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 #include <linux/fprobe.h>
  21 #include <linux/bsearch.h>
  22 #include <linux/sort.h>
  23
  24 #include <net/bpf_sk_storage.h>
  25
  26 #include <uapi/linux/bpf.h>
  27 #include <uapi/linux/btf.h>
  28
  29 #include <asm/tlb.h>
  30
  31 #include "trace_probe.h"
  32 #include "trace.h"
  33
  34 #define CREATE_TRACE_POINTS
  35 #include "bpf_trace.h"
  36
  37 #define bpf_event_rcu_dereference(p)                                    \
  38         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  39
  40 #ifdef CONFIG_MODULES
  41 struct bpf_trace_module {
  42         struct module *module;
  43         struct list_head list;
  44 };
  45
  46 static LIST_HEAD(bpf_trace_modules);
  47 static DEFINE_MUTEX(bpf_module_mutex);
  48
  49 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  50 {
  51         struct bpf_raw_event_map *btp, *ret = NULL;
  52         struct bpf_trace_module *btm;
  53         unsigned int i;
  54
  55         mutex_lock(&bpf_module_mutex);
  56         list_for_each_entry(btm, &bpf_trace_modules, list) {
  57                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  58                         btp = &btm->module->bpf_raw_events[i];
  59                         if (!strcmp(btp->tp->name, name)) {
  60                                 if (try_module_get(btm->module))
  61                                         ret = btp;
  62                                 goto out;
  63                         }
  64                 }
  65         }
  66 out:
  67         mutex_unlock(&bpf_module_mutex);
  68         return ret;
  69 }
  70 #else
  71 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  72 {
  73         return NULL;
  74 }
  75 #endif /* CONFIG_MODULES */
  76
  77 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  78 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  79
  80 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
  81                                   u64 flags, const struct btf **btf,
  82                                   s32 *btf_id);
  83 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
  84 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
  85
  86 /**
  87  * trace_call_bpf - invoke BPF program
  88  * @call: tracepoint event
  89  * @ctx: opaque context pointer
  90  *
  91  * kprobe handlers execute BPF programs via this helper.
  92  * Can be used from static tracepoints in the future.
  93  *
  94  * Return: BPF programs always return an integer which is interpreted by
  95  * kprobe handler as:
  96  * 0 - return from kprobe (event is filtered out)
  97  * 1 - store kprobe event into ring buffer
  98  * Other values are reserved and currently alias to 1
  99  */
 100 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 101 {
 102         unsigned int ret;
 103
 104         cant_sleep();
 105
 106         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 107                 /*
 108                  * since some bpf program is already running on this cpu,
 109                  * don't call into another bpf program (same or different)
 110                  * and don't send kprobe event into ring-buffer,
 111                  * so return zero here
 112                  */
 113                 ret = 0;
 114                 goto out;
 115         }
 116
 117         /*
 118          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 119          * to all call sites, we did a bpf_prog_array_valid() there to check
 120          * whether call->prog_array is empty or not, which is
 121          * a heuristic to speed up execution.
 122          *
 123          * If bpf_prog_array_valid() fetched prog_array was
 124          * non-NULL, we go into trace_call_bpf() and do the actual
 125          * proper rcu_dereference() under RCU lock.
 126          * If it turns out that prog_array is NULL then, we bail out.
 127          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 128          * was NULL, you'll skip the prog_array with the risk of missing
 129          * out of events when it was updated in between this and the
 130          * rcu_dereference() which is accepted risk.
 131          */
 132         rcu_read_lock();
 133         ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
 134                                  ctx, bpf_prog_run);
 135         rcu_read_unlock();
 136
 137  out:
 138         __this_cpu_dec(bpf_prog_active);
 139
 140         return ret;
 141 }
 142
 143 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 144 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 145 {
 146         regs_set_return_value(regs, rc);
 147         override_function_with_return(regs);
 148         return 0;
 149 }
 150
 151 static const struct bpf_func_proto bpf_override_return_proto = {
 152         .func           = bpf_override_return,
 153         .gpl_only       = true,
 154         .ret_type       = RET_INTEGER,
 155         .arg1_type      = ARG_PTR_TO_CTX,
 156         .arg2_type      = ARG_ANYTHING,
 157 };
 158 #endif
 159
 160 static __always_inline int
 161 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 162 {
 163         int ret;
 164
 165         ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 166         if (unlikely(ret < 0))
 167                 memset(dst, 0, size);
 168         return ret;
 169 }
 170
 171 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 172            const void __user *, unsafe_ptr)
 173 {
 174         return bpf_probe_read_user_common(dst, size, unsafe_ptr);
 175 }
 176
 177 const struct bpf_func_proto bpf_probe_read_user_proto = {
 178         .func           = bpf_probe_read_user,
 179         .gpl_only       = true,
 180         .ret_type       = RET_INTEGER,
 181         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 182         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 183         .arg3_type      = ARG_ANYTHING,
 184 };
 185
 186 static __always_inline int
 187 bpf_probe_read_user_str_common(void *dst, u32 size,
 188                                const void __user *unsafe_ptr)
 189 {
 190         int ret;
 191
 192         /*
 193          * NB: We rely on strncpy_from_user() not copying junk past the NUL
 194          * terminator into `dst`.
 195          *
 196          * strncpy_from_user() does long-sized strides in the fast path. If the
 197          * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
 198          * then there could be junk after the NUL in `dst`. If user takes `dst`
 199          * and keys a hash map with it, then semantically identical strings can
 200          * occupy multiple entries in the map.
 201          */
 202         ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 203         if (unlikely(ret < 0))
 204                 memset(dst, 0, size);
 205         return ret;
 206 }
 207
 208 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 209            const void __user *, unsafe_ptr)
 210 {
 211         return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
 212 }
 213
 214 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 215         .func           = bpf_probe_read_user_str,
 216         .gpl_only       = true,
 217         .ret_type       = RET_INTEGER,
 218         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 219         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 220         .arg3_type      = ARG_ANYTHING,
 221 };
 222
 223 static __always_inline int
 224 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
 225 {
 226         int ret;
 227
 228         ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
 229         if (unlikely(ret < 0))
 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;
 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                 memset(dst, 0, size);
 266         return ret;
 267 }
 268
 269 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 270            const void *, unsafe_ptr)
 271 {
 272         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 273 }
 274
 275 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 276         .func           = bpf_probe_read_kernel_str,
 277         .gpl_only       = true,
 278         .ret_type       = RET_INTEGER,
 279         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 280         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 281         .arg3_type      = ARG_ANYTHING,
 282 };
 283
 284 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 285 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 286            const void *, unsafe_ptr)
 287 {
 288         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 289                 return bpf_probe_read_user_common(dst, size,
 290                                 (__force void __user *)unsafe_ptr);
 291         }
 292         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 293 }
 294
 295 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 296         .func           = bpf_probe_read_compat,
 297         .gpl_only       = true,
 298         .ret_type       = RET_INTEGER,
 299         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 300         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 301         .arg3_type      = ARG_ANYTHING,
 302 };
 303
 304 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 305            const void *, unsafe_ptr)
 306 {
 307         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 308                 return bpf_probe_read_user_str_common(dst, size,
 309                                 (__force void __user *)unsafe_ptr);
 310         }
 311         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 312 }
 313
 314 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 315         .func           = bpf_probe_read_compat_str,
 316         .gpl_only       = true,
 317         .ret_type       = RET_INTEGER,
 318         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 319         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 320         .arg3_type      = ARG_ANYTHING,
 321 };
 322 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 323
 324 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 325            u32, size)
 326 {
 327         /*
 328          * Ensure we're in user context which is safe for the helper to
 329          * run. This helper has no business in a kthread.
 330          *
 331          * access_ok() should prevent writing to non-user memory, but in
 332          * some situations (nommu, temporary switch, etc) access_ok() does
 333          * not provide enough validation, hence the check on KERNEL_DS.
 334          *
 335          * nmi_uaccess_okay() ensures the probe is not run in an interim
 336          * state, when the task or mm are switched. This is specifically
 337          * required to prevent the use of temporary mm.
 338          */
 339
 340         if (unlikely(in_interrupt() ||
 341                      current->flags & (PF_KTHREAD | PF_EXITING)))
 342                 return -EPERM;
 343         if (unlikely(!nmi_uaccess_okay()))
 344                 return -EPERM;
 345
 346         return copy_to_user_nofault(unsafe_ptr, src, size);
 347 }
 348
 349 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 350         .func           = bpf_probe_write_user,
 351         .gpl_only       = true,
 352         .ret_type       = RET_INTEGER,
 353         .arg1_type      = ARG_ANYTHING,
 354         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 355         .arg3_type      = ARG_CONST_SIZE,
 356 };
 357
 358 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 359 {
 360         if (!capable(CAP_SYS_ADMIN))
 361                 return NULL;
 362
 363         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 364                             current->comm, task_pid_nr(current));
 365
 366         return &bpf_probe_write_user_proto;
 367 }
 368
 369 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
 370
 371 #define MAX_TRACE_PRINTK_VARARGS        3
 372 #define BPF_TRACE_PRINTK_SIZE           1024
 373
 374 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 375            u64, arg2, u64, arg3)
 376 {
 377         u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
 378         u32 *bin_args;
 379         static char buf[BPF_TRACE_PRINTK_SIZE];
 380         unsigned long flags;
 381         int ret;
 382
 383         ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
 384                                   MAX_TRACE_PRINTK_VARARGS);
 385         if (ret < 0)
 386                 return ret;
 387
 388         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 389         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 390
 391         trace_bpf_trace_printk(buf);
 392         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 393
 394         bpf_bprintf_cleanup();
 395
 396         return ret;
 397 }
 398
 399 static const struct bpf_func_proto bpf_trace_printk_proto = {
 400         .func           = bpf_trace_printk,
 401         .gpl_only       = true,
 402         .ret_type       = RET_INTEGER,
 403         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 404         .arg2_type      = ARG_CONST_SIZE,
 405 };
 406
 407 static void __set_printk_clr_event(void)
 408 {
 409         /*
 410          * This program might be calling bpf_trace_printk,
 411          * so enable the associated bpf_trace/bpf_trace_printk event.
 412          * Repeat this each time as it is possible a user has
 413          * disabled bpf_trace_printk events.  By loading a program
 414          * calling bpf_trace_printk() however the user has expressed
 415          * the intent to see such events.
 416          */
 417         if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 418                 pr_warn_ratelimited("could not enable bpf_trace_printk events");
 419 }
 420
 421 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 422 {
 423         __set_printk_clr_event();
 424         return &bpf_trace_printk_proto;
 425 }
 426
 427 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
 428            u32, data_len)
 429 {
 430         static char buf[BPF_TRACE_PRINTK_SIZE];
 431         unsigned long flags;
 432         int ret, num_args;
 433         u32 *bin_args;
 434
 435         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 436             (data_len && !data))
 437                 return -EINVAL;
 438         num_args = data_len / 8;
 439
 440         ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 441         if (ret < 0)
 442                 return ret;
 443
 444         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 445         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 446
 447         trace_bpf_trace_printk(buf);
 448         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 449
 450         bpf_bprintf_cleanup();
 451
 452         return ret;
 453 }
 454
 455 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
 456         .func           = bpf_trace_vprintk,
 457         .gpl_only       = true,
 458         .ret_type       = RET_INTEGER,
 459         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 460         .arg2_type      = ARG_CONST_SIZE,
 461         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 462         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
 463 };
 464
 465 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
 466 {
 467         __set_printk_clr_event();
 468         return &bpf_trace_vprintk_proto;
 469 }
 470
 471 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 472            const void *, data, u32, data_len)
 473 {
 474         int err, num_args;
 475         u32 *bin_args;
 476
 477         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 478             (data_len && !data))
 479                 return -EINVAL;
 480         num_args = data_len / 8;
 481
 482         err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 483         if (err < 0)
 484                 return err;
 485
 486         seq_bprintf(m, fmt, bin_args);
 487
 488         bpf_bprintf_cleanup();
 489
 490         return seq_has_overflowed(m) ? -EOVERFLOW : 0;
 491 }
 492
 493 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
 494
 495 static const struct bpf_func_proto bpf_seq_printf_proto = {
 496         .func           = bpf_seq_printf,
 497         .gpl_only       = true,
 498         .ret_type       = RET_INTEGER,
 499         .arg1_type      = ARG_PTR_TO_BTF_ID,
 500         .arg1_btf_id    = &btf_seq_file_ids[0],
 501         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 502         .arg3_type      = ARG_CONST_SIZE,
 503         .arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 504         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 505 };
 506
 507 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
 508 {
 509         return seq_write(m, data, len) ? -EOVERFLOW : 0;
 510 }
 511
 512 static const struct bpf_func_proto bpf_seq_write_proto = {
 513         .func           = bpf_seq_write,
 514         .gpl_only       = true,
 515         .ret_type       = RET_INTEGER,
 516         .arg1_type      = ARG_PTR_TO_BTF_ID,
 517         .arg1_btf_id    = &btf_seq_file_ids[0],
 518         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 519         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 520 };
 521
 522 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
 523            u32, btf_ptr_size, u64, flags)
 524 {
 525         const struct btf *btf;
 526         s32 btf_id;
 527         int ret;
 528
 529         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 530         if (ret)
 531                 return ret;
 532
 533         return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
 534 }
 535
 536 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
 537         .func           = bpf_seq_printf_btf,
 538         .gpl_only       = true,
 539         .ret_type       = RET_INTEGER,
 540         .arg1_type      = ARG_PTR_TO_BTF_ID,
 541         .arg1_btf_id    = &btf_seq_file_ids[0],
 542         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 543         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 544         .arg4_type      = ARG_ANYTHING,
 545 };
 546
 547 static __always_inline int
 548 get_map_perf_counter(struct bpf_map *map, u64 flags,
 549                      u64 *value, u64 *enabled, u64 *running)
 550 {
 551         struct bpf_array *array = container_of(map, struct bpf_array, map);
 552         unsigned int cpu = smp_processor_id();
 553         u64 index = flags & BPF_F_INDEX_MASK;
 554         struct bpf_event_entry *ee;
 555
 556         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 557                 return -EINVAL;
 558         if (index == BPF_F_CURRENT_CPU)
 559                 index = cpu;
 560         if (unlikely(index >= array->map.max_entries))
 561                 return -E2BIG;
 562
 563         ee = READ_ONCE(array->ptrs[index]);
 564         if (!ee)
 565                 return -ENOENT;
 566
 567         return perf_event_read_local(ee->event, value, enabled, running);
 568 }
 569
 570 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 571 {
 572         u64 value = 0;
 573         int err;
 574
 575         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 576         /*
 577          * this api is ugly since we miss [-22..-2] range of valid
 578          * counter values, but that's uapi
 579          */
 580         if (err)
 581                 return err;
 582         return value;
 583 }
 584
 585 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 586         .func           = bpf_perf_event_read,
 587         .gpl_only       = true,
 588         .ret_type       = RET_INTEGER,
 589         .arg1_type      = ARG_CONST_MAP_PTR,
 590         .arg2_type      = ARG_ANYTHING,
 591 };
 592
 593 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 594            struct bpf_perf_event_value *, buf, u32, size)
 595 {
 596         int err = -EINVAL;
 597
 598         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 599                 goto clear;
 600         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 601                                    &buf->running);
 602         if (unlikely(err))
 603                 goto clear;
 604         return 0;
 605 clear:
 606         memset(buf, 0, size);
 607         return err;
 608 }
 609
 610 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 611         .func           = bpf_perf_event_read_value,
 612         .gpl_only       = true,
 613         .ret_type       = RET_INTEGER,
 614         .arg1_type      = ARG_CONST_MAP_PTR,
 615         .arg2_type      = ARG_ANYTHING,
 616         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 617         .arg4_type      = ARG_CONST_SIZE,
 618 };
 619
 620 static __always_inline u64
 621 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 622                         u64 flags, struct perf_sample_data *sd)
 623 {
 624         struct bpf_array *array = container_of(map, struct bpf_array, map);
 625         unsigned int cpu = smp_processor_id();
 626         u64 index = flags & BPF_F_INDEX_MASK;
 627         struct bpf_event_entry *ee;
 628         struct perf_event *event;
 629
 630         if (index == BPF_F_CURRENT_CPU)
 631                 index = cpu;
 632         if (unlikely(index >= array->map.max_entries))
 633                 return -E2BIG;
 634
 635         ee = READ_ONCE(array->ptrs[index]);
 636         if (!ee)
 637                 return -ENOENT;
 638
 639         event = ee->event;
 640         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 641                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 642                 return -EINVAL;
 643
 644         if (unlikely(event->oncpu != cpu))
 645                 return -EOPNOTSUPP;
 646
 647         return perf_event_output(event, sd, regs);
 648 }
 649
 650 /*
 651  * Support executing tracepoints in normal, irq, and nmi context that each call
 652  * bpf_perf_event_output
 653  */
 654 struct bpf_trace_sample_data {
 655         struct perf_sample_data sds[3];
 656 };
 657
 658 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 659 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 660 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 661            u64, flags, void *, data, u64, size)
 662 {
 663         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 664         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 665         struct perf_raw_record raw = {
 666                 .frag = {
 667                         .size = size,
 668                         .data = data,
 669                 },
 670         };
 671         struct perf_sample_data *sd;
 672         int err;
 673
 674         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 675                 err = -EBUSY;
 676                 goto out;
 677         }
 678
 679         sd = &sds->sds[nest_level - 1];
 680
 681         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 682                 err = -EINVAL;
 683                 goto out;
 684         }
 685
 686         perf_sample_data_init(sd, 0, 0);
 687         sd->raw = &raw;
 688
 689         err = __bpf_perf_event_output(regs, map, flags, sd);
 690
 691 out:
 692         this_cpu_dec(bpf_trace_nest_level);
 693         return err;
 694 }
 695
 696 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 697         .func           = bpf_perf_event_output,
 698         .gpl_only       = true,
 699         .ret_type       = RET_INTEGER,
 700         .arg1_type      = ARG_PTR_TO_CTX,
 701         .arg2_type      = ARG_CONST_MAP_PTR,
 702         .arg3_type      = ARG_ANYTHING,
 703         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 704         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 705 };
 706
 707 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 708 struct bpf_nested_pt_regs {
 709         struct pt_regs regs[3];
 710 };
 711 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 712 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 713
 714 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 715                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 716 {
 717         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 718         struct perf_raw_frag frag = {
 719                 .copy           = ctx_copy,
 720                 .size           = ctx_size,
 721                 .data           = ctx,
 722         };
 723         struct perf_raw_record raw = {
 724                 .frag = {
 725                         {
 726                                 .next   = ctx_size ? &frag : NULL,
 727                         },
 728                         .size   = meta_size,
 729                         .data   = meta,
 730                 },
 731         };
 732         struct perf_sample_data *sd;
 733         struct pt_regs *regs;
 734         u64 ret;
 735
 736         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 737                 ret = -EBUSY;
 738                 goto out;
 739         }
 740         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 741         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 742
 743         perf_fetch_caller_regs(regs);
 744         perf_sample_data_init(sd, 0, 0);
 745         sd->raw = &raw;
 746
 747         ret = __bpf_perf_event_output(regs, map, flags, sd);
 748 out:
 749         this_cpu_dec(bpf_event_output_nest_level);
 750         return ret;
 751 }
 752
 753 BPF_CALL_0(bpf_get_current_task)
 754 {
 755         return (long) current;
 756 }
 757
 758 const struct bpf_func_proto bpf_get_current_task_proto = {
 759         .func           = bpf_get_current_task,
 760         .gpl_only       = true,
 761         .ret_type       = RET_INTEGER,
 762 };
 763
 764 BPF_CALL_0(bpf_get_current_task_btf)
 765 {
 766         return (unsigned long) current;
 767 }
 768
 769 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
 770         .func           = bpf_get_current_task_btf,
 771         .gpl_only       = true,
 772         .ret_type       = RET_PTR_TO_BTF_ID,
 773         .ret_btf_id     = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 774 };
 775
 776 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
 777 {
 778         return (unsigned long) task_pt_regs(task);
 779 }
 780
 781 BTF_ID_LIST(bpf_task_pt_regs_ids)
 782 BTF_ID(struct, pt_regs)
 783
 784 const struct bpf_func_proto bpf_task_pt_regs_proto = {
 785         .func           = bpf_task_pt_regs,
 786         .gpl_only       = true,
 787         .arg1_type      = ARG_PTR_TO_BTF_ID,
 788         .arg1_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 789         .ret_type       = RET_PTR_TO_BTF_ID,
 790         .ret_btf_id     = &bpf_task_pt_regs_ids[0],
 791 };
 792
 793 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 794 {
 795         struct bpf_array *array = container_of(map, struct bpf_array, map);
 796         struct cgroup *cgrp;
 797
 798         if (unlikely(idx >= array->map.max_entries))
 799                 return -E2BIG;
 800
 801         cgrp = READ_ONCE(array->ptrs[idx]);
 802         if (unlikely(!cgrp))
 803                 return -EAGAIN;
 804
 805         return task_under_cgroup_hierarchy(current, cgrp);
 806 }
 807
 808 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 809         .func           = bpf_current_task_under_cgroup,
 810         .gpl_only       = false,
 811         .ret_type       = RET_INTEGER,
 812         .arg1_type      = ARG_CONST_MAP_PTR,
 813         .arg2_type      = ARG_ANYTHING,
 814 };
 815
 816 struct send_signal_irq_work {
 817         struct irq_work irq_work;
 818         struct task_struct *task;
 819         u32 sig;
 820         enum pid_type type;
 821 };
 822
 823 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 824
 825 static void do_bpf_send_signal(struct irq_work *entry)
 826 {
 827         struct send_signal_irq_work *work;
 828
 829         work = container_of(entry, struct send_signal_irq_work, irq_work);
 830         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 831 }
 832
 833 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 834 {
 835         struct send_signal_irq_work *work = NULL;
 836
 837         /* Similar to bpf_probe_write_user, task needs to be
 838          * in a sound condition and kernel memory access be
 839          * permitted in order to send signal to the current
 840          * task.
 841          */
 842         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 843                 return -EPERM;
 844         if (unlikely(!nmi_uaccess_okay()))
 845                 return -EPERM;
 846
 847         if (irqs_disabled()) {
 848                 /* Do an early check on signal validity. Otherwise,
 849                  * the error is lost in deferred irq_work.
 850                  */
 851                 if (unlikely(!valid_signal(sig)))
 852                         return -EINVAL;
 853
 854                 work = this_cpu_ptr(&send_signal_work);
 855                 if (irq_work_is_busy(&work->irq_work))
 856                         return -EBUSY;
 857
 858                 /* Add the current task, which is the target of sending signal,
 859                  * to the irq_work. The current task may change when queued
 860                  * irq works get executed.
 861                  */
 862                 work->task = current;
 863                 work->sig = sig;
 864                 work->type = type;
 865                 irq_work_queue(&work->irq_work);
 866                 return 0;
 867         }
 868
 869         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 870 }
 871
 872 BPF_CALL_1(bpf_send_signal, u32, sig)
 873 {
 874         return bpf_send_signal_common(sig, PIDTYPE_TGID);
 875 }
 876
 877 static const struct bpf_func_proto bpf_send_signal_proto = {
 878         .func           = bpf_send_signal,
 879         .gpl_only       = false,
 880         .ret_type       = RET_INTEGER,
 881         .arg1_type      = ARG_ANYTHING,
 882 };
 883
 884 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 885 {
 886         return bpf_send_signal_common(sig, PIDTYPE_PID);
 887 }
 888
 889 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 890         .func           = bpf_send_signal_thread,
 891         .gpl_only       = false,
 892         .ret_type       = RET_INTEGER,
 893         .arg1_type      = ARG_ANYTHING,
 894 };
 895
 896 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 897 {
 898         long len;
 899         char *p;
 900
 901         if (!sz)
 902                 return 0;
 903
 904         p = d_path(path, buf, sz);
 905         if (IS_ERR(p)) {
 906                 len = PTR_ERR(p);
 907         } else {
 908                 len = buf + sz - p;
 909                 memmove(buf, p, len);
 910         }
 911
 912         return len;
 913 }
 914
 915 BTF_SET_START(btf_allowlist_d_path)
 916 #ifdef CONFIG_SECURITY
 917 BTF_ID(func, security_file_permission)
 918 BTF_ID(func, security_inode_getattr)
 919 BTF_ID(func, security_file_open)
 920 #endif
 921 #ifdef CONFIG_SECURITY_PATH
 922 BTF_ID(func, security_path_truncate)
 923 #endif
 924 BTF_ID(func, vfs_truncate)
 925 BTF_ID(func, vfs_fallocate)
 926 BTF_ID(func, dentry_open)
 927 BTF_ID(func, vfs_getattr)
 928 BTF_ID(func, filp_close)
 929 BTF_SET_END(btf_allowlist_d_path)
 930
 931 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 932 {
 933         if (prog->type == BPF_PROG_TYPE_TRACING &&
 934             prog->expected_attach_type == BPF_TRACE_ITER)
 935                 return true;
 936
 937         if (prog->type == BPF_PROG_TYPE_LSM)
 938                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 939
 940         return btf_id_set_contains(&btf_allowlist_d_path,
 941                                    prog->aux->attach_btf_id);
 942 }
 943
 944 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 945
 946 static const struct bpf_func_proto bpf_d_path_proto = {
 947         .func           = bpf_d_path,
 948         .gpl_only       = false,
 949         .ret_type       = RET_INTEGER,
 950         .arg1_type      = ARG_PTR_TO_BTF_ID,
 951         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
 952         .arg2_type      = ARG_PTR_TO_MEM,
 953         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 954         .allowed        = bpf_d_path_allowed,
 955 };
 956
 957 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
 958                          BTF_F_PTR_RAW | BTF_F_ZERO)
 959
 960 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 961                                   u64 flags, const struct btf **btf,
 962                                   s32 *btf_id)
 963 {
 964         const struct btf_type *t;
 965
 966         if (unlikely(flags & ~(BTF_F_ALL)))
 967                 return -EINVAL;
 968
 969         if (btf_ptr_size != sizeof(struct btf_ptr))
 970                 return -EINVAL;
 971
 972         *btf = bpf_get_btf_vmlinux();
 973
 974         if (IS_ERR_OR_NULL(*btf))
 975                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 976
 977         if (ptr->type_id > 0)
 978                 *btf_id = ptr->type_id;
 979         else
 980                 return -EINVAL;
 981
 982         if (*btf_id > 0)
 983                 t = btf_type_by_id(*btf, *btf_id);
 984         if (*btf_id <= 0 || !t)
 985                 return -ENOENT;
 986
 987         return 0;
 988 }
 989
 990 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
 991            u32, btf_ptr_size, u64, flags)
 992 {
 993         const struct btf *btf;
 994         s32 btf_id;
 995         int ret;
 996
 997         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 998         if (ret)
 999                 return ret;
1000
1001         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1002                                       flags);
1003 }
1004
1005 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1006         .func           = bpf_snprintf_btf,
1007         .gpl_only       = false,
1008         .ret_type       = RET_INTEGER,
1009         .arg1_type      = ARG_PTR_TO_MEM,
1010         .arg2_type      = ARG_CONST_SIZE,
1011         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1012         .arg4_type      = ARG_CONST_SIZE,
1013         .arg5_type      = ARG_ANYTHING,
1014 };
1015
1016 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1017 {
1018         /* This helper call is inlined by verifier. */
1019         return ((u64 *)ctx)[-2];
1020 }
1021
1022 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1023         .func           = bpf_get_func_ip_tracing,
1024         .gpl_only       = true,
1025         .ret_type       = RET_INTEGER,
1026         .arg1_type      = ARG_PTR_TO_CTX,
1027 };
1028
1029 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1030 {
1031         struct kprobe *kp = kprobe_running();
1032
1033         return kp ? (uintptr_t)kp->addr : 0;
1034 }
1035
1036 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1037         .func           = bpf_get_func_ip_kprobe,
1038         .gpl_only       = true,
1039         .ret_type       = RET_INTEGER,
1040         .arg1_type      = ARG_PTR_TO_CTX,
1041 };
1042
1043 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1044 {
1045         return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1046 }
1047
1048 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1049         .func           = bpf_get_func_ip_kprobe_multi,
1050         .gpl_only       = false,
1051         .ret_type       = RET_INTEGER,
1052         .arg1_type      = ARG_PTR_TO_CTX,
1053 };
1054
1055 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1056 {
1057         return bpf_kprobe_multi_cookie(current->bpf_ctx);
1058 }
1059
1060 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1061         .func           = bpf_get_attach_cookie_kprobe_multi,
1062         .gpl_only       = false,
1063         .ret_type       = RET_INTEGER,
1064         .arg1_type      = ARG_PTR_TO_CTX,
1065 };
1066
1067 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1068 {
1069         struct bpf_trace_run_ctx *run_ctx;
1070
1071         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1072         return run_ctx->bpf_cookie;
1073 }
1074
1075 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1076         .func           = bpf_get_attach_cookie_trace,
1077         .gpl_only       = false,
1078         .ret_type       = RET_INTEGER,
1079         .arg1_type      = ARG_PTR_TO_CTX,
1080 };
1081
1082 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1083 {
1084         return ctx->event->bpf_cookie;
1085 }
1086
1087 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1088         .func           = bpf_get_attach_cookie_pe,
1089         .gpl_only       = false,
1090         .ret_type       = RET_INTEGER,
1091         .arg1_type      = ARG_PTR_TO_CTX,
1092 };
1093
1094 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1095 {
1096         struct bpf_trace_run_ctx *run_ctx;
1097
1098         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1099         return run_ctx->bpf_cookie;
1100 }
1101
1102 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1103         .func           = bpf_get_attach_cookie_tracing,
1104         .gpl_only       = false,
1105         .ret_type       = RET_INTEGER,
1106         .arg1_type      = ARG_PTR_TO_CTX,
1107 };
1108
1109 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1110 {
1111 #ifndef CONFIG_X86
1112         return -ENOENT;
1113 #else
1114         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1115         u32 entry_cnt = size / br_entry_size;
1116
1117         entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1118
1119         if (unlikely(flags))
1120                 return -EINVAL;
1121
1122         if (!entry_cnt)
1123                 return -ENOENT;
1124
1125         return entry_cnt * br_entry_size;
1126 #endif
1127 }
1128
1129 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1130         .func           = bpf_get_branch_snapshot,
1131         .gpl_only       = true,
1132         .ret_type       = RET_INTEGER,
1133         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1134         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1135 };
1136
1137 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1138 {
1139         /* This helper call is inlined by verifier. */
1140         u64 nr_args = ((u64 *)ctx)[-1];
1141
1142         if ((u64) n >= nr_args)
1143                 return -EINVAL;
1144         *value = ((u64 *)ctx)[n];
1145         return 0;
1146 }
1147
1148 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1149         .func           = get_func_arg,
1150         .ret_type       = RET_INTEGER,
1151         .arg1_type      = ARG_PTR_TO_CTX,
1152         .arg2_type      = ARG_ANYTHING,
1153         .arg3_type      = ARG_PTR_TO_LONG,
1154 };
1155
1156 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1157 {
1158         /* This helper call is inlined by verifier. */
1159         u64 nr_args = ((u64 *)ctx)[-1];
1160
1161         *value = ((u64 *)ctx)[nr_args];
1162         return 0;
1163 }
1164
1165 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1166         .func           = get_func_ret,
1167         .ret_type       = RET_INTEGER,
1168         .arg1_type      = ARG_PTR_TO_CTX,
1169         .arg2_type      = ARG_PTR_TO_LONG,
1170 };
1171
1172 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1173 {
1174         /* This helper call is inlined by verifier. */
1175         return ((u64 *)ctx)[-1];
1176 }
1177
1178 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1179         .func           = get_func_arg_cnt,
1180         .ret_type       = RET_INTEGER,
1181         .arg1_type      = ARG_PTR_TO_CTX,
1182 };
1183
1184 static const struct bpf_func_proto *
1185 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1186 {
1187         switch (func_id) {
1188         case BPF_FUNC_map_lookup_elem:
1189                 return &bpf_map_lookup_elem_proto;
1190         case BPF_FUNC_map_update_elem:
1191                 return &bpf_map_update_elem_proto;
1192         case BPF_FUNC_map_delete_elem:
1193                 return &bpf_map_delete_elem_proto;
1194         case BPF_FUNC_map_push_elem:
1195                 return &bpf_map_push_elem_proto;
1196         case BPF_FUNC_map_pop_elem:
1197                 return &bpf_map_pop_elem_proto;
1198         case BPF_FUNC_map_peek_elem:
1199                 return &bpf_map_peek_elem_proto;
1200         case BPF_FUNC_map_lookup_percpu_elem:
1201                 return &bpf_map_lookup_percpu_elem_proto;
1202         case BPF_FUNC_ktime_get_ns:
1203                 return &bpf_ktime_get_ns_proto;
1204         case BPF_FUNC_ktime_get_boot_ns:
1205                 return &bpf_ktime_get_boot_ns_proto;
1206         case BPF_FUNC_tail_call:
1207                 return &bpf_tail_call_proto;
1208         case BPF_FUNC_get_current_pid_tgid:
1209                 return &bpf_get_current_pid_tgid_proto;
1210         case BPF_FUNC_get_current_task:
1211                 return &bpf_get_current_task_proto;
1212         case BPF_FUNC_get_current_task_btf:
1213                 return &bpf_get_current_task_btf_proto;
1214         case BPF_FUNC_task_pt_regs:
1215                 return &bpf_task_pt_regs_proto;
1216         case BPF_FUNC_get_current_uid_gid:
1217                 return &bpf_get_current_uid_gid_proto;
1218         case BPF_FUNC_get_current_comm:
1219                 return &bpf_get_current_comm_proto;
1220         case BPF_FUNC_trace_printk:
1221                 return bpf_get_trace_printk_proto();
1222         case BPF_FUNC_get_smp_processor_id:
1223                 return &bpf_get_smp_processor_id_proto;
1224         case BPF_FUNC_get_numa_node_id:
1225                 return &bpf_get_numa_node_id_proto;
1226         case BPF_FUNC_perf_event_read:
1227                 return &bpf_perf_event_read_proto;
1228         case BPF_FUNC_current_task_under_cgroup:
1229                 return &bpf_current_task_under_cgroup_proto;
1230         case BPF_FUNC_get_prandom_u32:
1231                 return &bpf_get_prandom_u32_proto;
1232         case BPF_FUNC_probe_write_user:
1233                 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1234                        NULL : bpf_get_probe_write_proto();
1235         case BPF_FUNC_probe_read_user:
1236                 return &bpf_probe_read_user_proto;
1237         case BPF_FUNC_probe_read_kernel:
1238                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1239                        NULL : &bpf_probe_read_kernel_proto;
1240         case BPF_FUNC_probe_read_user_str:
1241                 return &bpf_probe_read_user_str_proto;
1242         case BPF_FUNC_probe_read_kernel_str:
1243                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1244                        NULL : &bpf_probe_read_kernel_str_proto;
1245 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1246         case BPF_FUNC_probe_read:
1247                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1248                        NULL : &bpf_probe_read_compat_proto;
1249         case BPF_FUNC_probe_read_str:
1250                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1251                        NULL : &bpf_probe_read_compat_str_proto;
1252 #endif
1253 #ifdef CONFIG_CGROUPS
1254         case BPF_FUNC_get_current_cgroup_id:
1255                 return &bpf_get_current_cgroup_id_proto;
1256         case BPF_FUNC_get_current_ancestor_cgroup_id:
1257                 return &bpf_get_current_ancestor_cgroup_id_proto;
1258 #endif
1259         case BPF_FUNC_send_signal:
1260                 return &bpf_send_signal_proto;
1261         case BPF_FUNC_send_signal_thread:
1262                 return &bpf_send_signal_thread_proto;
1263         case BPF_FUNC_perf_event_read_value:
1264                 return &bpf_perf_event_read_value_proto;
1265         case BPF_FUNC_get_ns_current_pid_tgid:
1266                 return &bpf_get_ns_current_pid_tgid_proto;
1267         case BPF_FUNC_ringbuf_output:
1268                 return &bpf_ringbuf_output_proto;
1269         case BPF_FUNC_ringbuf_reserve:
1270                 return &bpf_ringbuf_reserve_proto;
1271         case BPF_FUNC_ringbuf_submit:
1272                 return &bpf_ringbuf_submit_proto;
1273         case BPF_FUNC_ringbuf_discard:
1274                 return &bpf_ringbuf_discard_proto;
1275         case BPF_FUNC_ringbuf_query:
1276                 return &bpf_ringbuf_query_proto;
1277         case BPF_FUNC_jiffies64:
1278                 return &bpf_jiffies64_proto;
1279         case BPF_FUNC_get_task_stack:
1280                 return &bpf_get_task_stack_proto;
1281         case BPF_FUNC_copy_from_user:
1282                 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1283         case BPF_FUNC_copy_from_user_task:
1284                 return prog->aux->sleepable ? &bpf_copy_from_user_task_proto : NULL;
1285         case BPF_FUNC_snprintf_btf:
1286                 return &bpf_snprintf_btf_proto;
1287         case BPF_FUNC_per_cpu_ptr:
1288                 return &bpf_per_cpu_ptr_proto;
1289         case BPF_FUNC_this_cpu_ptr:
1290                 return &bpf_this_cpu_ptr_proto;
1291         case BPF_FUNC_task_storage_get:
1292                 return &bpf_task_storage_get_proto;
1293         case BPF_FUNC_task_storage_delete:
1294                 return &bpf_task_storage_delete_proto;
1295         case BPF_FUNC_for_each_map_elem:
1296                 return &bpf_for_each_map_elem_proto;
1297         case BPF_FUNC_snprintf:
1298                 return &bpf_snprintf_proto;
1299         case BPF_FUNC_get_func_ip:
1300                 return &bpf_get_func_ip_proto_tracing;
1301         case BPF_FUNC_get_branch_snapshot:
1302                 return &bpf_get_branch_snapshot_proto;
1303         case BPF_FUNC_find_vma:
1304                 return &bpf_find_vma_proto;
1305         case BPF_FUNC_trace_vprintk:
1306                 return bpf_get_trace_vprintk_proto();
1307         default:
1308                 return bpf_base_func_proto(func_id);
1309         }
1310 }
1311
1312 static const struct bpf_func_proto *
1313 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1314 {
1315         switch (func_id) {
1316         case BPF_FUNC_perf_event_output:
1317                 return &bpf_perf_event_output_proto;
1318         case BPF_FUNC_get_stackid:
1319                 return &bpf_get_stackid_proto;
1320         case BPF_FUNC_get_stack:
1321                 return &bpf_get_stack_proto;
1322 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1323         case BPF_FUNC_override_return:
1324                 return &bpf_override_return_proto;
1325 #endif
1326         case BPF_FUNC_get_func_ip:
1327                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1328                         &bpf_get_func_ip_proto_kprobe_multi :
1329                         &bpf_get_func_ip_proto_kprobe;
1330         case BPF_FUNC_get_attach_cookie:
1331                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1332                         &bpf_get_attach_cookie_proto_kmulti :
1333                         &bpf_get_attach_cookie_proto_trace;
1334         default:
1335                 return bpf_tracing_func_proto(func_id, prog);
1336         }
1337 }
1338
1339 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1340 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1341                                         const struct bpf_prog *prog,
1342                                         struct bpf_insn_access_aux *info)
1343 {
1344         if (off < 0 || off >= sizeof(struct pt_regs))
1345                 return false;
1346         if (type != BPF_READ)
1347                 return false;
1348         if (off % size != 0)
1349                 return false;
1350         /*
1351          * Assertion for 32 bit to make sure last 8 byte access
1352          * (BPF_DW) to the last 4 byte member is disallowed.
1353          */
1354         if (off + size > sizeof(struct pt_regs))
1355                 return false;
1356
1357         return true;
1358 }
1359
1360 const struct bpf_verifier_ops kprobe_verifier_ops = {
1361         .get_func_proto  = kprobe_prog_func_proto,
1362         .is_valid_access = kprobe_prog_is_valid_access,
1363 };
1364
1365 const struct bpf_prog_ops kprobe_prog_ops = {
1366 };
1367
1368 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1369            u64, flags, void *, data, u64, size)
1370 {
1371         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1372
1373         /*
1374          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1375          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1376          * from there and call the same bpf_perf_event_output() helper inline.
1377          */
1378         return ____bpf_perf_event_output(regs, map, flags, data, size);
1379 }
1380
1381 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1382         .func           = bpf_perf_event_output_tp,
1383         .gpl_only       = true,
1384         .ret_type       = RET_INTEGER,
1385         .arg1_type      = ARG_PTR_TO_CTX,
1386         .arg2_type      = ARG_CONST_MAP_PTR,
1387         .arg3_type      = ARG_ANYTHING,
1388         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1389         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1390 };
1391
1392 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1393            u64, flags)
1394 {
1395         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1396
1397         /*
1398          * Same comment as in bpf_perf_event_output_tp(), only that this time
1399          * the other helper's function body cannot be inlined due to being
1400          * external, thus we need to call raw helper function.
1401          */
1402         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1403                                flags, 0, 0);
1404 }
1405
1406 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1407         .func           = bpf_get_stackid_tp,
1408         .gpl_only       = true,
1409         .ret_type       = RET_INTEGER,
1410         .arg1_type      = ARG_PTR_TO_CTX,
1411         .arg2_type      = ARG_CONST_MAP_PTR,
1412         .arg3_type      = ARG_ANYTHING,
1413 };
1414
1415 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1416            u64, flags)
1417 {
1418         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1419
1420         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1421                              (unsigned long) size, flags, 0);
1422 }
1423
1424 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1425         .func           = bpf_get_stack_tp,
1426         .gpl_only       = true,
1427         .ret_type       = RET_INTEGER,
1428         .arg1_type      = ARG_PTR_TO_CTX,
1429         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1430         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1431         .arg4_type      = ARG_ANYTHING,
1432 };
1433
1434 static const struct bpf_func_proto *
1435 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1436 {
1437         switch (func_id) {
1438         case BPF_FUNC_perf_event_output:
1439                 return &bpf_perf_event_output_proto_tp;
1440         case BPF_FUNC_get_stackid:
1441                 return &bpf_get_stackid_proto_tp;
1442         case BPF_FUNC_get_stack:
1443                 return &bpf_get_stack_proto_tp;
1444         case BPF_FUNC_get_attach_cookie:
1445                 return &bpf_get_attach_cookie_proto_trace;
1446         default:
1447                 return bpf_tracing_func_proto(func_id, prog);
1448         }
1449 }
1450
1451 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1452                                     const struct bpf_prog *prog,
1453                                     struct bpf_insn_access_aux *info)
1454 {
1455         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1456                 return false;
1457         if (type != BPF_READ)
1458                 return false;
1459         if (off % size != 0)
1460                 return false;
1461
1462         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1463         return true;
1464 }
1465
1466 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1467         .get_func_proto  = tp_prog_func_proto,
1468         .is_valid_access = tp_prog_is_valid_access,
1469 };
1470
1471 const struct bpf_prog_ops tracepoint_prog_ops = {
1472 };
1473
1474 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1475            struct bpf_perf_event_value *, buf, u32, size)
1476 {
1477         int err = -EINVAL;
1478
1479         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1480                 goto clear;
1481         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1482                                     &buf->running);
1483         if (unlikely(err))
1484                 goto clear;
1485         return 0;
1486 clear:
1487         memset(buf, 0, size);
1488         return err;
1489 }
1490
1491 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1492          .func           = bpf_perf_prog_read_value,
1493          .gpl_only       = true,
1494          .ret_type       = RET_INTEGER,
1495          .arg1_type      = ARG_PTR_TO_CTX,
1496          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1497          .arg3_type      = ARG_CONST_SIZE,
1498 };
1499
1500 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1501            void *, buf, u32, size, u64, flags)
1502 {
1503         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1504         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1505         u32 to_copy;
1506
1507         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1508                 return -EINVAL;
1509
1510         if (unlikely(!br_stack))
1511                 return -ENOENT;
1512
1513         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1514                 return br_stack->nr * br_entry_size;
1515
1516         if (!buf || (size % br_entry_size != 0))
1517                 return -EINVAL;
1518
1519         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1520         memcpy(buf, br_stack->entries, to_copy);
1521
1522         return to_copy;
1523 }
1524
1525 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1526         .func           = bpf_read_branch_records,
1527         .gpl_only       = true,
1528         .ret_type       = RET_INTEGER,
1529         .arg1_type      = ARG_PTR_TO_CTX,
1530         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1531         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1532         .arg4_type      = ARG_ANYTHING,
1533 };
1534
1535 static const struct bpf_func_proto *
1536 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1537 {
1538         switch (func_id) {
1539         case BPF_FUNC_perf_event_output:
1540                 return &bpf_perf_event_output_proto_tp;
1541         case BPF_FUNC_get_stackid:
1542                 return &bpf_get_stackid_proto_pe;
1543         case BPF_FUNC_get_stack:
1544                 return &bpf_get_stack_proto_pe;
1545         case BPF_FUNC_perf_prog_read_value:
1546                 return &bpf_perf_prog_read_value_proto;
1547         case BPF_FUNC_read_branch_records:
1548                 return &bpf_read_branch_records_proto;
1549         case BPF_FUNC_get_attach_cookie:
1550                 return &bpf_get_attach_cookie_proto_pe;
1551         default:
1552                 return bpf_tracing_func_proto(func_id, prog);
1553         }
1554 }
1555
1556 /*
1557  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1558  * to avoid potential recursive reuse issue when/if tracepoints are added
1559  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1560  *
1561  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1562  * in normal, irq, and nmi context.
1563  */
1564 struct bpf_raw_tp_regs {
1565         struct pt_regs regs[3];
1566 };
1567 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1568 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1569 static struct pt_regs *get_bpf_raw_tp_regs(void)
1570 {
1571         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1572         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1573
1574         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1575                 this_cpu_dec(bpf_raw_tp_nest_level);
1576                 return ERR_PTR(-EBUSY);
1577         }
1578
1579         return &tp_regs->regs[nest_level - 1];
1580 }
1581
1582 static void put_bpf_raw_tp_regs(void)
1583 {
1584         this_cpu_dec(bpf_raw_tp_nest_level);
1585 }
1586
1587 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1588            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1589 {
1590         struct pt_regs *regs = get_bpf_raw_tp_regs();
1591         int ret;
1592
1593         if (IS_ERR(regs))
1594                 return PTR_ERR(regs);
1595
1596         perf_fetch_caller_regs(regs);
1597         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1598
1599         put_bpf_raw_tp_regs();
1600         return ret;
1601 }
1602
1603 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1604         .func           = bpf_perf_event_output_raw_tp,
1605         .gpl_only       = true,
1606         .ret_type       = RET_INTEGER,
1607         .arg1_type      = ARG_PTR_TO_CTX,
1608         .arg2_type      = ARG_CONST_MAP_PTR,
1609         .arg3_type      = ARG_ANYTHING,
1610         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1611         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1612 };
1613
1614 extern const struct bpf_func_proto bpf_skb_output_proto;
1615 extern const struct bpf_func_proto bpf_xdp_output_proto;
1616 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1617
1618 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1619            struct bpf_map *, map, u64, flags)
1620 {
1621         struct pt_regs *regs = get_bpf_raw_tp_regs();
1622         int ret;
1623
1624         if (IS_ERR(regs))
1625                 return PTR_ERR(regs);
1626
1627         perf_fetch_caller_regs(regs);
1628         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1629         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1630                               flags, 0, 0);
1631         put_bpf_raw_tp_regs();
1632         return ret;
1633 }
1634
1635 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1636         .func           = bpf_get_stackid_raw_tp,
1637         .gpl_only       = true,
1638         .ret_type       = RET_INTEGER,
1639         .arg1_type      = ARG_PTR_TO_CTX,
1640         .arg2_type      = ARG_CONST_MAP_PTR,
1641         .arg3_type      = ARG_ANYTHING,
1642 };
1643
1644 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1645            void *, buf, u32, size, u64, flags)
1646 {
1647         struct pt_regs *regs = get_bpf_raw_tp_regs();
1648         int ret;
1649
1650         if (IS_ERR(regs))
1651                 return PTR_ERR(regs);
1652
1653         perf_fetch_caller_regs(regs);
1654         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1655                             (unsigned long) size, flags, 0);
1656         put_bpf_raw_tp_regs();
1657         return ret;
1658 }
1659
1660 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1661         .func           = bpf_get_stack_raw_tp,
1662         .gpl_only       = true,
1663         .ret_type       = RET_INTEGER,
1664         .arg1_type      = ARG_PTR_TO_CTX,
1665         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1666         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1667         .arg4_type      = ARG_ANYTHING,
1668 };
1669
1670 static const struct bpf_func_proto *
1671 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1672 {
1673         switch (func_id) {
1674         case BPF_FUNC_perf_event_output:
1675                 return &bpf_perf_event_output_proto_raw_tp;
1676         case BPF_FUNC_get_stackid:
1677                 return &bpf_get_stackid_proto_raw_tp;
1678         case BPF_FUNC_get_stack:
1679                 return &bpf_get_stack_proto_raw_tp;
1680         default:
1681                 return bpf_tracing_func_proto(func_id, prog);
1682         }
1683 }
1684
1685 const struct bpf_func_proto *
1686 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1687 {
1688         const struct bpf_func_proto *fn;
1689
1690         switch (func_id) {
1691 #ifdef CONFIG_NET
1692         case BPF_FUNC_skb_output:
1693                 return &bpf_skb_output_proto;
1694         case BPF_FUNC_xdp_output:
1695                 return &bpf_xdp_output_proto;
1696         case BPF_FUNC_skc_to_tcp6_sock:
1697                 return &bpf_skc_to_tcp6_sock_proto;
1698         case BPF_FUNC_skc_to_tcp_sock:
1699                 return &bpf_skc_to_tcp_sock_proto;
1700         case BPF_FUNC_skc_to_tcp_timewait_sock:
1701                 return &bpf_skc_to_tcp_timewait_sock_proto;
1702         case BPF_FUNC_skc_to_tcp_request_sock:
1703                 return &bpf_skc_to_tcp_request_sock_proto;
1704         case BPF_FUNC_skc_to_udp6_sock:
1705                 return &bpf_skc_to_udp6_sock_proto;
1706         case BPF_FUNC_skc_to_unix_sock:
1707                 return &bpf_skc_to_unix_sock_proto;
1708         case BPF_FUNC_skc_to_mptcp_sock:
1709                 return &bpf_skc_to_mptcp_sock_proto;
1710         case BPF_FUNC_sk_storage_get:
1711                 return &bpf_sk_storage_get_tracing_proto;
1712         case BPF_FUNC_sk_storage_delete:
1713                 return &bpf_sk_storage_delete_tracing_proto;
1714         case BPF_FUNC_sock_from_file:
1715                 return &bpf_sock_from_file_proto;
1716         case BPF_FUNC_get_socket_cookie:
1717                 return &bpf_get_socket_ptr_cookie_proto;
1718         case BPF_FUNC_xdp_get_buff_len:
1719                 return &bpf_xdp_get_buff_len_trace_proto;
1720 #endif
1721         case BPF_FUNC_seq_printf:
1722                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1723                        &bpf_seq_printf_proto :
1724                        NULL;
1725         case BPF_FUNC_seq_write:
1726                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1727                        &bpf_seq_write_proto :
1728                        NULL;
1729         case BPF_FUNC_seq_printf_btf:
1730                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1731                        &bpf_seq_printf_btf_proto :
1732                        NULL;
1733         case BPF_FUNC_d_path:
1734                 return &bpf_d_path_proto;
1735         case BPF_FUNC_get_func_arg:
1736                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1737         case BPF_FUNC_get_func_ret:
1738                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1739         case BPF_FUNC_get_func_arg_cnt:
1740                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1741         case BPF_FUNC_get_attach_cookie:
1742                 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
1743         default:
1744                 fn = raw_tp_prog_func_proto(func_id, prog);
1745                 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1746                         fn = bpf_iter_get_func_proto(func_id, prog);
1747                 return fn;
1748         }
1749 }
1750
1751 static bool raw_tp_prog_is_valid_access(int off, int size,
1752                                         enum bpf_access_type type,
1753                                         const struct bpf_prog *prog,
1754                                         struct bpf_insn_access_aux *info)
1755 {
1756         return bpf_tracing_ctx_access(off, size, type);
1757 }
1758
1759 static bool tracing_prog_is_valid_access(int off, int size,
1760                                          enum bpf_access_type type,
1761                                          const struct bpf_prog *prog,
1762                                          struct bpf_insn_access_aux *info)
1763 {
1764         return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1765 }
1766
1767 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1768                                      const union bpf_attr *kattr,
1769                                      union bpf_attr __user *uattr)
1770 {
1771         return -ENOTSUPP;
1772 }
1773
1774 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1775         .get_func_proto  = raw_tp_prog_func_proto,
1776         .is_valid_access = raw_tp_prog_is_valid_access,
1777 };
1778
1779 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1780 #ifdef CONFIG_NET
1781         .test_run = bpf_prog_test_run_raw_tp,
1782 #endif
1783 };
1784
1785 const struct bpf_verifier_ops tracing_verifier_ops = {
1786         .get_func_proto  = tracing_prog_func_proto,
1787         .is_valid_access = tracing_prog_is_valid_access,
1788 };
1789
1790 const struct bpf_prog_ops tracing_prog_ops = {
1791         .test_run = bpf_prog_test_run_tracing,
1792 };
1793
1794 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1795                                                  enum bpf_access_type type,
1796                                                  const struct bpf_prog *prog,
1797                                                  struct bpf_insn_access_aux *info)
1798 {
1799         if (off == 0) {
1800                 if (size != sizeof(u64) || type != BPF_READ)
1801                         return false;
1802                 info->reg_type = PTR_TO_TP_BUFFER;
1803         }
1804         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1805 }
1806
1807 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1808         .get_func_proto  = raw_tp_prog_func_proto,
1809         .is_valid_access = raw_tp_writable_prog_is_valid_access,
1810 };
1811
1812 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1813 };
1814
1815 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1816                                     const struct bpf_prog *prog,
1817                                     struct bpf_insn_access_aux *info)
1818 {
1819         const int size_u64 = sizeof(u64);
1820
1821         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1822                 return false;
1823         if (type != BPF_READ)
1824                 return false;
1825         if (off % size != 0) {
1826                 if (sizeof(unsigned long) != 4)
1827                         return false;
1828                 if (size != 8)
1829                         return false;
1830                 if (off % size != 4)
1831                         return false;
1832         }
1833
1834         switch (off) {
1835         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1836                 bpf_ctx_record_field_size(info, size_u64);
1837                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1838                         return false;
1839                 break;
1840         case bpf_ctx_range(struct bpf_perf_event_data, addr):
1841                 bpf_ctx_record_field_size(info, size_u64);
1842                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1843                         return false;
1844                 break;
1845         default:
1846                 if (size != sizeof(long))
1847                         return false;
1848         }
1849
1850         return true;
1851 }
1852
1853 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1854                                       const struct bpf_insn *si,
1855                                       struct bpf_insn *insn_buf,
1856                                       struct bpf_prog *prog, u32 *target_size)
1857 {
1858         struct bpf_insn *insn = insn_buf;
1859
1860         switch (si->off) {
1861         case offsetof(struct bpf_perf_event_data, sample_period):
1862                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1863                                                        data), si->dst_reg, si->src_reg,
1864                                       offsetof(struct bpf_perf_event_data_kern, data));
1865                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1866                                       bpf_target_off(struct perf_sample_data, period, 8,
1867                                                      target_size));
1868                 break;
1869         case offsetof(struct bpf_perf_event_data, addr):
1870                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1871                                                        data), si->dst_reg, si->src_reg,
1872                                       offsetof(struct bpf_perf_event_data_kern, data));
1873                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1874                                       bpf_target_off(struct perf_sample_data, addr, 8,
1875                                                      target_size));
1876                 break;
1877         default:
1878                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1879                                                        regs), si->dst_reg, si->src_reg,
1880                                       offsetof(struct bpf_perf_event_data_kern, regs));
1881                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1882                                       si->off);
1883                 break;
1884         }
1885
1886         return insn - insn_buf;
1887 }
1888
1889 const struct bpf_verifier_ops perf_event_verifier_ops = {
1890         .get_func_proto         = pe_prog_func_proto,
1891         .is_valid_access        = pe_prog_is_valid_access,
1892         .convert_ctx_access     = pe_prog_convert_ctx_access,
1893 };
1894
1895 const struct bpf_prog_ops perf_event_prog_ops = {
1896 };
1897
1898 static DEFINE_MUTEX(bpf_event_mutex);
1899
1900 #define BPF_TRACE_MAX_PROGS 64
1901
1902 int perf_event_attach_bpf_prog(struct perf_event *event,
1903                                struct bpf_prog *prog,
1904                                u64 bpf_cookie)
1905 {
1906         struct bpf_prog_array *old_array;
1907         struct bpf_prog_array *new_array;
1908         int ret = -EEXIST;
1909
1910         /*
1911          * Kprobe override only works if they are on the function entry,
1912          * and only if they are on the opt-in list.
1913          */
1914         if (prog->kprobe_override &&
1915             (!trace_kprobe_on_func_entry(event->tp_event) ||
1916              !trace_kprobe_error_injectable(event->tp_event)))
1917                 return -EINVAL;
1918
1919         mutex_lock(&bpf_event_mutex);
1920
1921         if (event->prog)
1922                 goto unlock;
1923
1924         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1925         if (old_array &&
1926             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1927                 ret = -E2BIG;
1928                 goto unlock;
1929         }
1930
1931         ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
1932         if (ret < 0)
1933                 goto unlock;
1934
1935         /* set the new array to event->tp_event and set event->prog */
1936         event->prog = prog;
1937         event->bpf_cookie = bpf_cookie;
1938         rcu_assign_pointer(event->tp_event->prog_array, new_array);
1939         bpf_prog_array_free(old_array);
1940
1941 unlock:
1942         mutex_unlock(&bpf_event_mutex);
1943         return ret;
1944 }
1945
1946 void perf_event_detach_bpf_prog(struct perf_event *event)
1947 {
1948         struct bpf_prog_array *old_array;
1949         struct bpf_prog_array *new_array;
1950         int ret;
1951
1952         mutex_lock(&bpf_event_mutex);
1953
1954         if (!event->prog)
1955                 goto unlock;
1956
1957         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1958         ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
1959         if (ret == -ENOENT)
1960                 goto unlock;
1961         if (ret < 0) {
1962                 bpf_prog_array_delete_safe(old_array, event->prog);
1963         } else {
1964                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1965                 bpf_prog_array_free(old_array);
1966         }
1967
1968         bpf_prog_put(event->prog);
1969         event->prog = NULL;
1970
1971 unlock:
1972         mutex_unlock(&bpf_event_mutex);
1973 }
1974
1975 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1976 {
1977         struct perf_event_query_bpf __user *uquery = info;
1978         struct perf_event_query_bpf query = {};
1979         struct bpf_prog_array *progs;
1980         u32 *ids, prog_cnt, ids_len;
1981         int ret;
1982
1983         if (!perfmon_capable())
1984                 return -EPERM;
1985         if (event->attr.type != PERF_TYPE_TRACEPOINT)
1986                 return -EINVAL;
1987         if (copy_from_user(&query, uquery, sizeof(query)))
1988                 return -EFAULT;
1989
1990         ids_len = query.ids_len;
1991         if (ids_len > BPF_TRACE_MAX_PROGS)
1992                 return -E2BIG;
1993         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1994         if (!ids)
1995                 return -ENOMEM;
1996         /*
1997          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1998          * is required when user only wants to check for uquery->prog_cnt.
1999          * There is no need to check for it since the case is handled
2000          * gracefully in bpf_prog_array_copy_info.
2001          */
2002
2003         mutex_lock(&bpf_event_mutex);
2004         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2005         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2006         mutex_unlock(&bpf_event_mutex);
2007
2008         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2009             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2010                 ret = -EFAULT;
2011
2012         kfree(ids);
2013         return ret;
2014 }
2015
2016 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2017 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2018
2019 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2020 {
2021         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2022
2023         for (; btp < __stop__bpf_raw_tp; btp++) {
2024                 if (!strcmp(btp->tp->name, name))
2025                         return btp;
2026         }
2027
2028         return bpf_get_raw_tracepoint_module(name);
2029 }
2030
2031 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2032 {
2033         struct module *mod;
2034
2035         preempt_disable();
2036         mod = __module_address((unsigned long)btp);
2037         module_put(mod);
2038         preempt_enable();
2039 }
2040
2041 static __always_inline
2042 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2043 {
2044         cant_sleep();
2045         rcu_read_lock();
2046         (void) bpf_prog_run(prog, args);
2047         rcu_read_unlock();
2048 }
2049
2050 #define UNPACK(...)                     __VA_ARGS__
2051 #define REPEAT_1(FN, DL, X, ...)        FN(X)
2052 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2053 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2054 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2055 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2056 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2057 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2058 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2059 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2060 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2061 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2062 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2063 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
2064
2065 #define SARG(X)         u64 arg##X
2066 #define COPY(X)         args[X] = arg##X
2067
2068 #define __DL_COM        (,)
2069 #define __DL_SEM        (;)
2070
2071 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2072
2073 #define BPF_TRACE_DEFN_x(x)                                             \
2074         void bpf_trace_run##x(struct bpf_prog *prog,                    \
2075                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
2076         {                                                               \
2077                 u64 args[x];                                            \
2078                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
2079                 __bpf_trace_run(prog, args);                            \
2080         }                                                               \
2081         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2082 BPF_TRACE_DEFN_x(1);
2083 BPF_TRACE_DEFN_x(2);
2084 BPF_TRACE_DEFN_x(3);
2085 BPF_TRACE_DEFN_x(4);
2086 BPF_TRACE_DEFN_x(5);
2087 BPF_TRACE_DEFN_x(6);
2088 BPF_TRACE_DEFN_x(7);
2089 BPF_TRACE_DEFN_x(8);
2090 BPF_TRACE_DEFN_x(9);
2091 BPF_TRACE_DEFN_x(10);
2092 BPF_TRACE_DEFN_x(11);
2093 BPF_TRACE_DEFN_x(12);
2094
2095 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2096 {
2097         struct tracepoint *tp = btp->tp;
2098
2099         /*
2100          * check that program doesn't access arguments beyond what's
2101          * available in this tracepoint
2102          */
2103         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2104                 return -EINVAL;
2105
2106         if (prog->aux->max_tp_access > btp->writable_size)
2107                 return -EINVAL;
2108
2109         return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2110                                                    prog);
2111 }
2112
2113 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2114 {
2115         return __bpf_probe_register(btp, prog);
2116 }
2117
2118 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2119 {
2120         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2121 }
2122
2123 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2124                             u32 *fd_type, const char **buf,
2125                             u64 *probe_offset, u64 *probe_addr)
2126 {
2127         bool is_tracepoint, is_syscall_tp;
2128         struct bpf_prog *prog;
2129         int flags, err = 0;
2130
2131         prog = event->prog;
2132         if (!prog)
2133                 return -ENOENT;
2134
2135         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2136         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2137                 return -EOPNOTSUPP;
2138
2139         *prog_id = prog->aux->id;
2140         flags = event->tp_event->flags;
2141         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2142         is_syscall_tp = is_syscall_trace_event(event->tp_event);
2143
2144         if (is_tracepoint || is_syscall_tp) {
2145                 *buf = is_tracepoint ? event->tp_event->tp->name
2146                                      : event->tp_event->name;
2147                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2148                 *probe_offset = 0x0;
2149                 *probe_addr = 0x0;
2150         } else {
2151                 /* kprobe/uprobe */
2152                 err = -EOPNOTSUPP;
2153 #ifdef CONFIG_KPROBE_EVENTS
2154                 if (flags & TRACE_EVENT_FL_KPROBE)
2155                         err = bpf_get_kprobe_info(event, fd_type, buf,
2156                                                   probe_offset, probe_addr,
2157                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2158 #endif
2159 #ifdef CONFIG_UPROBE_EVENTS
2160                 if (flags & TRACE_EVENT_FL_UPROBE)
2161                         err = bpf_get_uprobe_info(event, fd_type, buf,
2162                                                   probe_offset,
2163                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2164 #endif
2165         }
2166
2167         return err;
2168 }
2169
2170 static int __init send_signal_irq_work_init(void)
2171 {
2172         int cpu;
2173         struct send_signal_irq_work *work;
2174
2175         for_each_possible_cpu(cpu) {
2176                 work = per_cpu_ptr(&send_signal_work, cpu);
2177                 init_irq_work(&work->irq_work, do_bpf_send_signal);
2178         }
2179         return 0;
2180 }
2181
2182 subsys_initcall(send_signal_irq_work_init);
2183
2184 #ifdef CONFIG_MODULES
2185 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2186                             void *module)
2187 {
2188         struct bpf_trace_module *btm, *tmp;
2189         struct module *mod = module;
2190         int ret = 0;
2191
2192         if (mod->num_bpf_raw_events == 0 ||
2193             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2194                 goto out;
2195
2196         mutex_lock(&bpf_module_mutex);
2197
2198         switch (op) {
2199         case MODULE_STATE_COMING:
2200                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2201                 if (btm) {
2202                         btm->module = module;
2203                         list_add(&btm->list, &bpf_trace_modules);
2204                 } else {
2205                         ret = -ENOMEM;
2206                 }
2207                 break;
2208         case MODULE_STATE_GOING:
2209                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2210                         if (btm->module == module) {
2211                                 list_del(&btm->list);
2212                                 kfree(btm);
2213                                 break;
2214                         }
2215                 }
2216                 break;
2217         }
2218
2219         mutex_unlock(&bpf_module_mutex);
2220
2221 out:
2222         return notifier_from_errno(ret);
2223 }
2224
2225 static struct notifier_block bpf_module_nb = {
2226         .notifier_call = bpf_event_notify,
2227 };
2228
2229 static int __init bpf_event_init(void)
2230 {
2231         register_module_notifier(&bpf_module_nb);
2232         return 0;
2233 }
2234
2235 fs_initcall(bpf_event_init);
2236 #endif /* CONFIG_MODULES */
2237
2238 #ifdef CONFIG_FPROBE
2239 struct bpf_kprobe_multi_link {
2240         struct bpf_link link;
2241         struct fprobe fp;
2242         unsigned long *addrs;
2243         u64 *cookies;
2244         u32 cnt;
2245 };
2246
2247 struct bpf_kprobe_multi_run_ctx {
2248         struct bpf_run_ctx run_ctx;
2249         struct bpf_kprobe_multi_link *link;
2250         unsigned long entry_ip;
2251 };
2252
2253 struct user_syms {
2254         const char **syms;
2255         char *buf;
2256 };
2257
2258 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2259 {
2260         unsigned long __user usymbol;
2261         const char **syms = NULL;
2262         char *buf = NULL, *p;
2263         int err = -ENOMEM;
2264         unsigned int i;
2265
2266         syms = kvmalloc(cnt * sizeof(*syms), GFP_KERNEL);
2267         if (!syms)
2268                 goto error;
2269
2270         buf = kvmalloc(cnt * KSYM_NAME_LEN, GFP_KERNEL);
2271         if (!buf)
2272                 goto error;
2273
2274         for (p = buf, i = 0; i < cnt; i++) {
2275                 if (__get_user(usymbol, usyms + i)) {
2276                         err = -EFAULT;
2277                         goto error;
2278                 }
2279                 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2280                 if (err == KSYM_NAME_LEN)
2281                         err = -E2BIG;
2282                 if (err < 0)
2283                         goto error;
2284                 syms[i] = p;
2285                 p += err + 1;
2286         }
2287
2288         us->syms = syms;
2289         us->buf = buf;
2290         return 0;
2291
2292 error:
2293         if (err) {
2294                 kvfree(syms);
2295                 kvfree(buf);
2296         }
2297         return err;
2298 }
2299
2300 static void free_user_syms(struct user_syms *us)
2301 {
2302         kvfree(us->syms);
2303         kvfree(us->buf);
2304 }
2305
2306 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2307 {
2308         struct bpf_kprobe_multi_link *kmulti_link;
2309
2310         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2311         unregister_fprobe(&kmulti_link->fp);
2312 }
2313
2314 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2315 {
2316         struct bpf_kprobe_multi_link *kmulti_link;
2317
2318         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2319         kvfree(kmulti_link->addrs);
2320         kvfree(kmulti_link->cookies);
2321         kfree(kmulti_link);
2322 }
2323
2324 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2325         .release = bpf_kprobe_multi_link_release,
2326         .dealloc = bpf_kprobe_multi_link_dealloc,
2327 };
2328
2329 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2330 {
2331         const struct bpf_kprobe_multi_link *link = priv;
2332         unsigned long *addr_a = a, *addr_b = b;
2333         u64 *cookie_a, *cookie_b;
2334
2335         cookie_a = link->cookies + (addr_a - link->addrs);
2336         cookie_b = link->cookies + (addr_b - link->addrs);
2337
2338         /* swap addr_a/addr_b and cookie_a/cookie_b values */
2339         swap(*addr_a, *addr_b);
2340         swap(*cookie_a, *cookie_b);
2341 }
2342
2343 static int __bpf_kprobe_multi_cookie_cmp(const void *a, const void *b)
2344 {
2345         const unsigned long *addr_a = a, *addr_b = b;
2346
2347         if (*addr_a == *addr_b)
2348                 return 0;
2349         return *addr_a < *addr_b ? -1 : 1;
2350 }
2351
2352 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2353 {
2354         return __bpf_kprobe_multi_cookie_cmp(a, b);
2355 }
2356
2357 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2358 {
2359         struct bpf_kprobe_multi_run_ctx *run_ctx;
2360         struct bpf_kprobe_multi_link *link;
2361         u64 *cookie, entry_ip;
2362         unsigned long *addr;
2363
2364         if (WARN_ON_ONCE(!ctx))
2365                 return 0;
2366         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2367         link = run_ctx->link;
2368         if (!link->cookies)
2369                 return 0;
2370         entry_ip = run_ctx->entry_ip;
2371         addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2372                        __bpf_kprobe_multi_cookie_cmp);
2373         if (!addr)
2374                 return 0;
2375         cookie = link->cookies + (addr - link->addrs);
2376         return *cookie;
2377 }
2378
2379 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2380 {
2381         struct bpf_kprobe_multi_run_ctx *run_ctx;
2382
2383         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2384         return run_ctx->entry_ip;
2385 }
2386
2387 static int
2388 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2389                            unsigned long entry_ip, struct pt_regs *regs)
2390 {
2391         struct bpf_kprobe_multi_run_ctx run_ctx = {
2392                 .link = link,
2393                 .entry_ip = entry_ip,
2394         };
2395         struct bpf_run_ctx *old_run_ctx;
2396         int err;
2397
2398         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2399                 err = 0;
2400                 goto out;
2401         }
2402
2403         migrate_disable();
2404         rcu_read_lock();
2405         old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2406         err = bpf_prog_run(link->link.prog, regs);
2407         bpf_reset_run_ctx(old_run_ctx);
2408         rcu_read_unlock();
2409         migrate_enable();
2410
2411  out:
2412         __this_cpu_dec(bpf_prog_active);
2413         return err;
2414 }
2415
2416 static void
2417 kprobe_multi_link_handler(struct fprobe *fp, unsigned long entry_ip,
2418                           struct pt_regs *regs)
2419 {
2420         struct bpf_kprobe_multi_link *link;
2421
2422         link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2423         kprobe_multi_link_prog_run(link, entry_ip, regs);
2424 }
2425
2426 static int symbols_cmp(const void *a, const void *b)
2427 {
2428         const char **str_a = (const char **) a;
2429         const char **str_b = (const char **) b;
2430
2431         return strcmp(*str_a, *str_b);
2432 }
2433
2434 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2435 {
2436         struct bpf_kprobe_multi_link *link = NULL;
2437         struct bpf_link_primer link_primer;
2438         void __user *ucookies;
2439         unsigned long *addrs;
2440         u32 flags, cnt, size;
2441         void __user *uaddrs;
2442         u64 *cookies = NULL;
2443         void __user *usyms;
2444         int err;
2445
2446         /* no support for 32bit archs yet */
2447         if (sizeof(u64) != sizeof(void *))
2448                 return -EOPNOTSUPP;
2449
2450         if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2451                 return -EINVAL;
2452
2453         flags = attr->link_create.kprobe_multi.flags;
2454         if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2455                 return -EINVAL;
2456
2457         uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2458         usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2459         if (!!uaddrs == !!usyms)
2460                 return -EINVAL;
2461
2462         cnt = attr->link_create.kprobe_multi.cnt;
2463         if (!cnt)
2464                 return -EINVAL;
2465
2466         size = cnt * sizeof(*addrs);
2467         addrs = kvmalloc(size, GFP_KERNEL);
2468         if (!addrs)
2469                 return -ENOMEM;
2470
2471         if (uaddrs) {
2472                 if (copy_from_user(addrs, uaddrs, size)) {
2473                         err = -EFAULT;
2474                         goto error;
2475                 }
2476         } else {
2477                 struct user_syms us;
2478
2479                 err = copy_user_syms(&us, usyms, cnt);
2480                 if (err)
2481                         goto error;
2482
2483                 sort(us.syms, cnt, sizeof(*us.syms), symbols_cmp, NULL);
2484                 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2485                 free_user_syms(&us);
2486                 if (err)
2487                         goto error;
2488         }
2489
2490         ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2491         if (ucookies) {
2492                 cookies = kvmalloc(size, GFP_KERNEL);
2493                 if (!cookies) {
2494                         err = -ENOMEM;
2495                         goto error;
2496                 }
2497                 if (copy_from_user(cookies, ucookies, size)) {
2498                         err = -EFAULT;
2499                         goto error;
2500                 }
2501         }
2502
2503         link = kzalloc(sizeof(*link), GFP_KERNEL);
2504         if (!link) {
2505                 err = -ENOMEM;
2506                 goto error;
2507         }
2508
2509         bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2510                       &bpf_kprobe_multi_link_lops, prog);
2511
2512         err = bpf_link_prime(&link->link, &link_primer);
2513         if (err)
2514                 goto error;
2515
2516         if (flags & BPF_F_KPROBE_MULTI_RETURN)
2517                 link->fp.exit_handler = kprobe_multi_link_handler;
2518         else
2519                 link->fp.entry_handler = kprobe_multi_link_handler;
2520
2521         link->addrs = addrs;
2522         link->cookies = cookies;
2523         link->cnt = cnt;
2524
2525         if (cookies) {
2526                 /*
2527                  * Sorting addresses will trigger sorting cookies as well
2528                  * (check bpf_kprobe_multi_cookie_swap). This way we can
2529                  * find cookie based on the address in bpf_get_attach_cookie
2530                  * helper.
2531                  */
2532                 sort_r(addrs, cnt, sizeof(*addrs),
2533                        bpf_kprobe_multi_cookie_cmp,
2534                        bpf_kprobe_multi_cookie_swap,
2535                        link);
2536         }
2537
2538         err = register_fprobe_ips(&link->fp, addrs, cnt);
2539         if (err) {
2540                 bpf_link_cleanup(&link_primer);
2541                 return err;
2542         }
2543
2544         return bpf_link_settle(&link_primer);
2545
2546 error:
2547         kfree(link);
2548         kvfree(addrs);
2549         kvfree(cookies);
2550         return err;
2551 }
2552 #else /* !CONFIG_FPROBE */
2553 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2554 {
2555         return -EOPNOTSUPP;
2556 }
2557 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2558 {
2559         return 0;
2560 }
2561 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2562 {
2563         return 0;
2564 }
2565 #endif