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