event_verifier_ops = {
1937         .get_func_proto         = pe_prog_func_proto,
1938         .is_valid_access        = pe_prog_is_valid_access,
1939         .convert_ctx_access     = pe_prog_convert_ctx_access,
1940 };
1941
1942 const struct bpf_prog_ops perf_event_prog_ops = {
1943 };
1944
1945 static DEFINE_MUTEX(bpf_event_mutex);
1946
1947 #define BPF_TRACE_MAX_PROGS 64
1948
1949 int perf_event_attach_bpf_prog(struct perf_event *event,
1950                                struct bpf_prog *prog)
1951 {
1952         struct bpf_prog_array *old_array;
1953         struct bpf_prog_array *new_array;
1954         int ret = -EEXIST;
1955
1956         /*
1957          * Kprobe override only works if they are on the function entry,
1958          * and only if they are on the opt-in list.
1959          */
1960         if (prog->kprobe_override &&
1961             (!trace_kprobe_on_func_entry(event->tp_event) ||
1962              !trace_kprobe_error_injectable(event->tp_event)))
1963                 return -EINVAL;
1964
1965         mutex_lock(&bpf_event_mutex);
1966
1967         if (event->prog)
1968                 goto unlock;
1969
1970         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1971         if (old_array &&
1972             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1973                 ret = -E2BIG;
1974                 goto unlock;
1975         }
1976
1977         ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1978         if (ret < 0)
1979                 goto unlock;
1980
1981         /* set the new array to event->tp_event and set event->prog */
1982         event->prog = prog;
1983         rcu_assign_pointer(event->tp_event->prog_array, new_array);
1984         bpf_prog_array_free(old_array);
1985
1986 unlock:
1987         mutex_unlock(&bpf_event_mutex);
1988         return ret;
1989 }
1990
1991 void perf_event_detach_bpf_prog(struct perf_event *event)
1992 {
1993         struct bpf_prog_array *old_array;
1994         struct bpf_prog_array *new_array;
1995         int ret;
1996
1997         mutex_lock(&bpf_event_mutex);
1998
1999         if (!event->prog)
2000                 goto unlock;
2001
2002         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2003         ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
2004         if (ret == -ENOENT)
2005                 goto unlock;
2006         if (ret < 0) {
2007                 bpf_prog_array_delete_safe(old_array, event->prog);
2008         } else {
2009                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2010                 bpf_prog_array_free(old_array);
2011         }
2012
2013         bpf_prog_put(event->prog);
2014         event->prog = NULL;
2015
2016 unlock:
2017         mutex_unlock(&bpf_event_mutex);
2018 }
2019
2020 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2021 {
2022         struct perf_event_query_bpf __user *uquery = info;
2023         struct perf_event_query_bpf query = {};
2024         struct bpf_prog_array *progs;
2025         u32 *ids, prog_cnt, ids_len;
2026         int ret;
2027
2028         if (!perfmon_capable())
2029                 return -EPERM;
2030         if (event->attr.type != PERF_TYPE_TRACEPOINT)
2031                 return -EINVAL;
2032         if (copy_from_user(&query, uquery, sizeof(query)))
2033                 return -EFAULT;
2034
2035         ids_len = query.ids_len;
2036         if (ids_len > BPF_TRACE_MAX_PROGS)
2037                 return -E2BIG;
2038         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2039         if (!ids)
2040                 return -ENOMEM;
2041         /*
2042          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2043          * is required when user only wants to check for uquery->prog_cnt.
2044          * There is no need to check for it since the case is handled
2045          * gracefully in bpf_prog_array_copy_info.
2046          */
2047
2048         mutex_lock(&bpf_event_mutex);
2049         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2050         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2051         mutex_unlock(&bpf_event_mutex);
2052
2053         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2054             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2055                 ret = -EFAULT;
2056
2057         kfree(ids);
2058         return ret;
2059 }
2060
2061 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2062 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2063
2064 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2065 {
2066         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2067
2068         for (; btp < __stop__bpf_raw_tp; btp++) {
2069                 if (!strcmp(btp->tp->name, name))
2070                         return btp;
2071         }
2072
2073         return bpf_get_raw_tracepoint_module(name);
2074 }
2075
2076 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2077 {
2078         struct module *mod;
2079
2080         preempt_disable();
2081         mod = __module_address((unsigned long)btp);
2082         module_put(mod);
2083         preempt_enable();
2084 }
2085
2086 static __always_inline
2087 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2088 {
2089         cant_sleep();
2090         rcu_read_lock();
2091         (void) BPF_PROG_RUN(prog, args);
2092         rcu_read_unlock();
2093 }
2094
2095 #define UNPACK(...)                     __VA_ARGS__
2096 #define REPEAT_1(FN, DL, X, ...)        FN(X)
2097 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2098 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2099 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2100 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2101 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2102 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2103 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2104 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2105 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2106 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2107 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2108 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
2109
2110 #define SARG(X)         u64 arg##X
2111 #define COPY(X)         args[X] = arg##X
2112
2113 #define __DL_COM        (,)
2114 #define __DL_SEM        (;)
2115
2116 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2117
2118 #define BPF_TRACE_DEFN_x(x)                                             \
2119         void bpf_trace_run##x(struct bpf_prog *prog,                    \
2120                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
2121         {                                                               \
2122                 u64 args[x];                                            \
2123                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
2124                 __bpf_trace_run(prog, args);                            \
2125         }                                                               \
2126         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2127 BPF_TRACE_DEFN_x(1);
2128 BPF_TRACE_DEFN_x(2);
2129 BPF_TRACE_DEFN_x(3);
2130 BPF_TRACE_DEFN_x(4);
2131 BPF_TRACE_DEFN_x(5);
2132 BPF_TRACE_DEFN_x(6);
2133 BPF_TRACE_DEFN_x(7);
2134 BPF_TRACE_DEFN_x(8);
2135 BPF_TRACE_DEFN_x(9);
2136 BPF_TRACE_DEFN_x(10);
2137 BPF_TRACE_DEFN_x(11);
2138 BPF_TRACE_DEFN_x(12);
2139
2140 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2141 {
2142         struct tracepoint *tp = btp->tp;
2143
2144         /*
2145          * check that program doesn't access arguments beyond what's
2146          * available in this tracepoint
2147          */
2148         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2149                 return -EINVAL;
2150
2151         if (prog->aux->max_tp_access > btp->writable_size)
2152                 return -EINVAL;
2153
2154         return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
2155 }
2156
2157 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2158 {
2159         return __bpf_probe_register(btp, prog);
2160 }
2161
2162 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2163 {
2164         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2165 }
2166
2167 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2168                             u32 *fd_type, const char **buf,
2169                             u64 *probe_offset, u64 *probe_addr)
2170 {
2171         bool is_tracepoint, is_syscall_tp;
2172         struct bpf_prog *prog;
2173         int flags, err = 0;
2174
2175         prog = event->prog;
2176         if (!prog)
2177                 return -ENOENT;
2178
2179         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2180         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2181                 return -EOPNOTSUPP;
2182
2183         *prog_id = prog->aux->id;
2184         flags = event->tp_event->flags;
2185         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2186         is_syscall_tp = is_syscall_trace_event(event->tp_event);
2187
2188         if (is_tracepoint || is_syscall_tp) {
2189                 *buf = is_tracepoint ? event->tp_event->tp->name
2190                                      : event->tp_event->name;
2191                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2192                 *probe_offset = 0x0;
2193                 *probe_addr = 0x0;
2194         } else {
2195                 /* kprobe/uprobe */
2196                 err = -EOPNOTSUPP;
2197 #ifdef CONFIG_KPROBE_EVENTS
2198                 if (flags & TRACE_EVENT_FL_KPROBE)
2199                         err = bpf_get_kprobe_info(event, fd_type, buf,
2200                                                   probe_offset, probe_addr,
2201                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2202 #endif
2203 #ifdef CONFIG_UPROBE_EVENTS
2204                 if (flags & TRACE_EVENT_FL_UPROBE)
2205                         err = bpf_get_uprobe_info(event, fd_type, buf,
2206                                                   probe_offset,
2207                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2208 #endif
2209         }
2210
2211         return err;
2212 }
2213
2214 static int __init send_signal_irq_work_init(void)
2215 {
2216         int cpu;
2217         struct send_signal_irq_work *work;
2218
2219         for_each_possible_cpu(cpu) {
2220                 work = per_cpu_ptr(&send_signal_work, cpu);
2221                 init_irq_work(&work->irq_work, do_bpf_send_signal);
2222         }
2223         return 0;
2224 }
2225
2226 subsys_initcall(send_signal_irq_work_init);
2227
2228 #ifdef CONFIG_MODULES
2229 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2230                             void *module)
2231 {
2232         struct bpf_trace_module *btm, *tmp;
2233         struct module *mod = module;
2234         int ret = 0;
2235
2236         if (mod->num_bpf_raw_events == 0 ||
2237             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2238                 goto out;
2239
2240         mutex_lock(&bpf_module_mutex);
2241
2242         switch (op) {
2243         case MODULE_STATE_COMING:
2244                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2245                 if (btm) {
2246                         btm->module = module;
2247                         list_add(&btm->list, &bpf_trace_modules);
2248                 } else {
2249                         ret = -ENOMEM;
2250                 }
2251                 break;
2252         case MODULE_STATE_GOING:
2253                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2254                         if (btm->module == module) {
2255                                 list_del(&btm->list);
2256                                 kfree(btm);
2257                                 break;
2258                         }
2259                 }
2260                 break;
2261         }
2262
2263         mutex_unlock(&bpf_module_mutex);
2264
2265 out:
2266         return notifier_from_errno(ret);
2267 }
2268
2269 static struct notifier_block bpf_module_nb = {
2270         .notifier_call = bpf_event_notify,
2271 };
2272
2273 static int __init bpf_event_init(void)
2274 {
2275         register_module_notifier(&bpf_module_nb);
2276         return 0;
2277 }
2278
2279 fs_initcall(bpf_event_init);
2280 #endif /* CONFIG_MODULES */