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 #include <linux/key.h>
  24 #include <linux/verification.h>
  25
  26 #include <net/bpf_sk_storage.h>
  27
  28 #include <uapi/linux/bpf.h>
  29 #include <uapi/linux/btf.h>
  30
  31 #include <asm/tlb.h>
  32
  33 #include "trace_probe.h"
  34 #include "trace.h"
  35
  36 #define CREATE_TRACE_POINTS
  37 #include "bpf_trace.h"
  38
  39 #define bpf_event_rcu_dereference(p)                                    \
  40         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  41
  42 #ifdef CONFIG_MODULES
  43 struct bpf_trace_module {
  44         struct module *module;
  45         struct list_head list;
  46 };
  47
  48 static LIST_HEAD(bpf_trace_modules);
  49 static DEFINE_MUTEX(bpf_module_mutex);
  50
  51 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  52 {
  53         struct bpf_raw_event_map *btp, *ret = NULL;
  54         struct bpf_trace_module *btm;
  55         unsigned int i;
  56
  57         mutex_lock(&bpf_module_mutex);
  58         list_for_each_entry(btm, &bpf_trace_modules, list) {
  59                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  60                         btp = &btm->module->bpf_raw_events[i];
  61                         if (!strcmp(btp->tp->name, name)) {
  62                                 if (try_module_get(btm->module))
  63                                         ret = btp;
  64                                 goto out;
  65                         }
  66                 }
  67         }
  68 out:
  69         mutex_unlock(&bpf_module_mutex);
  70         return ret;
  71 }
  72 #else
  73 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  74 {
  75         return NULL;
  76 }
  77 #endif /* CONFIG_MODULES */
  78
  79 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  80 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  81
  82 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
  83                                   u64 flags, const struct btf **btf,
  84                                   s32 *btf_id);
  85 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
  86 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
  87
  88 /**
  89  * trace_call_bpf - invoke BPF program
  90  * @call: tracepoint event
  91  * @ctx: opaque context pointer
  92  *
  93  * kprobe handlers execute BPF programs via this helper.
  94  * Can be used from static tracepoints in the future.
  95  *
  96  * Return: BPF programs always return an integer which is interpreted by
  97  * kprobe handler as:
  98  * 0 - return from kprobe (event is filtered out)
  99  * 1 - store kprobe event into ring buffer
 100  * Other values are reserved and currently alias to 1
 101  */
 102 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 103 {
 104         unsigned int ret;
 105
 106         cant_sleep();
 107
 108         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 109                 /*
 110                  * since some bpf program is already running on this cpu,
 111                  * don't call into another bpf program (same or different)
 112                  * and don't send kprobe event into ring-buffer,
 113                  * so return zero here
 114                  */
 115                 ret = 0;
 116                 goto out;
 117         }
 118
 119         /*
 120          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 121          * to all call sites, we did a bpf_prog_array_valid() there to check
 122          * whether call->prog_array is empty or not, which is
 123          * a heuristic to speed up execution.
 124          *
 125          * If bpf_prog_array_valid() fetched prog_array was
 126          * non-NULL, we go into trace_call_bpf() and do the actual
 127          * proper rcu_dereference() under RCU lock.
 128          * If it turns out that prog_array is NULL then, we bail out.
 129          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 130          * was NULL, you'll skip the prog_array with the risk of missing
 131          * out of events when it was updated in between this and the
 132          * rcu_dereference() which is accepted risk.
 133          */
 134         rcu_read_lock();
 135         ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
 136                                  ctx, bpf_prog_run);
 137         rcu_read_unlock();
 138
 139  out:
 140         __this_cpu_dec(bpf_prog_active);
 141
 142         return ret;
 143 }
 144
 145 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 146 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 147 {
 148         regs_set_return_value(regs, rc);
 149         override_function_with_return(regs);
 150         return 0;
 151 }
 152
 153 static const struct bpf_func_proto bpf_override_return_proto = {
 154         .func           = bpf_override_return,
 155         .gpl_only       = true,
 156         .ret_type       = RET_INTEGER,
 157         .arg1_type      = ARG_PTR_TO_CTX,
 158         .arg2_type      = ARG_ANYTHING,
 159 };
 160 #endif
 161
 162 static __always_inline int
 163 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 164 {
 165         int ret;
 166
 167         ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 168         if (unlikely(ret < 0))
 169                 memset(dst, 0, size);
 170         return ret;
 171 }
 172
 173 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 174            const void __user *, unsafe_ptr)
 175 {
 176         return bpf_probe_read_user_common(dst, size, unsafe_ptr);
 177 }
 178
 179 const struct bpf_func_proto bpf_probe_read_user_proto = {
 180         .func           = bpf_probe_read_user,
 181         .gpl_only       = true,
 182         .ret_type       = RET_INTEGER,
 183         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 184         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 185         .arg3_type      = ARG_ANYTHING,
 186 };
 187
 188 static __always_inline int
 189 bpf_probe_read_user_str_common(void *dst, u32 size,
 190                                const void __user *unsafe_ptr)
 191 {
 192         int ret;
 193
 194         /*
 195          * NB: We rely on strncpy_from_user() not copying junk past the NUL
 196          * terminator into `dst`.
 197          *
 198          * strncpy_from_user() does long-sized strides in the fast path. If the
 199          * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
 200          * then there could be junk after the NUL in `dst`. If user takes `dst`
 201          * and keys a hash map with it, then semantically identical strings can
 202          * occupy multiple entries in the map.
 203          */
 204         ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 205         if (unlikely(ret < 0))
 206                 memset(dst, 0, size);
 207         return ret;
 208 }
 209
 210 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 211            const void __user *, unsafe_ptr)
 212 {
 213         return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
 214 }
 215
 216 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 217         .func           = bpf_probe_read_user_str,
 218         .gpl_only       = true,
 219         .ret_type       = RET_INTEGER,
 220         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 221         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 222         .arg3_type      = ARG_ANYTHING,
 223 };
 224
 225 static __always_inline int
 226 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
 227 {
 228         int ret;
 229
 230         ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
 231         if (unlikely(ret < 0))
 232                 memset(dst, 0, size);
 233         return ret;
 234 }
 235
 236 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 237            const void *, unsafe_ptr)
 238 {
 239         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 240 }
 241
 242 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 243         .func           = bpf_probe_read_kernel,
 244         .gpl_only       = true,
 245         .ret_type       = RET_INTEGER,
 246         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 247         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 248         .arg3_type      = ARG_ANYTHING,
 249 };
 250
 251 static __always_inline int
 252 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
 253 {
 254         int ret;
 255
 256         /*
 257          * The strncpy_from_kernel_nofault() call will likely not fill the
 258          * entire buffer, but that's okay in this circumstance as we're probing
 259          * arbitrary memory anyway similar to bpf_probe_read_*() and might
 260          * as well probe the stack. Thus, memory is explicitly cleared
 261          * only in error case, so that improper users ignoring return
 262          * code altogether don't copy garbage; otherwise length of string
 263          * is returned that can be used for bpf_perf_event_output() et al.
 264          */
 265         ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
 266         if (unlikely(ret < 0))
 267                 memset(dst, 0, size);
 268         return ret;
 269 }
 270
 271 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 272            const void *, unsafe_ptr)
 273 {
 274         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 275 }
 276
 277 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 278         .func           = bpf_probe_read_kernel_str,
 279         .gpl_only       = true,
 280         .ret_type       = RET_INTEGER,
 281         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 282         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 283         .arg3_type      = ARG_ANYTHING,
 284 };
 285
 286 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 287 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 288            const void *, unsafe_ptr)
 289 {
 290         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 291                 return bpf_probe_read_user_common(dst, size,
 292                                 (__force void __user *)unsafe_ptr);
 293         }
 294         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 295 }
 296
 297 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 298         .func           = bpf_probe_read_compat,
 299         .gpl_only       = true,
 300         .ret_type       = RET_INTEGER,
 301         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 302         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 303         .arg3_type      = ARG_ANYTHING,
 304 };
 305
 306 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 307            const void *, unsafe_ptr)
 308 {
 309         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 310                 return bpf_probe_read_user_str_common(dst, size,
 311                                 (__force void __user *)unsafe_ptr);
 312         }
 313         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 314 }
 315
 316 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 317         .func           = bpf_probe_read_compat_str,
 318         .gpl_only       = true,
 319         .ret_type       = RET_INTEGER,
 320         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 321         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 322         .arg3_type      = ARG_ANYTHING,
 323 };
 324 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 325
 326 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 327            u32, size)
 328 {
 329         /*
 330          * Ensure we're in user context which is safe for the helper to
 331          * run. This helper has no business in a kthread.
 332          *
 333          * access_ok() should prevent writing to non-user memory, but in
 334          * some situations (nommu, temporary switch, etc) access_ok() does
 335          * not provide enough validation, hence the check on KERNEL_DS.
 336          *
 337          * nmi_uaccess_okay() ensures the probe is not run in an interim
 338          * state, when the task or mm are switched. This is specifically
 339          * required to prevent the use of temporary mm.
 340          */
 341
 342         if (unlikely(in_interrupt() ||
 343                      current->flags & (PF_KTHREAD | PF_EXITING)))
 344                 return -EPERM;
 345         if (unlikely(!nmi_uaccess_okay()))
 346                 return -EPERM;
 347
 348         return copy_to_user_nofault(unsafe_ptr, src, size);
 349 }
 350
 351 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 352         .func           = bpf_probe_write_user,
 353         .gpl_only       = true,
 354         .ret_type       = RET_INTEGER,
 355         .arg1_type      = ARG_ANYTHING,
 356         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 357         .arg3_type      = ARG_CONST_SIZE,
 358 };
 359
 360 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 361 {
 362         if (!capable(CAP_SYS_ADMIN))
 363                 return NULL;
 364
 365         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 366                             current->comm, task_pid_nr(current));
 367
 368         return &bpf_probe_write_user_proto;
 369 }
 370
 371 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
 372
 373 #define MAX_TRACE_PRINTK_VARARGS        3
 374 #define BPF_TRACE_PRINTK_SIZE           1024
 375
 376 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 377            u64, arg2, u64, arg3)
 378 {
 379         u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
 380         u32 *bin_args;
 381         static char buf[BPF_TRACE_PRINTK_SIZE];
 382         unsigned long flags;
 383         int ret;
 384
 385         ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
 386                                   MAX_TRACE_PRINTK_VARARGS);
 387         if (ret < 0)
 388                 return ret;
 389
 390         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 391         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 392
 393         trace_bpf_trace_printk(buf);
 394         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 395
 396         bpf_bprintf_cleanup();
 397
 398         return ret;
 399 }
 400
 401 static const struct bpf_func_proto bpf_trace_printk_proto = {
 402         .func           = bpf_trace_printk,
 403         .gpl_only       = true,
 404         .ret_type       = RET_INTEGER,
 405         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 406         .arg2_type      = ARG_CONST_SIZE,
 407 };
 408
 409 static void __set_printk_clr_event(void)
 410 {
 411         /*
 412          * This program might be calling bpf_trace_printk,
 413          * so enable the associated bpf_trace/bpf_trace_printk event.
 414          * Repeat this each time as it is possible a user has
 415          * disabled bpf_trace_printk events.  By loading a program
 416          * calling bpf_trace_printk() however the user has expressed
 417          * the intent to see such events.
 418          */
 419         if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 420                 pr_warn_ratelimited("could not enable bpf_trace_printk events");
 421 }
 422
 423 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 424 {
 425         __set_printk_clr_event();
 426         return &bpf_trace_printk_proto;
 427 }
 428
 429 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
 430            u32, data_len)
 431 {
 432         static char buf[BPF_TRACE_PRINTK_SIZE];
 433         unsigned long flags;
 434         int ret, num_args;
 435         u32 *bin_args;
 436
 437         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 438             (data_len && !data))
 439                 return -EINVAL;
 440         num_args = data_len / 8;
 441
 442         ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 443         if (ret < 0)
 444                 return ret;
 445
 446         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 447         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 448
 449         trace_bpf_trace_printk(buf);
 450         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 451
 452         bpf_bprintf_cleanup();
 453
 454         return ret;
 455 }
 456
 457 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
 458         .func           = bpf_trace_vprintk,
 459         .gpl_only       = true,
 460         .ret_type       = RET_INTEGER,
 461         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 462         .arg2_type      = ARG_CONST_SIZE,
 463         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 464         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
 465 };
 466
 467 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
 468 {
 469         __set_printk_clr_event();
 470         return &bpf_trace_vprintk_proto;
 471 }
 472
 473 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 474            const void *, data, u32, data_len)
 475 {
 476         int err, num_args;
 477         u32 *bin_args;
 478
 479         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 480             (data_len && !data))
 481                 return -EINVAL;
 482         num_args = data_len / 8;
 483
 484         err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 485         if (err < 0)
 486                 return err;
 487
 488         seq_bprintf(m, fmt, bin_args);
 489
 490         bpf_bprintf_cleanup();
 491
 492         return seq_has_overflowed(m) ? -EOVERFLOW : 0;
 493 }
 494
 495 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
 496
 497 static const struct bpf_func_proto bpf_seq_printf_proto = {
 498         .func           = bpf_seq_printf,
 499         .gpl_only       = true,
 500         .ret_type       = RET_INTEGER,
 501         .arg1_type      = ARG_PTR_TO_BTF_ID,
 502         .arg1_btf_id    = &btf_seq_file_ids[0],
 503         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 504         .arg3_type      = ARG_CONST_SIZE,
 505         .arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 506         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 507 };
 508
 509 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
 510 {
 511         return seq_write(m, data, len) ? -EOVERFLOW : 0;
 512 }
 513
 514 static const struct bpf_func_proto bpf_seq_write_proto = {
 515         .func           = bpf_seq_write,
 516         .gpl_only       = true,
 517         .ret_type       = RET_INTEGER,
 518         .arg1_type      = ARG_PTR_TO_BTF_ID,
 519         .arg1_btf_id    = &btf_seq_file_ids[0],
 520         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 521         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 522 };
 523
 524 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
 525            u32, btf_ptr_size, u64, flags)
 526 {
 527         const struct btf *btf;
 528         s32 btf_id;
 529         int ret;
 530
 531         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 532         if (ret)
 533                 return ret;
 534
 535         return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
 536 }
 537
 538 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
 539         .func           = bpf_seq_printf_btf,
 540         .gpl_only       = true,
 541         .ret_type       = RET_INTEGER,
 542         .arg1_type      = ARG_PTR_TO_BTF_ID,
 543         .arg1_btf_id    = &btf_seq_file_ids[0],
 544         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 545         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 546         .arg4_type      = ARG_ANYTHING,
 547 };
 548
 549 static __always_inline int
 550 get_map_perf_counter(struct bpf_map *map, u64 flags,
 551                      u64 *value, u64 *enabled, u64 *running)
 552 {
 553         struct bpf_array *array = container_of(map, struct bpf_array, map);
 554         unsigned int cpu = smp_processor_id();
 555         u64 index = flags & BPF_F_INDEX_MASK;
 556         struct bpf_event_entry *ee;
 557
 558         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 559                 return -EINVAL;
 560         if (index == BPF_F_CURRENT_CPU)
 561                 index = cpu;
 562         if (unlikely(index >= array->map.max_entries))
 563                 return -E2BIG;
 564
 565         ee = READ_ONCE(array->ptrs[index]);
 566         if (!ee)
 567                 return -ENOENT;
 568
 569         return perf_event_read_local(ee->event, value, enabled, running);
 570 }
 571
 572 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 573 {
 574         u64 value = 0;
 575         int err;
 576
 577         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 578         /*
 579          * this api is ugly since we miss [-22..-2] range of valid
 580          * counter values, but that's uapi
 581          */
 582         if (err)
 583                 return err;
 584         return value;
 585 }
 586
 587 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 588         .func           = bpf_perf_event_read,
 589         .gpl_only       = true,
 590         .ret_type       = RET_INTEGER,
 591         .arg1_type      = ARG_CONST_MAP_PTR,
 592         .arg2_type      = ARG_ANYTHING,
 593 };
 594
 595 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 596            struct bpf_perf_event_value *, buf, u32, size)
 597 {
 598         int err = -EINVAL;
 599
 600         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 601                 goto clear;
 602         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 603                                    &buf->running);
 604         if (unlikely(err))
 605                 goto clear;
 606         return 0;
 607 clear:
 608         memset(buf, 0, size);
 609         return err;
 610 }
 611
 612 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 613         .func           = bpf_perf_event_read_value,
 614         .gpl_only       = true,
 615         .ret_type       = RET_INTEGER,
 616         .arg1_type      = ARG_CONST_MAP_PTR,
 617         .arg2_type      = ARG_ANYTHING,
 618         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 619         .arg4_type      = ARG_CONST_SIZE,
 620 };
 621
 622 static __always_inline u64
 623 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 624                         u64 flags, struct perf_sample_data *sd)
 625 {
 626         struct bpf_array *array = container_of(map, struct bpf_array, map);
 627         unsigned int cpu = smp_processor_id();
 628         u64 index = flags & BPF_F_INDEX_MASK;
 629         struct bpf_event_entry *ee;
 630         struct perf_event *event;
 631
 632         if (index == BPF_F_CURRENT_CPU)
 633                 index = cpu;
 634         if (unlikely(index >= array->map.max_entries))
 635                 return -E2BIG;
 636
 637         ee = READ_ONCE(array->ptrs[index]);
 638         if (!ee)
 639                 return -ENOENT;
 640
 641         event = ee->event;
 642         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 643                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 644                 return -EINVAL;
 645
 646         if (unlikely(event->oncpu != cpu))
 647                 return -EOPNOTSUPP;
 648
 649         return perf_event_output(event, sd, regs);
 650 }
 651
 652 /*
 653  * Support executing tracepoints in normal, irq, and nmi context that each call
 654  * bpf_perf_event_output
 655  */
 656 struct bpf_trace_sample_data {
 657         struct perf_sample_data sds[3];
 658 };
 659
 660 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 661 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 662 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 663            u64, flags, void *, data, u64, size)
 664 {
 665         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 666         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 667         struct perf_raw_record raw = {
 668                 .frag = {
 669                         .size = size,
 670                         .data = data,
 671                 },
 672         };
 673         struct perf_sample_data *sd;
 674         int err;
 675
 676         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 677                 err = -EBUSY;
 678                 goto out;
 679         }
 680
 681         sd = &sds->sds[nest_level - 1];
 682
 683         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 684                 err = -EINVAL;
 685                 goto out;
 686         }
 687
 688         perf_sample_data_init(sd, 0, 0);
 689         sd->raw = &raw;
 690         sd->sample_flags |= PERF_SAMPLE_RAW;
 691
 692         err = __bpf_perf_event_output(regs, map, flags, sd);
 693
 694 out:
 695         this_cpu_dec(bpf_trace_nest_level);
 696         return err;
 697 }
 698
 699 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 700         .func           = bpf_perf_event_output,
 701         .gpl_only       = true,
 702         .ret_type       = RET_INTEGER,
 703         .arg1_type      = ARG_PTR_TO_CTX,
 704         .arg2_type      = ARG_CONST_MAP_PTR,
 705         .arg3_type      = ARG_ANYTHING,
 706         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 707         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 708 };
 709
 710 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 711 struct bpf_nested_pt_regs {
 712         struct pt_regs regs[3];
 713 };
 714 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 715 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 716
 717 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 718                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 719 {
 720         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 721         struct perf_raw_frag frag = {
 722                 .copy           = ctx_copy,
 723                 .size           = ctx_size,
 724                 .data           = ctx,
 725         };
 726         struct perf_raw_record raw = {
 727                 .frag = {
 728                         {
 729                                 .next   = ctx_size ? &frag : NULL,
 730                         },
 731                         .size   = meta_size,
 732                         .data   = meta,
 733                 },
 734         };
 735         struct perf_sample_data *sd;
 736         struct pt_regs *regs;
 737         u64 ret;
 738
 739         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 740                 ret = -EBUSY;
 741                 goto out;
 742         }
 743         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 744         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 745
 746         perf_fetch_caller_regs(regs);
 747         perf_sample_data_init(sd, 0, 0);
 748         sd->raw = &raw;
 749         sd->sample_flags |= PERF_SAMPLE_RAW;
 750
 751         ret = __bpf_perf_event_output(regs, map, flags, sd);
 752 out:
 753         this_cpu_dec(bpf_event_output_nest_level);
 754         return ret;
 755 }
 756
 757 BPF_CALL_0(bpf_get_current_task)
 758 {
 759         return (long) current;
 760 }
 761
 762 const struct bpf_func_proto bpf_get_current_task_proto = {
 763         .func           = bpf_get_current_task,
 764         .gpl_only       = true,
 765         .ret_type       = RET_INTEGER,
 766 };
 767
 768 BPF_CALL_0(bpf_get_current_task_btf)
 769 {
 770         return (unsigned long) current;
 771 }
 772
 773 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
 774         .func           = bpf_get_current_task_btf,
 775         .gpl_only       = true,
 776         .ret_type       = RET_PTR_TO_BTF_ID,
 777         .ret_btf_id     = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 778 };
 779
 780 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
 781 {
 782         return (unsigned long) task_pt_regs(task);
 783 }
 784
 785 BTF_ID_LIST(bpf_task_pt_regs_ids)
 786 BTF_ID(struct, pt_regs)
 787
 788 const struct bpf_func_proto bpf_task_pt_regs_proto = {
 789         .func           = bpf_task_pt_regs,
 790         .gpl_only       = true,
 791         .arg1_type      = ARG_PTR_TO_BTF_ID,
 792         .arg1_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 793         .ret_type       = RET_PTR_TO_BTF_ID,
 794         .ret_btf_id     = &bpf_task_pt_regs_ids[0],
 795 };
 796
 797 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 798 {
 799         struct bpf_array *array = container_of(map, struct bpf_array, map);
 800         struct cgroup *cgrp;
 801
 802         if (unlikely(idx >= array->map.max_entries))
 803                 return -E2BIG;
 804
 805         cgrp = READ_ONCE(array->ptrs[idx]);
 806         if (unlikely(!cgrp))
 807                 return -EAGAIN;
 808
 809         return task_under_cgroup_hierarchy(current, cgrp);
 810 }
 811
 812 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 813         .func           = bpf_current_task_under_cgroup,
 814         .gpl_only       = false,
 815         .ret_type       = RET_INTEGER,
 816         .arg1_type      = ARG_CONST_MAP_PTR,
 817         .arg2_type      = ARG_ANYTHING,
 818 };
 819
 820 struct send_signal_irq_work {
 821         struct irq_work irq_work;
 822         struct task_struct *task;
 823         u32 sig;
 824         enum pid_type type;
 825 };
 826
 827 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 828
 829 static void do_bpf_send_signal(struct irq_work *entry)
 830 {
 831         struct send_signal_irq_work *work;
 832
 833         work = container_of(entry, struct send_signal_irq_work, irq_work);
 834         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 835 }
 836
 837 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 838 {
 839         struct send_signal_irq_work *work = NULL;
 840
 841         /* Similar to bpf_probe_write_user, task needs to be
 842          * in a sound condition and kernel memory access be
 843          * permitted in order to send signal to the current
 844          * task.
 845          */
 846         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 847                 return -EPERM;
 848         if (unlikely(!nmi_uaccess_okay()))
 849                 return -EPERM;
 850
 851         if (irqs_disabled()) {
 852                 /* Do an early check on signal validity. Otherwise,
 853                  * the error is lost in deferred irq_work.
 854                  */
 855                 if (unlikely(!valid_signal(sig)))
 856                         return -EINVAL;
 857
 858                 work = this_cpu_ptr(&send_signal_work);
 859                 if (irq_work_is_busy(&work->irq_work))
 860                         return -EBUSY;
 861
 862                 /* Add the current task, which is the target of sending signal,
 863                  * to the irq_work. The current task may change when queued
 864                  * irq works get executed.
 865                  */
 866                 work->task = current;
 867                 work->sig = sig;
 868                 work->type = type;
 869                 irq_work_queue(&work->irq_work);
 870                 return 0;
 871         }
 872
 873         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 874 }
 875
 876 BPF_CALL_1(bpf_send_signal, u32, sig)
 877 {
 878         return bpf_send_signal_common(sig, PIDTYPE_TGID);
 879 }
 880
 881 static const struct bpf_func_proto bpf_send_signal_proto = {
 882         .func           = bpf_send_signal,
 883         .gpl_only       = false,
 884         .ret_type       = RET_INTEGER,
 885         .arg1_type      = ARG_ANYTHING,
 886 };
 887
 888 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 889 {
 890         return bpf_send_signal_common(sig, PIDTYPE_PID);
 891 }
 892
 893 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 894         .func           = bpf_send_signal_thread,
 895         .gpl_only       = false,
 896         .ret_type       = RET_INTEGER,
 897         .arg1_type      = ARG_ANYTHING,
 898 };
 899
 900 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 901 {
 902         long len;
 903         char *p;
 904
 905         if (!sz)
 906                 return 0;
 907
 908         p = d_path(path, buf, sz);
 909         if (IS_ERR(p)) {
 910                 len = PTR_ERR(p);
 911         } else {
 912                 len = buf + sz - p;
 913                 memmove(buf, p, len);
 914         }
 915
 916         return len;
 917 }
 918
 919 BTF_SET_START(btf_allowlist_d_path)
 920 #ifdef CONFIG_SECURITY
 921 BTF_ID(func, security_file_permission)
 922 BTF_ID(func, security_inode_getattr)
 923 BTF_ID(func, security_file_open)
 924 #endif
 925 #ifdef CONFIG_SECURITY_PATH
 926 BTF_ID(func, security_path_truncate)
 927 #endif
 928 BTF_ID(func, vfs_truncate)
 929 BTF_ID(func, vfs_fallocate)
 930 BTF_ID(func, dentry_open)
 931 BTF_ID(func, vfs_getattr)
 932 BTF_ID(func, filp_close)
 933 BTF_SET_END(btf_allowlist_d_path)
 934
 935 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 936 {
 937         if (prog->type == BPF_PROG_TYPE_TRACING &&
 938             prog->expected_attach_type == BPF_TRACE_ITER)
 939                 return true;
 940
 941         if (prog->type == BPF_PROG_TYPE_LSM)
 942                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 943
 944         return btf_id_set_contains(&btf_allowlist_d_path,
 945                                    prog->aux->attach_btf_id);
 946 }
 947
 948 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 949
 950 static const struct bpf_func_proto bpf_d_path_proto = {
 951         .func           = bpf_d_path,
 952         .gpl_only       = false,
 953         .ret_type       = RET_INTEGER,
 954         .arg1_type      = ARG_PTR_TO_BTF_ID,
 955         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
 956         .arg2_type      = ARG_PTR_TO_MEM,
 957         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 958         .allowed        = bpf_d_path_allowed,
 959 };
 960
 961 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
 962                          BTF_F_PTR_RAW | BTF_F_ZERO)
 963
 964 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 965                                   u64 flags, const struct btf **btf,
 966                                   s32 *btf_id)
 967 {
 968         const struct btf_type *t;
 969
 970         if (unlikely(flags & ~(BTF_F_ALL)))
 971                 return -EINVAL;
 972
 973         if (btf_ptr_size != sizeof(struct btf_ptr))
 974                 return -EINVAL;
 975
 976         *btf = bpf_get_btf_vmlinux();
 977
 978         if (IS_ERR_OR_NULL(*btf))
 979                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 980
 981         if (ptr->type_id > 0)
 982                 *btf_id = ptr->type_id;
 983         else
 984                 return -EINVAL;
 985
 986         if (*btf_id > 0)
 987                 t = btf_type_by_id(*btf, *btf_id);
 988         if (*btf_id <= 0 || !t)
 989                 return -ENOENT;
 990
 991         return 0;
 992 }
 993
 994 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
 995            u32, btf_ptr_size, u64, flags)
 996 {
 997         const struct btf *btf;
 998         s32 btf_id;
 999         int ret;
1000
1001         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1002         if (ret)
1003                 return ret;
1004
1005         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1006                                       flags);
1007 }
1008
1009 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1010         .func           = bpf_snprintf_btf,
1011         .gpl_only       = false,
1012         .ret_type       = RET_INTEGER,
1013         .arg1_type      = ARG_PTR_TO_MEM,
1014         .arg2_type      = ARG_CONST_SIZE,
1015         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1016         .arg4_type      = ARG_CONST_SIZE,
1017         .arg5_type      = ARG_ANYTHING,
1018 };
1019
1020 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1021 {
1022         /* This helper call is inlined by verifier. */
1023         return ((u64 *)ctx)[-2];
1024 }
1025
1026 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1027         .func           = bpf_get_func_ip_tracing,
1028         .gpl_only       = true,
1029         .ret_type       = RET_INTEGER,
1030         .arg1_type      = ARG_PTR_TO_CTX,
1031 };
1032
1033 #ifdef CONFIG_X86_KERNEL_IBT
1034 static unsigned long get_entry_ip(unsigned long fentry_ip)
1035 {
1036         u32 instr;
1037
1038         /* Being extra safe in here in case entry ip is on the page-edge. */
1039         if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1))
1040                 return fentry_ip;
1041         if (is_endbr(instr))
1042                 fentry_ip -= ENDBR_INSN_SIZE;
1043         return fentry_ip;
1044 }
1045 #else
1046 #define get_entry_ip(fentry_ip) fentry_ip
1047 #endif
1048
1049 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1050 {
1051         struct kprobe *kp = kprobe_running();
1052
1053         if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1054                 return 0;
1055
1056         return get_entry_ip((uintptr_t)kp->addr);
1057 }
1058
1059 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1060         .func           = bpf_get_func_ip_kprobe,
1061         .gpl_only       = true,
1062         .ret_type       = RET_INTEGER,
1063         .arg1_type      = ARG_PTR_TO_CTX,
1064 };
1065
1066 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1067 {
1068         return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1069 }
1070
1071 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1072         .func           = bpf_get_func_ip_kprobe_multi,
1073         .gpl_only       = false,
1074         .ret_type       = RET_INTEGER,
1075         .arg1_type      = ARG_PTR_TO_CTX,
1076 };
1077
1078 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1079 {
1080         return bpf_kprobe_multi_cookie(current->bpf_ctx);
1081 }
1082
1083 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1084         .func           = bpf_get_attach_cookie_kprobe_multi,
1085         .gpl_only       = false,
1086         .ret_type       = RET_INTEGER,
1087         .arg1_type      = ARG_PTR_TO_CTX,
1088 };
1089
1090 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1091 {
1092         struct bpf_trace_run_ctx *run_ctx;
1093
1094         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1095         return run_ctx->bpf_cookie;
1096 }
1097
1098 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1099         .func           = bpf_get_attach_cookie_trace,
1100         .gpl_only       = false,
1101         .ret_type       = RET_INTEGER,
1102         .arg1_type      = ARG_PTR_TO_CTX,
1103 };
1104
1105 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1106 {
1107         return ctx->event->bpf_cookie;
1108 }
1109
1110 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1111         .func           = bpf_get_attach_cookie_pe,
1112         .gpl_only       = false,
1113         .ret_type       = RET_INTEGER,
1114         .arg1_type      = ARG_PTR_TO_CTX,
1115 };
1116
1117 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1118 {
1119         struct bpf_trace_run_ctx *run_ctx;
1120
1121         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1122         return run_ctx->bpf_cookie;
1123 }
1124
1125 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1126         .func           = bpf_get_attach_cookie_tracing,
1127         .gpl_only       = false,
1128         .ret_type       = RET_INTEGER,
1129         .arg1_type      = ARG_PTR_TO_CTX,
1130 };
1131
1132 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1133 {
1134 #ifndef CONFIG_X86
1135         return -ENOENT;
1136 #else
1137         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1138         u32 entry_cnt = size / br_entry_size;
1139
1140         entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1141
1142         if (unlikely(flags))
1143                 return -EINVAL;
1144
1145         if (!entry_cnt)
1146                 return -ENOENT;
1147
1148         return entry_cnt * br_entry_size;
1149 #endif
1150 }
1151
1152 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1153         .func           = bpf_get_branch_snapshot,
1154         .gpl_only       = true,
1155         .ret_type       = RET_INTEGER,
1156         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1157         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1158 };
1159
1160 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1161 {
1162         /* This helper call is inlined by verifier. */
1163         u64 nr_args = ((u64 *)ctx)[-1];
1164
1165         if ((u64) n >= nr_args)
1166                 return -EINVAL;
1167         *value = ((u64 *)ctx)[n];
1168         return 0;
1169 }
1170
1171 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1172         .func           = get_func_arg,
1173         .ret_type       = RET_INTEGER,
1174         .arg1_type      = ARG_PTR_TO_CTX,
1175         .arg2_type      = ARG_ANYTHING,
1176         .arg3_type      = ARG_PTR_TO_LONG,
1177 };
1178
1179 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1180 {
1181         /* This helper call is inlined by verifier. */
1182         u64 nr_args = ((u64 *)ctx)[-1];
1183
1184         *value = ((u64 *)ctx)[nr_args];
1185         return 0;
1186 }
1187
1188 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1189         .func           = get_func_ret,
1190         .ret_type       = RET_INTEGER,
1191         .arg1_type      = ARG_PTR_TO_CTX,
1192         .arg2_type      = ARG_PTR_TO_LONG,
1193 };
1194
1195 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1196 {
1197         /* This helper call is inlined by verifier. */
1198         return ((u64 *)ctx)[-1];
1199 }
1200
1201 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1202         .func           = get_func_arg_cnt,
1203         .ret_type       = RET_INTEGER,
1204         .arg1_type      = ARG_PTR_TO_CTX,
1205 };
1206
1207 #ifdef CONFIG_KEYS
1208 __diag_push();
1209 __diag_ignore_all("-Wmissing-prototypes",
1210                   "kfuncs which will be used in BPF programs");
1211
1212 /**
1213  * bpf_lookup_user_key - lookup a key by its serial
1214  * @serial: key handle serial number
1215  * @flags: lookup-specific flags
1216  *
1217  * Search a key with a given *serial* and the provided *flags*.
1218  * If found, increment the reference count of the key by one, and
1219  * return it in the bpf_key structure.
1220  *
1221  * The bpf_key structure must be passed to bpf_key_put() when done
1222  * with it, so that the key reference count is decremented and the
1223  * bpf_key structure is freed.
1224  *
1225  * Permission checks are deferred to the time the key is used by
1226  * one of the available key-specific kfuncs.
1227  *
1228  * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1229  * special keyring (e.g. session keyring), if it doesn't yet exist.
1230  * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1231  * for the key construction, and to retrieve uninstantiated keys (keys
1232  * without data attached to them).
1233  *
1234  * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1235  *         NULL pointer otherwise.
1236  */
1237 struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1238 {
1239         key_ref_t key_ref;
1240         struct bpf_key *bkey;
1241
1242         if (flags & ~KEY_LOOKUP_ALL)
1243                 return NULL;
1244
1245         /*
1246          * Permission check is deferred until the key is used, as the
1247          * intent of the caller is unknown here.
1248          */
1249         key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1250         if (IS_ERR(key_ref))
1251                 return NULL;
1252
1253         bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1254         if (!bkey) {
1255                 key_put(key_ref_to_ptr(key_ref));
1256                 return NULL;
1257         }
1258
1259         bkey->key = key_ref_to_ptr(key_ref);
1260         bkey->has_ref = true;
1261
1262         return bkey;
1263 }
1264
1265 /**
1266  * bpf_lookup_system_key - lookup a key by a system-defined ID
1267  * @id: key ID
1268  *
1269  * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1270  * The key pointer is marked as invalid, to prevent bpf_key_put() from
1271  * attempting to decrement the key reference count on that pointer. The key
1272  * pointer set in such way is currently understood only by
1273  * verify_pkcs7_signature().
1274  *
1275  * Set *id* to one of the values defined in include/linux/verification.h:
1276  * 0 for the primary keyring (immutable keyring of system keys);
1277  * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1278  * (where keys can be added only if they are vouched for by existing keys
1279  * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1280  * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1281  * kerned image and, possibly, the initramfs signature).
1282  *
1283  * Return: a bpf_key pointer with an invalid key pointer set from the
1284  *         pre-determined ID on success, a NULL pointer otherwise
1285  */
1286 struct bpf_key *bpf_lookup_system_key(u64 id)
1287 {
1288         struct bpf_key *bkey;
1289
1290         if (system_keyring_id_check(id) < 0)
1291                 return NULL;
1292
1293         bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1294         if (!bkey)
1295                 return NULL;
1296
1297         bkey->key = (struct key *)(unsigned long)id;
1298         bkey->has_ref = false;
1299
1300         return bkey;
1301 }
1302
1303 /**
1304  * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1305  * @bkey: bpf_key structure
1306  *
1307  * Decrement the reference count of the key inside *bkey*, if the pointer
1308  * is valid, and free *bkey*.
1309  */
1310 void bpf_key_put(struct bpf_key *bkey)
1311 {
1312         if (bkey->has_ref)
1313                 key_put(bkey->key);
1314
1315         kfree(bkey);
1316 }
1317
1318 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1319 /**
1320  * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1321  * @data_ptr: data to verify
1322  * @sig_ptr: signature of the data
1323  * @trusted_keyring: keyring with keys trusted for signature verification
1324  *
1325  * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1326  * with keys in a keyring referenced by *trusted_keyring*.
1327  *
1328  * Return: 0 on success, a negative value on error.
1329  */
1330 int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr,
1331                                struct bpf_dynptr_kern *sig_ptr,
1332                                struct bpf_key *trusted_keyring)
1333 {
1334         int ret;
1335
1336         if (trusted_keyring->has_ref) {
1337                 /*
1338                  * Do the permission check deferred in bpf_lookup_user_key().
1339                  * See bpf_lookup_user_key() for more details.
1340                  *
1341                  * A call to key_task_permission() here would be redundant, as
1342                  * it is already done by keyring_search() called by
1343                  * find_asymmetric_key().
1344                  */
1345                 ret = key_validate(trusted_keyring->key);
1346                 if (ret < 0)
1347                         return ret;
1348         }
1349
1350         return verify_pkcs7_signature(data_ptr->data,
1351                                       bpf_dynptr_get_size(data_ptr),
1352                                       sig_ptr->data,
1353                                       bpf_dynptr_get_size(sig_ptr),
1354                                       trusted_keyring->key,
1355                                       VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1356                                       NULL);
1357 }
1358 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1359
1360 __diag_pop();
1361
1362 BTF_SET8_START(key_sig_kfunc_set)
1363 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1364 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1365 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1366 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1367 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1368 #endif
1369 BTF_SET8_END(key_sig_kfunc_set)
1370
1371 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1372         .owner = THIS_MODULE,
1373         .set = &key_sig_kfunc_set,
1374 };
1375
1376 static int __init bpf_key_sig_kfuncs_init(void)
1377 {
1378         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1379                                          &bpf_key_sig_kfunc_set);
1380 }
1381
1382 late_initcall(bpf_key_sig_kfuncs_init);
1383 #endif /* CONFIG_KEYS */
1384
1385 static const struct bpf_func_proto *
1386 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1387 {
1388         switch (func_id) {
1389         case BPF_FUNC_map_lookup_elem:
1390                 return &bpf_map_lookup_elem_proto;
1391         case BPF_FUNC_map_update_elem:
1392                 return &bpf_map_update_elem_proto;
1393         case BPF_FUNC_map_delete_elem:
1394                 return &bpf_map_delete_elem_proto;
1395         case BPF_FUNC_map_push_elem:
1396                 return &bpf_map_push_elem_proto;
1397         case BPF_FUNC_map_pop_elem:
1398                 return &bpf_map_pop_elem_proto;
1399         case BPF_FUNC_map_peek_elem:
1400                 return &bpf_map_peek_elem_proto;
1401         case BPF_FUNC_map_lookup_percpu_elem:
1402                 return &bpf_map_lookup_percpu_elem_proto;
1403         case BPF_FUNC_ktime_get_ns:
1404                 return &bpf_ktime_get_ns_proto;
1405         case BPF_FUNC_ktime_get_boot_ns:
1406                 return &bpf_ktime_get_boot_ns_proto;
1407         case BPF_FUNC_tail_call:
1408                 return &bpf_tail_call_proto;
1409         case BPF_FUNC_get_current_pid_tgid:
1410                 return &bpf_get_current_pid_tgid_proto;
1411         case BPF_FUNC_get_current_task:
1412                 return &bpf_get_current_task_proto;
1413         case BPF_FUNC_get_current_task_btf:
1414                 return &bpf_get_current_task_btf_proto;
1415         case BPF_FUNC_task_pt_regs:
1416                 return &bpf_task_pt_regs_proto;
1417         case BPF_FUNC_get_current_uid_gid:
1418                 return &bpf_get_current_uid_gid_proto;
1419         case BPF_FUNC_get_current_comm:
1420                 return &bpf_get_current_comm_proto;
1421         case BPF_FUNC_trace_printk:
1422                 return bpf_get_trace_printk_proto();
1423         case BPF_FUNC_get_smp_processor_id:
1424                 return &bpf_get_smp_processor_id_proto;
1425         case BPF_FUNC_get_numa_node_id:
1426                 return &bpf_get_numa_node_id_proto;
1427         case BPF_FUNC_perf_event_read:
1428                 return &bpf_perf_event_read_proto;
1429         case BPF_FUNC_current_task_under_cgroup:
1430                 return &bpf_current_task_under_cgroup_proto;
1431         case BPF_FUNC_get_prandom_u32:
1432                 return &bpf_get_prandom_u32_proto;
1433         case BPF_FUNC_probe_write_user:
1434                 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1435                        NULL : bpf_get_probe_write_proto();
1436         case BPF_FUNC_probe_read_user:
1437                 return &bpf_probe_read_user_proto;
1438         case BPF_FUNC_probe_read_kernel:
1439                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1440                        NULL : &bpf_probe_read_kernel_proto;
1441         case BPF_FUNC_probe_read_user_str:
1442                 return &bpf_probe_read_user_str_proto;
1443         case BPF_FUNC_probe_read_kernel_str:
1444                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1445                        NULL : &bpf_probe_read_kernel_str_proto;
1446 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1447         case BPF_FUNC_probe_read:
1448                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1449                        NULL : &bpf_probe_read_compat_proto;
1450         case BPF_FUNC_probe_read_str:
1451                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1452                        NULL : &bpf_probe_read_compat_str_proto;
1453 #endif
1454 #ifdef CONFIG_CGROUPS
1455         case BPF_FUNC_get_current_cgroup_id:
1456                 return &bpf_get_current_cgroup_id_proto;
1457         case BPF_FUNC_get_current_ancestor_cgroup_id:
1458                 return &bpf_get_current_ancestor_cgroup_id_proto;
1459 #endif
1460         case BPF_FUNC_send_signal:
1461                 return &bpf_send_signal_proto;
1462         case BPF_FUNC_send_signal_thread:
1463                 return &bpf_send_signal_thread_proto;
1464         case BPF_FUNC_perf_event_read_value:
1465                 return &bpf_perf_event_read_value_proto;
1466         case BPF_FUNC_get_ns_current_pid_tgid:
1467                 return &bpf_get_ns_current_pid_tgid_proto;
1468         case BPF_FUNC_ringbuf_output:
1469                 return &bpf_ringbuf_output_proto;
1470         case BPF_FUNC_ringbuf_reserve:
1471                 return &bpf_ringbuf_reserve_proto;
1472         case BPF_FUNC_ringbuf_submit:
1473                 return &bpf_ringbuf_submit_proto;
1474         case BPF_FUNC_ringbuf_discard:
1475                 return &bpf_ringbuf_discard_proto;
1476         case BPF_FUNC_ringbuf_query:
1477                 return &bpf_ringbuf_query_proto;
1478         case BPF_FUNC_jiffies64:
1479                 return &bpf_jiffies64_proto;
1480         case BPF_FUNC_get_task_stack:
1481                 return &bpf_get_task_stack_proto;
1482         case BPF_FUNC_copy_from_user:
1483                 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1484         case BPF_FUNC_copy_from_user_task:
1485                 return prog->aux->sleepable ? &bpf_copy_from_user_task_proto : NULL;
1486         case BPF_FUNC_snprintf_btf:
1487                 return &bpf_snprintf_btf_proto;
1488         case BPF_FUNC_per_cpu_ptr:
1489                 return &bpf_per_cpu_ptr_proto;
1490         case BPF_FUNC_this_cpu_ptr:
1491                 return &bpf_this_cpu_ptr_proto;
1492         case BPF_FUNC_task_storage_get:
1493                 return &bpf_task_storage_get_proto;
1494         case BPF_FUNC_task_storage_delete:
1495                 return &bpf_task_storage_delete_proto;
1496         case BPF_FUNC_for_each_map_elem:
1497                 return &bpf_for_each_map_elem_proto;
1498         case BPF_FUNC_snprintf:
1499                 return &bpf_snprintf_proto;
1500         case BPF_FUNC_get_func_ip:
1501                 return &bpf_get_func_ip_proto_tracing;
1502         case BPF_FUNC_get_branch_snapshot:
1503                 return &bpf_get_branch_snapshot_proto;
1504         case BPF_FUNC_find_vma:
1505                 return &bpf_find_vma_proto;
1506         case BPF_FUNC_trace_vprintk:
1507                 return bpf_get_trace_vprintk_proto();
1508         default:
1509                 return bpf_base_func_proto(func_id);
1510         }
1511 }
1512
1513 static const struct bpf_func_proto *
1514 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1515 {
1516         switch (func_id) {
1517         case BPF_FUNC_perf_event_output:
1518                 return &bpf_perf_event_output_proto;
1519         case BPF_FUNC_get_stackid:
1520                 return &bpf_get_stackid_proto;
1521         case BPF_FUNC_get_stack:
1522                 return &bpf_get_stack_proto;
1523 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1524         case BPF_FUNC_override_return:
1525                 return &bpf_override_return_proto;
1526 #endif
1527         case BPF_FUNC_get_func_ip:
1528                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1529                         &bpf_get_func_ip_proto_kprobe_multi :
1530                         &bpf_get_func_ip_proto_kprobe;
1531         case BPF_FUNC_get_attach_cookie:
1532                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1533                         &bpf_get_attach_cookie_proto_kmulti :
1534                         &bpf_get_attach_cookie_proto_trace;
1535         default:
1536                 return bpf_tracing_func_proto(func_id, prog);
1537         }
1538 }
1539
1540 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1541 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1542                                         const struct bpf_prog *prog,
1543                                         struct bpf_insn_access_aux *info)
1544 {
1545         if (off < 0 || off >= sizeof(struct pt_regs))
1546                 return false;
1547         if (type != BPF_READ)
1548                 return false;
1549         if (off % size != 0)
1550                 return false;
1551         /*
1552          * Assertion for 32 bit to make sure last 8 byte access
1553          * (BPF_DW) to the last 4 byte member is disallowed.
1554          */
1555         if (off + size > sizeof(struct pt_regs))
1556                 return false;
1557
1558         return true;
1559 }
1560
1561 const struct bpf_verifier_ops kprobe_verifier_ops = {
1562         .get_func_proto  = kprobe_prog_func_proto,
1563         .is_valid_access = kprobe_prog_is_valid_access,
1564 };
1565
1566 const struct bpf_prog_ops kprobe_prog_ops = {
1567 };
1568
1569 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1570            u64, flags, void *, data, u64, size)
1571 {
1572         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1573
1574         /*
1575          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1576          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1577          * from there and call the same bpf_perf_event_output() helper inline.
1578          */
1579         return ____bpf_perf_event_output(regs, map, flags, data, size);
1580 }
1581
1582 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1583         .func           = bpf_perf_event_output_tp,
1584         .gpl_only       = true,
1585         .ret_type       = RET_INTEGER,
1586         .arg1_type      = ARG_PTR_TO_CTX,
1587         .arg2_type      = ARG_CONST_MAP_PTR,
1588         .arg3_type      = ARG_ANYTHING,
1589         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1590         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1591 };
1592
1593 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1594            u64, flags)
1595 {
1596         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1597
1598         /*
1599          * Same comment as in bpf_perf_event_output_tp(), only that this time
1600          * the other helper's function body cannot be inlined due to being
1601          * external, thus we need to call raw helper function.
1602          */
1603         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1604                                flags, 0, 0);
1605 }
1606
1607 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1608         .func           = bpf_get_stackid_tp,
1609         .gpl_only       = true,
1610         .ret_type       = RET_INTEGER,
1611         .arg1_type      = ARG_PTR_TO_CTX,
1612         .arg2_type      = ARG_CONST_MAP_PTR,
1613         .arg3_type      = ARG_ANYTHING,
1614 };
1615
1616 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1617            u64, flags)
1618 {
1619         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1620
1621         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1622                              (unsigned long) size, flags, 0);
1623 }
1624
1625 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1626         .func           = bpf_get_stack_tp,
1627         .gpl_only       = true,
1628         .ret_type       = RET_INTEGER,
1629         .arg1_type      = ARG_PTR_TO_CTX,
1630         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1631         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1632         .arg4_type      = ARG_ANYTHING,
1633 };
1634
1635 static const struct bpf_func_proto *
1636 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1637 {
1638         switch (func_id) {
1639         case BPF_FUNC_perf_event_output:
1640                 return &bpf_perf_event_output_proto_tp;
1641         case BPF_FUNC_get_stackid:
1642                 return &bpf_get_stackid_proto_tp;
1643         case BPF_FUNC_get_stack:
1644                 return &bpf_get_stack_proto_tp;
1645         case BPF_FUNC_get_attach_cookie:
1646                 return &bpf_get_attach_cookie_proto_trace;
1647         default:
1648                 return bpf_tracing_func_proto(func_id, prog);
1649         }
1650 }
1651
1652 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1653                                     const struct bpf_prog *prog,
1654                                     struct bpf_insn_access_aux *info)
1655 {
1656         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1657                 return false;
1658         if (type != BPF_READ)
1659                 return false;
1660         if (off % size != 0)
1661                 return false;
1662
1663         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1664         return true;
1665 }
1666
1667 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1668         .get_func_proto  = tp_prog_func_proto,
1669         .is_valid_access = tp_prog_is_valid_access,
1670 };
1671
1672 const struct bpf_prog_ops tracepoint_prog_ops = {
1673 };
1674
1675 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1676            struct bpf_perf_event_value *, buf, u32, size)
1677 {
1678         int err = -EINVAL;
1679
1680         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1681                 goto clear;
1682         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1683                                     &buf->running);
1684         if (unlikely(err))
1685                 goto clear;
1686         return 0;
1687 clear:
1688         memset(buf, 0, size);
1689         return err;
1690 }
1691
1692 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1693          .func           = bpf_perf_prog_read_value,
1694          .gpl_only       = true,
1695          .ret_type       = RET_INTEGER,
1696          .arg1_type      = ARG_PTR_TO_CTX,
1697          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1698          .arg3_type      = ARG_CONST_SIZE,
1699 };
1700
1701 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1702            void *, buf, u32, size, u64, flags)
1703 {
1704         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1705         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1706         u32 to_copy;
1707
1708         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1709                 return -EINVAL;
1710
1711         if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1712                 return -ENOENT;
1713
1714         if (unlikely(!br_stack))
1715                 return -ENOENT;
1716
1717         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1718                 return br_stack->nr * br_entry_size;
1719
1720         if (!buf || (size % br_entry_size != 0))
1721                 return -EINVAL;
1722
1723         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1724         memcpy(buf, br_stack->entries, to_copy);
1725
1726         return to_copy;
1727 }
1728
1729 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1730         .func           = bpf_read_branch_records,
1731         .gpl_only       = true,
1732         .ret_type       = RET_INTEGER,
1733         .arg1_type      = ARG_PTR_TO_CTX,
1734         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1735         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1736         .arg4_type      = ARG_ANYTHING,
1737 };
1738
1739 static const struct bpf_func_proto *
1740 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1741 {
1742         switch (func_id) {
1743         case BPF_FUNC_perf_event_output:
1744                 return &bpf_perf_event_output_proto_tp;
1745         case BPF_FUNC_get_stackid:
1746                 return &bpf_get_stackid_proto_pe;
1747         case BPF_FUNC_get_stack:
1748                 return &bpf_get_stack_proto_pe;
1749         case BPF_FUNC_perf_prog_read_value:
1750                 return &bpf_perf_prog_read_value_proto;
1751         case BPF_FUNC_read_branch_records:
1752                 return &bpf_read_branch_records_proto;
1753         case BPF_FUNC_get_attach_cookie:
1754                 return &bpf_get_attach_cookie_proto_pe;
1755         default:
1756                 return bpf_tracing_func_proto(func_id, prog);
1757         }
1758 }
1759
1760 /*
1761  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1762  * to avoid potential recursive reuse issue when/if tracepoints are added
1763  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1764  *
1765  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1766  * in normal, irq, and nmi context.
1767  */
1768 struct bpf_raw_tp_regs {
1769         struct pt_regs regs[3];
1770 };
1771 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1772 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1773 static struct pt_regs *get_bpf_raw_tp_regs(void)
1774 {
1775         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1776         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1777
1778         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1779                 this_cpu_dec(bpf_raw_tp_nest_level);
1780                 return ERR_PTR(-EBUSY);
1781         }
1782
1783         return &tp_regs->regs[nest_level - 1];
1784 }
1785
1786 static void put_bpf_raw_tp_regs(void)
1787 {
1788         this_cpu_dec(bpf_raw_tp_nest_level);
1789 }
1790
1791 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1792            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1793 {
1794         struct pt_regs *regs = get_bpf_raw_tp_regs();
1795         int ret;
1796
1797         if (IS_ERR(regs))
1798                 return PTR_ERR(regs);
1799
1800         perf_fetch_caller_regs(regs);
1801         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1802
1803         put_bpf_raw_tp_regs();
1804         return ret;
1805 }
1806
1807 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1808         .func           = bpf_perf_event_output_raw_tp,
1809         .gpl_only       = true,
1810         .ret_type       = RET_INTEGER,
1811         .arg1_type      = ARG_PTR_TO_CTX,
1812         .arg2_type      = ARG_CONST_MAP_PTR,
1813         .arg3_type      = ARG_ANYTHING,
1814         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1815         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1816 };
1817
1818 extern const struct bpf_func_proto bpf_skb_output_proto;
1819 extern const struct bpf_func_proto bpf_xdp_output_proto;
1820 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1821
1822 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1823            struct bpf_map *, map, u64, flags)
1824 {
1825         struct pt_regs *regs = get_bpf_raw_tp_regs();
1826         int ret;
1827
1828         if (IS_ERR(regs))
1829                 return PTR_ERR(regs);
1830
1831         perf_fetch_caller_regs(regs);
1832         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1833         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1834                               flags, 0, 0);
1835         put_bpf_raw_tp_regs();
1836         return ret;
1837 }
1838
1839 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1840         .func           = bpf_get_stackid_raw_tp,
1841         .gpl_only       = true,
1842         .ret_type       = RET_INTEGER,
1843         .arg1_type      = ARG_PTR_TO_CTX,
1844         .arg2_type      = ARG_CONST_MAP_PTR,
1845         .arg3_type      = ARG_ANYTHING,
1846 };
1847
1848 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1849            void *, buf, u32, size, u64, flags)
1850 {
1851         struct pt_regs *regs = get_bpf_raw_tp_regs();
1852         int ret;
1853
1854         if (IS_ERR(regs))
1855                 return PTR_ERR(regs);
1856
1857         perf_fetch_caller_regs(regs);
1858         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1859                             (unsigned long) size, flags, 0);
1860         put_bpf_raw_tp_regs();
1861         return ret;
1862 }
1863
1864 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1865         .func           = bpf_get_stack_raw_tp,
1866         .gpl_only       = true,
1867         .ret_type       = RET_INTEGER,
1868         .arg1_type      = ARG_PTR_TO_CTX,
1869         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1870         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1871         .arg4_type      = ARG_ANYTHING,
1872 };
1873
1874 static const struct bpf_func_proto *
1875 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1876 {
1877         switch (func_id) {
1878         case BPF_FUNC_perf_event_output:
1879                 return &bpf_perf_event_output_proto_raw_tp;
1880         case BPF_FUNC_get_stackid:
1881                 return &bpf_get_stackid_proto_raw_tp;
1882         case BPF_FUNC_get_stack:
1883                 return &bpf_get_stack_proto_raw_tp;
1884         default:
1885                 return bpf_tracing_func_proto(func_id, prog);
1886         }
1887 }
1888
1889 const struct bpf_func_proto *
1890 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1891 {
1892         const struct bpf_func_proto *fn;
1893
1894         switch (func_id) {
1895 #ifdef CONFIG_NET
1896         case BPF_FUNC_skb_output:
1897                 return &bpf_skb_output_proto;
1898         case BPF_FUNC_xdp_output:
1899                 return &bpf_xdp_output_proto;
1900         case BPF_FUNC_skc_to_tcp6_sock:
1901                 return &bpf_skc_to_tcp6_sock_proto;
1902         case BPF_FUNC_skc_to_tcp_sock:
1903                 return &bpf_skc_to_tcp_sock_proto;
1904         case BPF_FUNC_skc_to_tcp_timewait_sock:
1905                 return &bpf_skc_to_tcp_timewait_sock_proto;
1906         case BPF_FUNC_skc_to_tcp_request_sock:
1907                 return &bpf_skc_to_tcp_request_sock_proto;
1908         case BPF_FUNC_skc_to_udp6_sock:
1909                 return &bpf_skc_to_udp6_sock_proto;
1910         case BPF_FUNC_skc_to_unix_sock:
1911                 return &bpf_skc_to_unix_sock_proto;
1912         case BPF_FUNC_skc_to_mptcp_sock:
1913                 return &bpf_skc_to_mptcp_sock_proto;
1914         case BPF_FUNC_sk_storage_get:
1915                 return &bpf_sk_storage_get_tracing_proto;
1916         case BPF_FUNC_sk_storage_delete:
1917                 return &bpf_sk_storage_delete_tracing_proto;
1918         case BPF_FUNC_sock_from_file:
1919                 return &bpf_sock_from_file_proto;
1920         case BPF_FUNC_get_socket_cookie:
1921                 return &bpf_get_socket_ptr_cookie_proto;
1922         case BPF_FUNC_xdp_get_buff_len:
1923                 return &bpf_xdp_get_buff_len_trace_proto;
1924 #endif
1925         case BPF_FUNC_seq_printf:
1926                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1927                        &bpf_seq_printf_proto :
1928                        NULL;
1929         case BPF_FUNC_seq_write:
1930                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1931                        &bpf_seq_write_proto :
1932                        NULL;
1933         case BPF_FUNC_seq_printf_btf:
1934                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1935                        &bpf_seq_printf_btf_proto :
1936                        NULL;
1937         case BPF_FUNC_d_path:
1938                 return &bpf_d_path_proto;
1939         case BPF_FUNC_get_func_arg:
1940                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1941         case BPF_FUNC_get_func_ret:
1942                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1943         case BPF_FUNC_get_func_arg_cnt:
1944                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1945         case BPF_FUNC_get_attach_cookie:
1946                 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
1947         default:
1948                 fn = raw_tp_prog_func_proto(func_id, prog);
1949                 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1950                         fn = bpf_iter_get_func_proto(func_id, prog);
1951                 return fn;
1952         }
1953 }
1954
1955 static bool raw_tp_prog_is_valid_access(int off, int size,
1956                                         enum bpf_access_type type,
1957                                         const struct bpf_prog *prog,
1958                                         struct bpf_insn_access_aux *info)
1959 {
1960         return bpf_tracing_ctx_access(off, size, type);
1961 }
1962
1963 static bool tracing_prog_is_valid_access(int off, int size,
1964                                          enum bpf_access_type type,
1965                                          const struct bpf_prog *prog,
1966                                          struct bpf_insn_access_aux *info)
1967 {
1968         return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1969 }
1970
1971 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1972                                      const union bpf_attr *kattr,
1973                                      union bpf_attr __user *uattr)
1974 {
1975         return -ENOTSUPP;
1976 }
1977
1978 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1979         .get_func_proto  = raw_tp_prog_func_proto,
1980         .is_valid_access = raw_tp_prog_is_valid_access,
1981 };
1982
1983 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1984 #ifdef CONFIG_NET
1985         .test_run = bpf_prog_test_run_raw_tp,
1986 #endif
1987 };
1988
1989 const struct bpf_verifier_ops tracing_verifier_ops = {
1990         .get_func_proto  = tracing_prog_func_proto,
1991         .is_valid_access = tracing_prog_is_valid_access,
1992 };
1993
1994 const struct bpf_prog_ops tracing_prog_ops = {
1995         .test_run = bpf_prog_test_run_tracing,
1996 };
1997
1998 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1999                                                  enum bpf_access_type type,
2000                                                  const struct bpf_prog *prog,
2001                                                  struct bpf_insn_access_aux *info)
2002 {
2003         if (off == 0) {
2004                 if (size != sizeof(u64) || type != BPF_READ)
2005                         return false;
2006                 info->reg_type = PTR_TO_TP_BUFFER;
2007         }
2008         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2009 }
2010
2011 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2012         .get_func_proto  = raw_tp_prog_func_proto,
2013         .is_valid_access = raw_tp_writable_prog_is_valid_access,
2014 };
2015
2016 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2017 };
2018
2019 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2020                                     const struct bpf_prog *prog,
2021                                     struct bpf_insn_access_aux *info)
2022 {
2023         const int size_u64 = sizeof(u64);
2024
2025         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2026                 return false;
2027         if (type != BPF_READ)
2028                 return false;
2029         if (off % size != 0) {
2030                 if (sizeof(unsigned long) != 4)
2031                         return false;
2032                 if (size != 8)
2033                         return false;
2034                 if (off % size != 4)
2035                         return false;
2036         }
2037
2038         switch (off) {
2039         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2040                 bpf_ctx_record_field_size(info, size_u64);
2041                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2042                         return false;
2043                 break;
2044         case bpf_ctx_range(struct bpf_perf_event_data, addr):
2045                 bpf_ctx_record_field_size(info, size_u64);
2046                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2047                         return false;
2048                 break;
2049         default:
2050                 if (size != sizeof(long))
2051                         return false;
2052         }
2053
2054         return true;
2055 }
2056
2057 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2058                                       const struct bpf_insn *si,
2059                                       struct bpf_insn *insn_buf,
2060                                       struct bpf_prog *prog, u32 *target_size)
2061 {
2062         struct bpf_insn *insn = insn_buf;
2063
2064         switch (si->off) {
2065         case offsetof(struct bpf_perf_event_data, sample_period):
2066                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2067                                                        data), si->dst_reg, si->src_reg,
2068                                       offsetof(struct bpf_perf_event_data_kern, data));
2069                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2070                                       bpf_target_off(struct perf_sample_data, period, 8,
2071                                                      target_size));
2072                 break;
2073         case offsetof(struct bpf_perf_event_data, addr):
2074                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2075                                                        data), si->dst_reg, si->src_reg,
2076                                       offsetof(struct bpf_perf_event_data_kern, data));
2077                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2078                                       bpf_target_off(struct perf_sample_data, addr, 8,
2079                                                      target_size));
2080                 break;
2081         default:
2082                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2083                                                        regs), si->dst_reg, si->src_reg,
2084                                       offsetof(struct bpf_perf_event_data_kern, regs));
2085                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2086                                       si->off);
2087                 break;
2088         }
2089
2090         return insn - insn_buf;
2091 }
2092
2093 const struct bpf_verifier_ops perf_event_verifier_ops = {
2094         .get_func_proto         = pe_prog_func_proto,
2095         .is_valid_access        = pe_prog_is_valid_access,
2096         .convert_ctx_access     = pe_prog_convert_ctx_access,
2097 };
2098
2099 const struct bpf_prog_ops perf_event_prog_ops = {
2100 };
2101
2102 static DEFINE_MUTEX(bpf_event_mutex);
2103
2104 #define BPF_TRACE_MAX_PROGS 64
2105
2106 int perf_event_attach_bpf_prog(struct perf_event *event,
2107                                struct bpf_prog *prog,
2108                                u64 bpf_cookie)
2109 {
2110         struct bpf_prog_array *old_array;
2111         struct bpf_prog_array *new_array;
2112         int ret = -EEXIST;
2113
2114         /*
2115          * Kprobe override only works if they are on the function entry,
2116          * and only if they are on the opt-in list.
2117          */
2118         if (prog->kprobe_override &&
2119             (!trace_kprobe_on_func_entry(event->tp_event) ||
2120              !trace_kprobe_error_injectable(event->tp_event)))
2121                 return -EINVAL;
2122
2123         mutex_lock(&bpf_event_mutex);
2124
2125         if (event->prog)
2126                 goto unlock;
2127
2128         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2129         if (old_array &&
2130             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2131                 ret = -E2BIG;
2132                 goto unlock;
2133         }
2134
2135         ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2136         if (ret < 0)
2137                 goto unlock;
2138
2139         /* set the new array to event->tp_event and set event->prog */
2140         event->prog = prog;
2141         event->bpf_cookie = bpf_cookie;
2142         rcu_assign_pointer(event->tp_event->prog_array, new_array);
2143         bpf_prog_array_free_sleepable(old_array);
2144
2145 unlock:
2146         mutex_unlock(&bpf_event_mutex);
2147         return ret;
2148 }
2149
2150 void perf_event_detach_bpf_prog(struct perf_event *event)
2151 {
2152         struct bpf_prog_array *old_array;
2153         struct bpf_prog_array *new_array;
2154         int ret;
2155
2156         mutex_lock(&bpf_event_mutex);
2157
2158         if (!event->prog)
2159                 goto unlock;
2160
2161         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2162         ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2163         if (ret == -ENOENT)
2164                 goto unlock;
2165         if (ret < 0) {
2166                 bpf_prog_array_delete_safe(old_array, event->prog);
2167         } else {
2168                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2169                 bpf_prog_array_free_sleepable(old_array);
2170         }
2171
2172         bpf_prog_put(event->prog);
2173         event->prog = NULL;
2174
2175 unlock:
2176         mutex_unlock(&bpf_event_mutex);
2177 }
2178
2179 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2180 {
2181         struct perf_event_query_bpf __user *uquery = info;
2182         struct perf_event_query_bpf query = {};
2183         struct bpf_prog_array *progs;
2184         u32 *ids, prog_cnt, ids_len;
2185         int ret;
2186
2187         if (!perfmon_capable())
2188                 return -EPERM;
2189         if (event->attr.type != PERF_TYPE_TRACEPOINT)
2190                 return -EINVAL;
2191         if (copy_from_user(&query, uquery, sizeof(query)))
2192                 return -EFAULT;
2193
2194         ids_len = query.ids_len;
2195         if (ids_len > BPF_TRACE_MAX_PROGS)
2196                 return -E2BIG;
2197         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2198         if (!ids)
2199                 return -ENOMEM;
2200         /*
2201          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2202          * is required when user only wants to check for uquery->prog_cnt.
2203          * There is no need to check for it since the case is handled
2204          * gracefully in bpf_prog_array_copy_info.
2205          */
2206
2207         mutex_lock(&bpf_event_mutex);
2208         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2209         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2210         mutex_unlock(&bpf_event_mutex);
2211
2212         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2213             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2214                 ret = -EFAULT;
2215
2216         kfree(ids);
2217         return ret;
2218 }
2219
2220 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2221 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2222
2223 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2224 {
2225         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2226
2227         for (; btp < __stop__bpf_raw_tp; btp++) {
2228                 if (!strcmp(btp->tp->name, name))
2229                         return btp;
2230         }
2231
2232         return bpf_get_raw_tracepoint_module(name);
2233 }
2234
2235 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2236 {
2237         struct module *mod;
2238
2239         preempt_disable();
2240         mod = __module_address((unsigned long)btp);
2241         module_put(mod);
2242         preempt_enable();
2243 }
2244
2245 static __always_inline
2246 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2247 {
2248         cant_sleep();
2249         if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2250                 bpf_prog_inc_misses_counter(prog);
2251                 goto out;
2252         }
2253         rcu_read_lock();
2254         (void) bpf_prog_run(prog, args);
2255         rcu_read_unlock();
2256 out:
2257         this_cpu_dec(*(prog->active));
2258 }
2259
2260 #define UNPACK(...)                     __VA_ARGS__
2261 #define REPEAT_1(FN, DL, X, ...)        FN(X)
2262 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2263 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2264 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2265 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2266 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2267 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2268 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2269 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2270 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2271 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2272 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2273 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
2274
2275 #define SARG(X)         u64 arg##X
2276 #define COPY(X)         args[X] = arg##X
2277
2278 #define __DL_COM        (,)
2279 #define __DL_SEM        (;)
2280
2281 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2282
2283 #define BPF_TRACE_DEFN_x(x)                                             \
2284         void bpf_trace_run##x(struct bpf_prog *prog,                    \
2285                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
2286         {                                                               \
2287                 u64 args[x];                                            \
2288                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
2289                 __bpf_trace_run(prog, args);                            \
2290         }                                                               \
2291         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2292 BPF_TRACE_DEFN_x(1);
2293 BPF_TRACE_DEFN_x(2);
2294 BPF_TRACE_DEFN_x(3);
2295 BPF_TRACE_DEFN_x(4);
2296 BPF_TRACE_DEFN_x(5);
2297 BPF_TRACE_DEFN_x(6);
2298 BPF_TRACE_DEFN_x(7);
2299 BPF_TRACE_DEFN_x(8);
2300 BPF_TRACE_DEFN_x(9);
2301 BPF_TRACE_DEFN_x(10);
2302 BPF_TRACE_DEFN_x(11);
2303 BPF_TRACE_DEFN_x(12);
2304
2305 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2306 {
2307         struct tracepoint *tp = btp->tp;
2308
2309         /*
2310          * check that program doesn't access arguments beyond what's
2311          * available in this tracepoint
2312          */
2313         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2314                 return -EINVAL;
2315
2316         if (prog->aux->max_tp_access > btp->writable_size)
2317                 return -EINVAL;
2318
2319         return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2320                                                    prog);
2321 }
2322
2323 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2324 {
2325         return __bpf_probe_register(btp, prog);
2326 }
2327
2328 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2329 {
2330         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2331 }
2332
2333 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2334                             u32 *fd_type, const char **buf,
2335                             u64 *probe_offset, u64 *probe_addr)
2336 {
2337         bool is_tracepoint, is_syscall_tp;
2338         struct bpf_prog *prog;
2339         int flags, err = 0;
2340
2341         prog = event->prog;
2342         if (!prog)
2343                 return -ENOENT;
2344
2345         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2346         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2347                 return -EOPNOTSUPP;
2348
2349         *prog_id = prog->aux->id;
2350         flags = event->tp_event->flags;
2351         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2352         is_syscall_tp = is_syscall_trace_event(event->tp_event);
2353
2354         if (is_tracepoint || is_syscall_tp) {
2355                 *buf = is_tracepoint ? event->tp_event->tp->name
2356                                      : event->tp_event->name;
2357                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2358                 *probe_offset = 0x0;
2359                 *probe_addr = 0x0;
2360         } else {
2361                 /* kprobe/uprobe */
2362                 err = -EOPNOTSUPP;
2363 #ifdef CONFIG_KPROBE_EVENTS
2364                 if (flags & TRACE_EVENT_FL_KPROBE)
2365                         err = bpf_get_kprobe_info(event, fd_type, buf,
2366                                                   probe_offset, probe_addr,
2367                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2368 #endif
2369 #ifdef CONFIG_UPROBE_EVENTS
2370                 if (flags & TRACE_EVENT_FL_UPROBE)
2371                         err = bpf_get_uprobe_info(event, fd_type, buf,
2372                                                   probe_offset,
2373                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2374 #endif
2375         }
2376
2377         return err;
2378 }
2379
2380 static int __init send_signal_irq_work_init(void)
2381 {
2382         int cpu;
2383         struct send_signal_irq_work *work;
2384
2385         for_each_possible_cpu(cpu) {
2386                 work = per_cpu_ptr(&send_signal_work, cpu);
2387                 init_irq_work(&work->irq_work, do_bpf_send_signal);
2388         }
2389         return 0;
2390 }
2391
2392 subsys_initcall(send_signal_irq_work_init);
2393
2394 #ifdef CONFIG_MODULES
2395 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2396                             void *module)
2397 {
2398         struct bpf_trace_module *btm, *tmp;
2399         struct module *mod = module;
2400         int ret = 0;
2401
2402         if (mod->num_bpf_raw_events == 0 ||
2403             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2404                 goto out;
2405
2406         mutex_lock(&bpf_module_mutex);
2407
2408         switch (op) {
2409         case MODULE_STATE_COMING:
2410                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2411                 if (btm) {
2412                         btm->module = module;
2413                         list_add(&btm->list, &bpf_trace_modules);
2414                 } else {
2415                         ret = -ENOMEM;
2416                 }
2417                 break;
2418         case MODULE_STATE_GOING:
2419                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2420                         if (btm->module == module) {
2421                                 list_del(&btm->list);
2422                                 kfree(btm);
2423                                 break;
2424                         }
2425                 }
2426                 break;
2427         }
2428
2429         mutex_unlock(&bpf_module_mutex);
2430
2431 out:
2432         return notifier_from_errno(ret);
2433 }
2434
2435 static struct notifier_block bpf_module_nb = {
2436         .notifier_call = bpf_event_notify,
2437 };
2438
2439 static int __init bpf_event_init(void)
2440 {
2441         register_module_notifier(&bpf_module_nb);
2442         return 0;
2443 }
2444
2445 fs_initcall(bpf_event_init);
2446 #endif /* CONFIG_MODULES */
2447
2448 #ifdef CONFIG_FPROBE
2449 struct bpf_kprobe_multi_link {
2450         struct bpf_link link;
2451         struct fprobe fp;
2452         unsigned long *addrs;
2453         u64 *cookies;
2454         u32 cnt;
2455 };
2456
2457 struct bpf_kprobe_multi_run_ctx {
2458         struct bpf_run_ctx run_ctx;
2459         struct bpf_kprobe_multi_link *link;
2460         unsigned long entry_ip;
2461 };
2462
2463 struct user_syms {
2464         const char **syms;
2465         char *buf;
2466 };
2467
2468 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2469 {
2470         unsigned long __user usymbol;
2471         const char **syms = NULL;
2472         char *buf = NULL, *p;
2473         int err = -ENOMEM;
2474         unsigned int i;
2475
2476         syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2477         if (!syms)
2478                 goto error;
2479
2480         buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2481         if (!buf)
2482                 goto error;
2483
2484         for (p = buf, i = 0; i < cnt; i++) {
2485                 if (__get_user(usymbol, usyms + i)) {
2486                         err = -EFAULT;
2487                         goto error;
2488                 }
2489                 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2490                 if (err == KSYM_NAME_LEN)
2491                         err = -E2BIG;
2492                 if (err < 0)
2493                         goto error;
2494                 syms[i] = p;
2495                 p += err + 1;
2496         }
2497
2498         us->syms = syms;
2499         us->buf = buf;
2500         return 0;
2501
2502 error:
2503         if (err) {
2504                 kvfree(syms);
2505                 kvfree(buf);
2506         }
2507         return err;
2508 }
2509
2510 static void free_user_syms(struct user_syms *us)
2511 {
2512         kvfree(us->syms);
2513         kvfree(us->buf);
2514 }
2515
2516 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2517 {
2518         struct bpf_kprobe_multi_link *kmulti_link;
2519
2520         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2521         unregister_fprobe(&kmulti_link->fp);
2522 }
2523
2524 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2525 {
2526         struct bpf_kprobe_multi_link *kmulti_link;
2527
2528         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2529         kvfree(kmulti_link->addrs);
2530         kvfree(kmulti_link->cookies);
2531         kfree(kmulti_link);
2532 }
2533
2534 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2535         .release = bpf_kprobe_multi_link_release,
2536         .dealloc = bpf_kprobe_multi_link_dealloc,
2537 };
2538
2539 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2540 {
2541         const struct bpf_kprobe_multi_link *link = priv;
2542         unsigned long *addr_a = a, *addr_b = b;
2543         u64 *cookie_a, *cookie_b;
2544
2545         cookie_a = link->cookies + (addr_a - link->addrs);
2546         cookie_b = link->cookies + (addr_b - link->addrs);
2547
2548         /* swap addr_a/addr_b and cookie_a/cookie_b values */
2549         swap(*addr_a, *addr_b);
2550         swap(*cookie_a, *cookie_b);
2551 }
2552
2553 static int __bpf_kprobe_multi_cookie_cmp(const void *a, const void *b)
2554 {
2555         const unsigned long *addr_a = a, *addr_b = b;
2556
2557         if (*addr_a == *addr_b)
2558                 return 0;
2559         return *addr_a < *addr_b ? -1 : 1;
2560 }
2561
2562 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2563 {
2564         return __bpf_kprobe_multi_cookie_cmp(a, b);
2565 }
2566
2567 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2568 {
2569         struct bpf_kprobe_multi_run_ctx *run_ctx;
2570         struct bpf_kprobe_multi_link *link;
2571         u64 *cookie, entry_ip;
2572         unsigned long *addr;
2573
2574         if (WARN_ON_ONCE(!ctx))
2575                 return 0;
2576         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2577         link = run_ctx->link;
2578         if (!link->cookies)
2579                 return 0;
2580         entry_ip = run_ctx->entry_ip;
2581         addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2582                        __bpf_kprobe_multi_cookie_cmp);
2583         if (!addr)
2584                 return 0;
2585         cookie = link->cookies + (addr - link->addrs);
2586         return *cookie;
2587 }
2588
2589 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2590 {
2591         struct bpf_kprobe_multi_run_ctx *run_ctx;
2592
2593         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2594         return run_ctx->entry_ip;
2595 }
2596
2597 static int
2598 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2599                            unsigned long entry_ip, struct pt_regs *regs)
2600 {
2601         struct bpf_kprobe_multi_run_ctx run_ctx = {
2602                 .link = link,
2603                 .entry_ip = entry_ip,
2604         };
2605         struct bpf_run_ctx *old_run_ctx;
2606         int err;
2607
2608         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2609                 err = 0;
2610                 goto out;
2611         }
2612
2613         migrate_disable();
2614         rcu_read_lock();
2615         old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2616         err = bpf_prog_run(link->link.prog, regs);
2617         bpf_reset_run_ctx(old_run_ctx);
2618         rcu_read_unlock();
2619         migrate_enable();
2620
2621  out:
2622         __this_cpu_dec(bpf_prog_active);
2623         return err;
2624 }
2625
2626 static void
2627 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2628                           struct pt_regs *regs)
2629 {
2630         struct bpf_kprobe_multi_link *link;
2631
2632         link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2633         kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2634 }
2635
2636 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2637 {
2638         const char **str_a = (const char **) a;
2639         const char **str_b = (const char **) b;
2640
2641         return strcmp(*str_a, *str_b);
2642 }
2643
2644 struct multi_symbols_sort {
2645         const char **funcs;
2646         u64 *cookies;
2647 };
2648
2649 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2650 {
2651         const struct multi_symbols_sort *data = priv;
2652         const char **name_a = a, **name_b = b;
2653
2654         swap(*name_a, *name_b);
2655
2656         /* If defined, swap also related cookies. */
2657         if (data->cookies) {
2658                 u64 *cookie_a, *cookie_b;
2659
2660                 cookie_a = data->cookies + (name_a - data->funcs);
2661                 cookie_b = data->cookies + (name_b - data->funcs);
2662                 swap(*cookie_a, *cookie_b);
2663         }
2664 }
2665
2666 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2667 {
2668         struct bpf_kprobe_multi_link *link = NULL;
2669         struct bpf_link_primer link_primer;
2670         void __user *ucookies;
2671         unsigned long *addrs;
2672         u32 flags, cnt, size;
2673         void __user *uaddrs;
2674         u64 *cookies = NULL;
2675         void __user *usyms;
2676         int err;
2677
2678         /* no support for 32bit archs yet */
2679         if (sizeof(u64) != sizeof(void *))
2680                 return -EOPNOTSUPP;
2681
2682         if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2683                 return -EINVAL;
2684
2685         flags = attr->link_create.kprobe_multi.flags;
2686         if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2687                 return -EINVAL;
2688
2689         uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2690         usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2691         if (!!uaddrs == !!usyms)
2692                 return -EINVAL;
2693
2694         cnt = attr->link_create.kprobe_multi.cnt;
2695         if (!cnt)
2696                 return -EINVAL;
2697
2698         size = cnt * sizeof(*addrs);
2699         addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2700         if (!addrs)
2701                 return -ENOMEM;
2702
2703         ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2704         if (ucookies) {
2705                 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2706                 if (!cookies) {
2707                         err = -ENOMEM;
2708                         goto error;
2709                 }
2710                 if (copy_from_user(cookies, ucookies, size)) {
2711                         err = -EFAULT;
2712                         goto error;
2713                 }
2714         }
2715
2716         if (uaddrs) {
2717                 if (copy_from_user(addrs, uaddrs, size)) {
2718                         err = -EFAULT;
2719                         goto error;
2720                 }
2721         } else {
2722                 struct multi_symbols_sort data = {
2723                         .cookies = cookies,
2724                 };
2725                 struct user_syms us;
2726
2727                 err = copy_user_syms(&us, usyms, cnt);
2728                 if (err)
2729                         goto error;
2730
2731                 if (cookies)
2732                         data.funcs = us.syms;
2733
2734                 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2735                        symbols_swap_r, &data);
2736
2737                 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2738                 free_user_syms(&us);
2739                 if (err)
2740                         goto error;
2741         }
2742
2743         link = kzalloc(sizeof(*link), GFP_KERNEL);
2744         if (!link) {
2745                 err = -ENOMEM;
2746                 goto error;
2747         }
2748
2749         bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2750                       &bpf_kprobe_multi_link_lops, prog);
2751
2752         err = bpf_link_prime(&link->link, &link_primer);
2753         if (err)
2754                 goto error;
2755
2756         if (flags & BPF_F_KPROBE_MULTI_RETURN)
2757                 link->fp.exit_handler = kprobe_multi_link_handler;
2758         else
2759                 link->fp.entry_handler = kprobe_multi_link_handler;
2760
2761         link->addrs = addrs;
2762         link->cookies = cookies;
2763         link->cnt = cnt;
2764
2765         if (cookies) {
2766                 /*
2767                  * Sorting addresses will trigger sorting cookies as well
2768                  * (check bpf_kprobe_multi_cookie_swap). This way we can
2769                  * find cookie based on the address in bpf_get_attach_cookie
2770                  * helper.
2771                  */
2772                 sort_r(addrs, cnt, sizeof(*addrs),
2773                        bpf_kprobe_multi_cookie_cmp,
2774                        bpf_kprobe_multi_cookie_swap,
2775                        link);
2776         }
2777
2778         err = register_fprobe_ips(&link->fp, addrs, cnt);
2779         if (err) {
2780                 bpf_link_cleanup(&link_primer);
2781                 return err;
2782         }
2783
2784         return bpf_link_settle(&link_primer);
2785
2786 error:
2787         kfree(link);
2788         kvfree(addrs);
2789         kvfree(cookies);
2790         return err;
2791 }
2792 #else /* !CONFIG_FPROBE */
2793 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2794 {
2795         return -EOPNOTSUPP;
2796 }
2797 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2798 {
2799         return 0;
2800 }
2801 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2802 {
2803         return 0;
2804 }
2805 #endif