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         put_task_struct(work->task);
 836 }
 837
 838 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 839 {
 840         struct send_signal_irq_work *work = NULL;
 841
 842         /* Similar to bpf_probe_write_user, task needs to be
 843          * in a sound condition and kernel memory access be
 844          * permitted in order to send signal to the current
 845          * task.
 846          */
 847         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 848                 return -EPERM;
 849         if (unlikely(!nmi_uaccess_okay()))
 850                 return -EPERM;
 851         /* Task should not be pid=1 to avoid kernel panic. */
 852         if (unlikely(is_global_init(current)))
 853                 return -EPERM;
 854
 855         if (irqs_disabled()) {
 856                 /* Do an early check on signal validity. Otherwise,
 857                  * the error is lost in deferred irq_work.
 858                  */
 859                 if (unlikely(!valid_signal(sig)))
 860                         return -EINVAL;
 861
 862                 work = this_cpu_ptr(&send_signal_work);
 863                 if (irq_work_is_busy(&work->irq_work))
 864                         return -EBUSY;
 865
 866                 /* Add the current task, which is the target of sending signal,
 867                  * to the irq_work. The current task may change when queued
 868                  * irq works get executed.
 869                  */
 870                 work->task = get_task_struct(current);
 871                 work->sig = sig;
 872                 work->type = type;
 873                 irq_work_queue(&work->irq_work);
 874                 return 0;
 875         }
 876
 877         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 878 }
 879
 880 BPF_CALL_1(bpf_send_signal, u32, sig)
 881 {
 882         return bpf_send_signal_common(sig, PIDTYPE_TGID);
 883 }
 884
 885 static const struct bpf_func_proto bpf_send_signal_proto = {
 886         .func           = bpf_send_signal,
 887         .gpl_only       = false,
 888         .ret_type       = RET_INTEGER,
 889         .arg1_type      = ARG_ANYTHING,
 890 };
 891
 892 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 893 {
 894         return bpf_send_signal_common(sig, PIDTYPE_PID);
 895 }
 896
 897 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 898         .func           = bpf_send_signal_thread,
 899         .gpl_only       = false,
 900         .ret_type       = RET_INTEGER,
 901         .arg1_type      = ARG_ANYTHING,
 902 };
 903
 904 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 905 {
 906         struct path copy;
 907         long len;
 908         char *p;
 909
 910         if (!sz)
 911                 return 0;
 912
 913         /*
 914          * The path pointer is verified as trusted and safe to use,
 915          * but let's double check it's valid anyway to workaround
 916          * potentially broken verifier.
 917          */
 918         len = copy_from_kernel_nofault(&copy, path, sizeof(*path));
 919         if (len < 0)
 920                 return len;
 921
 922         p = d_path(&copy, buf, sz);
 923         if (IS_ERR(p)) {
 924                 len = PTR_ERR(p);
 925         } else {
 926                 len = buf + sz - p;
 927                 memmove(buf, p, len);
 928         }
 929
 930         return len;
 931 }
 932
 933 BTF_SET_START(btf_allowlist_d_path)
 934 #ifdef CONFIG_SECURITY
 935 BTF_ID(func, security_file_permission)
 936 BTF_ID(func, security_inode_getattr)
 937 BTF_ID(func, security_file_open)
 938 #endif
 939 #ifdef CONFIG_SECURITY_PATH
 940 BTF_ID(func, security_path_truncate)
 941 #endif
 942 BTF_ID(func, vfs_truncate)
 943 BTF_ID(func, vfs_fallocate)
 944 BTF_ID(func, dentry_open)
 945 BTF_ID(func, vfs_getattr)
 946 BTF_ID(func, filp_close)
 947 BTF_SET_END(btf_allowlist_d_path)
 948
 949 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 950 {
 951         if (prog->type == BPF_PROG_TYPE_TRACING &&
 952             prog->expected_attach_type == BPF_TRACE_ITER)
 953                 return true;
 954
 955         if (prog->type == BPF_PROG_TYPE_LSM)
 956                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 957
 958         return btf_id_set_contains(&btf_allowlist_d_path,
 959                                    prog->aux->attach_btf_id);
 960 }
 961
 962 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 963
 964 static const struct bpf_func_proto bpf_d_path_proto = {
 965         .func           = bpf_d_path,
 966         .gpl_only       = false,
 967         .ret_type       = RET_INTEGER,
 968         .arg1_type      = ARG_PTR_TO_BTF_ID,
 969         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
 970         .arg2_type      = ARG_PTR_TO_MEM,
 971         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 972         .allowed        = bpf_d_path_allowed,
 973 };
 974
 975 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
 976                          BTF_F_PTR_RAW | BTF_F_ZERO)
 977
 978 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 979                                   u64 flags, const struct btf **btf,
 980                                   s32 *btf_id)
 981 {
 982         const struct btf_type *t;
 983
 984         if (unlikely(flags & ~(BTF_F_ALL)))
 985                 return -EINVAL;
 986
 987         if (btf_ptr_size != sizeof(struct btf_ptr))
 988                 return -EINVAL;
 989
 990         *btf = bpf_get_btf_vmlinux();
 991
 992         if (IS_ERR_OR_NULL(*btf))
 993                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 994
 995         if (ptr->type_id > 0)
 996                 *btf_id = ptr->type_id;
 997         else
 998                 return -EINVAL;
 999
1000         if (*btf_id > 0)
1001                 t = btf_type_by_id(*btf, *btf_id);
1002         if (*btf_id <= 0 || !t)
1003                 return -ENOENT;
1004
1005         return 0;
1006 }
1007
1008 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1009            u32, btf_ptr_size, u64, flags)
1010 {
1011         const struct btf *btf;
1012         s32 btf_id;
1013         int ret;
1014
1015         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1016         if (ret)
1017                 return ret;
1018
1019         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1020                                       flags);
1021 }
1022
1023 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1024         .func           = bpf_snprintf_btf,
1025         .gpl_only       = false,
1026         .ret_type       = RET_INTEGER,
1027         .arg1_type      = ARG_PTR_TO_MEM,
1028         .arg2_type      = ARG_CONST_SIZE,
1029         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1030         .arg4_type      = ARG_CONST_SIZE,
1031         .arg5_type      = ARG_ANYTHING,
1032 };
1033
1034 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1035 {
1036         /* This helper call is inlined by verifier. */
1037         return ((u64 *)ctx)[-2];
1038 }
1039
1040 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1041         .func           = bpf_get_func_ip_tracing,
1042         .gpl_only       = true,
1043         .ret_type       = RET_INTEGER,
1044         .arg1_type      = ARG_PTR_TO_CTX,
1045 };
1046
1047 #ifdef CONFIG_X86_KERNEL_IBT
1048 static unsigned long get_entry_ip(unsigned long fentry_ip)
1049 {
1050         u32 instr;
1051
1052         /* Being extra safe in here in case entry ip is on the page-edge. */
1053         if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1))
1054                 return fentry_ip;
1055         if (is_endbr(instr))
1056                 fentry_ip -= ENDBR_INSN_SIZE;
1057         return fentry_ip;
1058 }
1059 #else
1060 #define get_entry_ip(fentry_ip) fentry_ip
1061 #endif
1062
1063 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1064 {
1065         struct kprobe *kp = kprobe_running();
1066
1067         if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1068                 return 0;
1069
1070         return get_entry_ip((uintptr_t)kp->addr);
1071 }
1072
1073 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1074         .func           = bpf_get_func_ip_kprobe,
1075         .gpl_only       = true,
1076         .ret_type       = RET_INTEGER,
1077         .arg1_type      = ARG_PTR_TO_CTX,
1078 };
1079
1080 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1081 {
1082         return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1083 }
1084
1085 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1086         .func           = bpf_get_func_ip_kprobe_multi,
1087         .gpl_only       = false,
1088         .ret_type       = RET_INTEGER,
1089         .arg1_type      = ARG_PTR_TO_CTX,
1090 };
1091
1092 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1093 {
1094         return bpf_kprobe_multi_cookie(current->bpf_ctx);
1095 }
1096
1097 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1098         .func           = bpf_get_attach_cookie_kprobe_multi,
1099         .gpl_only       = false,
1100         .ret_type       = RET_INTEGER,
1101         .arg1_type      = ARG_PTR_TO_CTX,
1102 };
1103
1104 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1105 {
1106         struct bpf_trace_run_ctx *run_ctx;
1107
1108         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1109         return run_ctx->bpf_cookie;
1110 }
1111
1112 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1113         .func           = bpf_get_attach_cookie_trace,
1114         .gpl_only       = false,
1115         .ret_type       = RET_INTEGER,
1116         .arg1_type      = ARG_PTR_TO_CTX,
1117 };
1118
1119 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1120 {
1121         return ctx->event->bpf_cookie;
1122 }
1123
1124 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1125         .func           = bpf_get_attach_cookie_pe,
1126         .gpl_only       = false,
1127         .ret_type       = RET_INTEGER,
1128         .arg1_type      = ARG_PTR_TO_CTX,
1129 };
1130
1131 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1132 {
1133         struct bpf_trace_run_ctx *run_ctx;
1134
1135         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1136         return run_ctx->bpf_cookie;
1137 }
1138
1139 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1140         .func           = bpf_get_attach_cookie_tracing,
1141         .gpl_only       = false,
1142         .ret_type       = RET_INTEGER,
1143         .arg1_type      = ARG_PTR_TO_CTX,
1144 };
1145
1146 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1147 {
1148 #ifndef CONFIG_X86
1149         return -ENOENT;
1150 #else
1151         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1152         u32 entry_cnt = size / br_entry_size;
1153
1154         entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1155
1156         if (unlikely(flags))
1157                 return -EINVAL;
1158
1159         if (!entry_cnt)
1160                 return -ENOENT;
1161
1162         return entry_cnt * br_entry_size;
1163 #endif
1164 }
1165
1166 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1167         .func           = bpf_get_branch_snapshot,
1168         .gpl_only       = true,
1169         .ret_type       = RET_INTEGER,
1170         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1171         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1172 };
1173
1174 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1175 {
1176         /* This helper call is inlined by verifier. */
1177         u64 nr_args = ((u64 *)ctx)[-1];
1178
1179         if ((u64) n >= nr_args)
1180                 return -EINVAL;
1181         *value = ((u64 *)ctx)[n];
1182         return 0;
1183 }
1184
1185 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1186         .func           = get_func_arg,
1187         .ret_type       = RET_INTEGER,
1188         .arg1_type      = ARG_PTR_TO_CTX,
1189         .arg2_type      = ARG_ANYTHING,
1190         .arg3_type      = ARG_PTR_TO_LONG,
1191 };
1192
1193 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1194 {
1195         /* This helper call is inlined by verifier. */
1196         u64 nr_args = ((u64 *)ctx)[-1];
1197
1198         *value = ((u64 *)ctx)[nr_args];
1199         return 0;
1200 }
1201
1202 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1203         .func           = get_func_ret,
1204         .ret_type       = RET_INTEGER,
1205         .arg1_type      = ARG_PTR_TO_CTX,
1206         .arg2_type      = ARG_PTR_TO_LONG,
1207 };
1208
1209 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1210 {
1211         /* This helper call is inlined by verifier. */
1212         return ((u64 *)ctx)[-1];
1213 }
1214
1215 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1216         .func           = get_func_arg_cnt,
1217         .ret_type       = RET_INTEGER,
1218         .arg1_type      = ARG_PTR_TO_CTX,
1219 };
1220
1221 #ifdef CONFIG_KEYS
1222 __diag_push();
1223 __diag_ignore_all("-Wmissing-prototypes",
1224                   "kfuncs which will be used in BPF programs");
1225
1226 /**
1227  * bpf_lookup_user_key - lookup a key by its serial
1228  * @serial: key handle serial number
1229  * @flags: lookup-specific flags
1230  *
1231  * Search a key with a given *serial* and the provided *flags*.
1232  * If found, increment the reference count of the key by one, and
1233  * return it in the bpf_key structure.
1234  *
1235  * The bpf_key structure must be passed to bpf_key_put() when done
1236  * with it, so that the key reference count is decremented and the
1237  * bpf_key structure is freed.
1238  *
1239  * Permission checks are deferred to the time the key is used by
1240  * one of the available key-specific kfuncs.
1241  *
1242  * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1243  * special keyring (e.g. session keyring), if it doesn't yet exist.
1244  * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1245  * for the key construction, and to retrieve uninstantiated keys (keys
1246  * without data attached to them).
1247  *
1248  * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1249  *         NULL pointer otherwise.
1250  */
1251 struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1252 {
1253         key_ref_t key_ref;
1254         struct bpf_key *bkey;
1255
1256         if (flags & ~KEY_LOOKUP_ALL)
1257                 return NULL;
1258
1259         /*
1260          * Permission check is deferred until the key is used, as the
1261          * intent of the caller is unknown here.
1262          */
1263         key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1264         if (IS_ERR(key_ref))
1265                 return NULL;
1266
1267         bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1268         if (!bkey) {
1269                 key_put(key_ref_to_ptr(key_ref));
1270                 return NULL;
1271         }
1272
1273         bkey->key = key_ref_to_ptr(key_ref);
1274         bkey->has_ref = true;
1275
1276         return bkey;
1277 }
1278
1279 /**
1280  * bpf_lookup_system_key - lookup a key by a system-defined ID
1281  * @id: key ID
1282  *
1283  * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1284  * The key pointer is marked as invalid, to prevent bpf_key_put() from
1285  * attempting to decrement the key reference count on that pointer. The key
1286  * pointer set in such way is currently understood only by
1287  * verify_pkcs7_signature().
1288  *
1289  * Set *id* to one of the values defined in include/linux/verification.h:
1290  * 0 for the primary keyring (immutable keyring of system keys);
1291  * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1292  * (where keys can be added only if they are vouched for by existing keys
1293  * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1294  * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1295  * kerned image and, possibly, the initramfs signature).
1296  *
1297  * Return: a bpf_key pointer with an invalid key pointer set from the
1298  *         pre-determined ID on success, a NULL pointer otherwise
1299  */
1300 struct bpf_key *bpf_lookup_system_key(u64 id)
1301 {
1302         struct bpf_key *bkey;
1303
1304         if (system_keyring_id_check(id) < 0)
1305                 return NULL;
1306
1307         bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1308         if (!bkey)
1309                 return NULL;
1310
1311         bkey->key = (struct key *)(unsigned long)id;
1312         bkey->has_ref = false;
1313
1314         return bkey;
1315 }
1316
1317 /**
1318  * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1319  * @bkey: bpf_key structure
1320  *
1321  * Decrement the reference count of the key inside *bkey*, if the pointer
1322  * is valid, and free *bkey*.
1323  */
1324 void bpf_key_put(struct bpf_key *bkey)
1325 {
1326         if (bkey->has_ref)
1327                 key_put(bkey->key);
1328
1329         kfree(bkey);
1330 }
1331
1332 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1333 /**
1334  * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1335  * @data_ptr: data to verify
1336  * @sig_ptr: signature of the data
1337  * @trusted_keyring: keyring with keys trusted for signature verification
1338  *
1339  * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1340  * with keys in a keyring referenced by *trusted_keyring*.
1341  *
1342  * Return: 0 on success, a negative value on error.
1343  */
1344 int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr,
1345                                struct bpf_dynptr_kern *sig_ptr,
1346                                struct bpf_key *trusted_keyring)
1347 {
1348         int ret;
1349
1350         if (trusted_keyring->has_ref) {
1351                 /*
1352                  * Do the permission check deferred in bpf_lookup_user_key().
1353                  * See bpf_lookup_user_key() for more details.
1354                  *
1355                  * A call to key_task_permission() here would be redundant, as
1356                  * it is already done by keyring_search() called by
1357                  * find_asymmetric_key().
1358                  */
1359                 ret = key_validate(trusted_keyring->key);
1360                 if (ret < 0)
1361                         return ret;
1362         }
1363
1364         return verify_pkcs7_signature(data_ptr->data,
1365                                       bpf_dynptr_get_size(data_ptr),
1366                                       sig_ptr->data,
1367                                       bpf_dynptr_get_size(sig_ptr),
1368                                       trusted_keyring->key,
1369                                       VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1370                                       NULL);
1371 }
1372 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1373
1374 __diag_pop();
1375
1376 BTF_SET8_START(key_sig_kfunc_set)
1377 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1378 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1379 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1380 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1381 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1382 #endif
1383 BTF_SET8_END(key_sig_kfunc_set)
1384
1385 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1386         .owner = THIS_MODULE,
1387         .set = &key_sig_kfunc_set,
1388 };
1389
1390 static int __init bpf_key_sig_kfuncs_init(void)
1391 {
1392         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1393                                          &bpf_key_sig_kfunc_set);
1394 }
1395
1396 late_initcall(bpf_key_sig_kfuncs_init);
1397 #endif /* CONFIG_KEYS */
1398
1399 static const struct bpf_func_proto *
1400 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1401 {
1402         switch (func_id) {
1403         case BPF_FUNC_map_lookup_elem:
1404                 return &bpf_map_lookup_elem_proto;
1405         case BPF_FUNC_map_update_elem:
1406                 return &bpf_map_update_elem_proto;
1407         case BPF_FUNC_map_delete_elem:
1408                 return &bpf_map_delete_elem_proto;
1409         case BPF_FUNC_map_push_elem:
1410                 return &bpf_map_push_elem_proto;
1411         case BPF_FUNC_map_pop_elem:
1412                 return &bpf_map_pop_elem_proto;
1413         case BPF_FUNC_map_peek_elem:
1414                 return &bpf_map_peek_elem_proto;
1415         case BPF_FUNC_map_lookup_percpu_elem:
1416                 return &bpf_map_lookup_percpu_elem_proto;
1417         case BPF_FUNC_ktime_get_ns:
1418                 return &bpf_ktime_get_ns_proto;
1419         case BPF_FUNC_ktime_get_boot_ns:
1420                 return &bpf_ktime_get_boot_ns_proto;
1421         case BPF_FUNC_tail_call:
1422                 return &bpf_tail_call_proto;
1423         case BPF_FUNC_get_current_pid_tgid:
1424                 return &bpf_get_current_pid_tgid_proto;
1425         case BPF_FUNC_get_current_task:
1426                 return &bpf_get_current_task_proto;
1427         case BPF_FUNC_get_current_task_btf:
1428                 return &bpf_get_current_task_btf_proto;
1429         case BPF_FUNC_task_pt_regs:
1430                 return &bpf_task_pt_regs_proto;
1431         case BPF_FUNC_get_current_uid_gid:
1432                 return &bpf_get_current_uid_gid_proto;
1433         case BPF_FUNC_get_current_comm:
1434                 return &bpf_get_current_comm_proto;
1435         case BPF_FUNC_trace_printk:
1436                 return bpf_get_trace_printk_proto();
1437         case BPF_FUNC_get_smp_processor_id:
1438                 return &bpf_get_smp_processor_id_proto;
1439         case BPF_FUNC_get_numa_node_id:
1440                 return &bpf_get_numa_node_id_proto;
1441         case BPF_FUNC_perf_event_read:
1442                 return &bpf_perf_event_read_proto;
1443         case BPF_FUNC_current_task_under_cgroup:
1444                 return &bpf_current_task_under_cgroup_proto;
1445         case BPF_FUNC_get_prandom_u32:
1446                 return &bpf_get_prandom_u32_proto;
1447         case BPF_FUNC_probe_write_user:
1448                 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1449                        NULL : bpf_get_probe_write_proto();
1450         case BPF_FUNC_probe_read_user:
1451                 return &bpf_probe_read_user_proto;
1452         case BPF_FUNC_probe_read_kernel:
1453                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1454                        NULL : &bpf_probe_read_kernel_proto;
1455         case BPF_FUNC_probe_read_user_str:
1456                 return &bpf_probe_read_user_str_proto;
1457         case BPF_FUNC_probe_read_kernel_str:
1458                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1459                        NULL : &bpf_probe_read_kernel_str_proto;
1460 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1461         case BPF_FUNC_probe_read:
1462                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1463                        NULL : &bpf_probe_read_compat_proto;
1464         case BPF_FUNC_probe_read_str:
1465                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1466                        NULL : &bpf_probe_read_compat_str_proto;
1467 #endif
1468 #ifdef CONFIG_CGROUPS
1469         case BPF_FUNC_get_current_cgroup_id:
1470                 return &bpf_get_current_cgroup_id_proto;
1471         case BPF_FUNC_get_current_ancestor_cgroup_id:
1472                 return &bpf_get_current_ancestor_cgroup_id_proto;
1473 #endif
1474         case BPF_FUNC_send_signal:
1475                 return &bpf_send_signal_proto;
1476         case BPF_FUNC_send_signal_thread:
1477                 return &bpf_send_signal_thread_proto;
1478         case BPF_FUNC_perf_event_read_value:
1479                 return &bpf_perf_event_read_value_proto;
1480         case BPF_FUNC_get_ns_current_pid_tgid:
1481                 return &bpf_get_ns_current_pid_tgid_proto;
1482         case BPF_FUNC_ringbuf_output:
1483                 return &bpf_ringbuf_output_proto;
1484         case BPF_FUNC_ringbuf_reserve:
1485                 return &bpf_ringbuf_reserve_proto;
1486         case BPF_FUNC_ringbuf_submit:
1487                 return &bpf_ringbuf_submit_proto;
1488         case BPF_FUNC_ringbuf_discard:
1489                 return &bpf_ringbuf_discard_proto;
1490         case BPF_FUNC_ringbuf_query:
1491                 return &bpf_ringbuf_query_proto;
1492         case BPF_FUNC_jiffies64:
1493                 return &bpf_jiffies64_proto;
1494         case BPF_FUNC_get_task_stack:
1495                 return &bpf_get_task_stack_proto;
1496         case BPF_FUNC_copy_from_user:
1497                 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1498         case BPF_FUNC_copy_from_user_task:
1499                 return prog->aux->sleepable ? &bpf_copy_from_user_task_proto : NULL;
1500         case BPF_FUNC_snprintf_btf:
1501                 return &bpf_snprintf_btf_proto;
1502         case BPF_FUNC_per_cpu_ptr:
1503                 return &bpf_per_cpu_ptr_proto;
1504         case BPF_FUNC_this_cpu_ptr:
1505                 return &bpf_this_cpu_ptr_proto;
1506         case BPF_FUNC_task_storage_get:
1507                 return &bpf_task_storage_get_proto;
1508         case BPF_FUNC_task_storage_delete:
1509                 return &bpf_task_storage_delete_proto;
1510         case BPF_FUNC_for_each_map_elem:
1511                 return &bpf_for_each_map_elem_proto;
1512         case BPF_FUNC_snprintf:
1513                 return &bpf_snprintf_proto;
1514         case BPF_FUNC_get_func_ip:
1515                 return &bpf_get_func_ip_proto_tracing;
1516         case BPF_FUNC_get_branch_snapshot:
1517                 return &bpf_get_branch_snapshot_proto;
1518         case BPF_FUNC_find_vma:
1519                 return &bpf_find_vma_proto;
1520         case BPF_FUNC_trace_vprintk:
1521                 return bpf_get_trace_vprintk_proto();
1522         default:
1523                 return bpf_base_func_proto(func_id);
1524         }
1525 }
1526
1527 static const struct bpf_func_proto *
1528 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1529 {
1530         switch (func_id) {
1531         case BPF_FUNC_perf_event_output:
1532                 return &bpf_perf_event_output_proto;
1533         case BPF_FUNC_get_stackid:
1534                 return &bpf_get_stackid_proto;
1535         case BPF_FUNC_get_stack:
1536                 return &bpf_get_stack_proto;
1537 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1538         case BPF_FUNC_override_return:
1539                 return &bpf_override_return_proto;
1540 #endif
1541         case BPF_FUNC_get_func_ip:
1542                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1543                         &bpf_get_func_ip_proto_kprobe_multi :
1544                         &bpf_get_func_ip_proto_kprobe;
1545         case BPF_FUNC_get_attach_cookie:
1546                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1547                         &bpf_get_attach_cookie_proto_kmulti :
1548                         &bpf_get_attach_cookie_proto_trace;
1549         default:
1550                 return bpf_tracing_func_proto(func_id, prog);
1551         }
1552 }
1553
1554 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1555 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1556                                         const struct bpf_prog *prog,
1557                                         struct bpf_insn_access_aux *info)
1558 {
1559         if (off < 0 || off >= sizeof(struct pt_regs))
1560                 return false;
1561         if (type != BPF_READ)
1562                 return false;
1563         if (off % size != 0)
1564                 return false;
1565         /*
1566          * Assertion for 32 bit to make sure last 8 byte access
1567          * (BPF_DW) to the last 4 byte member is disallowed.
1568          */
1569         if (off + size > sizeof(struct pt_regs))
1570                 return false;
1571
1572         return true;
1573 }
1574
1575 const struct bpf_verifier_ops kprobe_verifier_ops = {
1576         .get_func_proto  = kprobe_prog_func_proto,
1577         .is_valid_access = kprobe_prog_is_valid_access,
1578 };
1579
1580 const struct bpf_prog_ops kprobe_prog_ops = {
1581 };
1582
1583 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1584            u64, flags, void *, data, u64, size)
1585 {
1586         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1587
1588         /*
1589          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1590          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1591          * from there and call the same bpf_perf_event_output() helper inline.
1592          */
1593         return ____bpf_perf_event_output(regs, map, flags, data, size);
1594 }
1595
1596 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1597         .func           = bpf_perf_event_output_tp,
1598         .gpl_only       = true,
1599         .ret_type       = RET_INTEGER,
1600         .arg1_type      = ARG_PTR_TO_CTX,
1601         .arg2_type      = ARG_CONST_MAP_PTR,
1602         .arg3_type      = ARG_ANYTHING,
1603         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1604         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1605 };
1606
1607 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1608            u64, flags)
1609 {
1610         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1611
1612         /*
1613          * Same comment as in bpf_perf_event_output_tp(), only that this time
1614          * the other helper's function body cannot be inlined due to being
1615          * external, thus we need to call raw helper function.
1616          */
1617         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1618                                flags, 0, 0);
1619 }
1620
1621 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1622         .func           = bpf_get_stackid_tp,
1623         .gpl_only       = true,
1624         .ret_type       = RET_INTEGER,
1625         .arg1_type      = ARG_PTR_TO_CTX,
1626         .arg2_type      = ARG_CONST_MAP_PTR,
1627         .arg3_type      = ARG_ANYTHING,
1628 };
1629
1630 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1631            u64, flags)
1632 {
1633         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1634
1635         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1636                              (unsigned long) size, flags, 0);
1637 }
1638
1639 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1640         .func           = bpf_get_stack_tp,
1641         .gpl_only       = true,
1642         .ret_type       = RET_INTEGER,
1643         .arg1_type      = ARG_PTR_TO_CTX,
1644         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1645         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1646         .arg4_type      = ARG_ANYTHING,
1647 };
1648
1649 static const struct bpf_func_proto *
1650 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1651 {
1652         switch (func_id) {
1653         case BPF_FUNC_perf_event_output:
1654                 return &bpf_perf_event_output_proto_tp;
1655         case BPF_FUNC_get_stackid:
1656                 return &bpf_get_stackid_proto_tp;
1657         case BPF_FUNC_get_stack:
1658                 return &bpf_get_stack_proto_tp;
1659         case BPF_FUNC_get_attach_cookie:
1660                 return &bpf_get_attach_cookie_proto_trace;
1661         default:
1662                 return bpf_tracing_func_proto(func_id, prog);
1663         }
1664 }
1665
1666 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1667                                     const struct bpf_prog *prog,
1668                                     struct bpf_insn_access_aux *info)
1669 {
1670         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1671                 return false;
1672         if (type != BPF_READ)
1673                 return false;
1674         if (off % size != 0)
1675                 return false;
1676
1677         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1678         return true;
1679 }
1680
1681 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1682         .get_func_proto  = tp_prog_func_proto,
1683         .is_valid_access = tp_prog_is_valid_access,
1684 };
1685
1686 const struct bpf_prog_ops tracepoint_prog_ops = {
1687 };
1688
1689 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1690            struct bpf_perf_event_value *, buf, u32, size)
1691 {
1692         int err = -EINVAL;
1693
1694         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1695                 goto clear;
1696         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1697                                     &buf->running);
1698         if (unlikely(err))
1699                 goto clear;
1700         return 0;
1701 clear:
1702         memset(buf, 0, size);
1703         return err;
1704 }
1705
1706 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1707          .func           = bpf_perf_prog_read_value,
1708          .gpl_only       = true,
1709          .ret_type       = RET_INTEGER,
1710          .arg1_type      = ARG_PTR_TO_CTX,
1711          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1712          .arg3_type      = ARG_CONST_SIZE,
1713 };
1714
1715 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1716            void *, buf, u32, size, u64, flags)
1717 {
1718         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1719         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1720         u32 to_copy;
1721
1722         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1723                 return -EINVAL;
1724
1725         if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1726                 return -ENOENT;
1727
1728         if (unlikely(!br_stack))
1729                 return -ENOENT;
1730
1731         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1732                 return br_stack->nr * br_entry_size;
1733
1734         if (!buf || (size % br_entry_size != 0))
1735                 return -EINVAL;
1736
1737         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1738         memcpy(buf, br_stack->entries, to_copy);
1739
1740         return to_copy;
1741 }
1742
1743 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1744         .func           = bpf_read_branch_records,
1745         .gpl_only       = true,
1746         .ret_type       = RET_INTEGER,
1747         .arg1_type      = ARG_PTR_TO_CTX,
1748         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1749         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1750         .arg4_type      = ARG_ANYTHING,
1751 };
1752
1753 static const struct bpf_func_proto *
1754 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1755 {
1756         switch (func_id) {
1757         case BPF_FUNC_perf_event_output:
1758                 return &bpf_perf_event_output_proto_tp;
1759         case BPF_FUNC_get_stackid:
1760                 return &bpf_get_stackid_proto_pe;
1761         case BPF_FUNC_get_stack:
1762                 return &bpf_get_stack_proto_pe;
1763         case BPF_FUNC_perf_prog_read_value:
1764                 return &bpf_perf_prog_read_value_proto;
1765         case BPF_FUNC_read_branch_records:
1766                 return &bpf_read_branch_records_proto;
1767         case BPF_FUNC_get_attach_cookie:
1768                 return &bpf_get_attach_cookie_proto_pe;
1769         default:
1770                 return bpf_tracing_func_proto(func_id, prog);
1771         }
1772 }
1773
1774 /*
1775  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1776  * to avoid potential recursive reuse issue when/if tracepoints are added
1777  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1778  *
1779  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1780  * in normal, irq, and nmi context.
1781  */
1782 struct bpf_raw_tp_regs {
1783         struct pt_regs regs[3];
1784 };
1785 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1786 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1787 static struct pt_regs *get_bpf_raw_tp_regs(void)
1788 {
1789         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1790         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1791
1792         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1793                 this_cpu_dec(bpf_raw_tp_nest_level);
1794                 return ERR_PTR(-EBUSY);
1795         }
1796
1797         return &tp_regs->regs[nest_level - 1];
1798 }
1799
1800 static void put_bpf_raw_tp_regs(void)
1801 {
1802         this_cpu_dec(bpf_raw_tp_nest_level);
1803 }
1804
1805 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1806            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1807 {
1808         struct pt_regs *regs = get_bpf_raw_tp_regs();
1809         int ret;
1810
1811         if (IS_ERR(regs))
1812                 return PTR_ERR(regs);
1813
1814         perf_fetch_caller_regs(regs);
1815         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1816
1817         put_bpf_raw_tp_regs();
1818         return ret;
1819 }
1820
1821 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1822         .func           = bpf_perf_event_output_raw_tp,
1823         .gpl_only       = true,
1824         .ret_type       = RET_INTEGER,
1825         .arg1_type      = ARG_PTR_TO_CTX,
1826         .arg2_type      = ARG_CONST_MAP_PTR,
1827         .arg3_type      = ARG_ANYTHING,
1828         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1829         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1830 };
1831
1832 extern const struct bpf_func_proto bpf_skb_output_proto;
1833 extern const struct bpf_func_proto bpf_xdp_output_proto;
1834 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1835
1836 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1837            struct bpf_map *, map, u64, flags)
1838 {
1839         struct pt_regs *regs = get_bpf_raw_tp_regs();
1840         int ret;
1841
1842         if (IS_ERR(regs))
1843                 return PTR_ERR(regs);
1844
1845         perf_fetch_caller_regs(regs);
1846         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1847         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1848                               flags, 0, 0);
1849         put_bpf_raw_tp_regs();
1850         return ret;
1851 }
1852
1853 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1854         .func           = bpf_get_stackid_raw_tp,
1855         .gpl_only       = true,
1856         .ret_type       = RET_INTEGER,
1857         .arg1_type      = ARG_PTR_TO_CTX,
1858         .arg2_type      = ARG_CONST_MAP_PTR,
1859         .arg3_type      = ARG_ANYTHING,
1860 };
1861
1862 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1863            void *, buf, u32, size, u64, flags)
1864 {
1865         struct pt_regs *regs = get_bpf_raw_tp_regs();
1866         int ret;
1867
1868         if (IS_ERR(regs))
1869                 return PTR_ERR(regs);
1870
1871         perf_fetch_caller_regs(regs);
1872         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1873                             (unsigned long) size, flags, 0);
1874         put_bpf_raw_tp_regs();
1875         return ret;
1876 }
1877
1878 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1879         .func           = bpf_get_stack_raw_tp,
1880         .gpl_only       = true,
1881         .ret_type       = RET_INTEGER,
1882         .arg1_type      = ARG_PTR_TO_CTX,
1883         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1884         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1885         .arg4_type      = ARG_ANYTHING,
1886 };
1887
1888 static const struct bpf_func_proto *
1889 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1890 {
1891         switch (func_id) {
1892         case BPF_FUNC_perf_event_output:
1893                 return &bpf_perf_event_output_proto_raw_tp;
1894         case BPF_FUNC_get_stackid:
1895                 return &bpf_get_stackid_proto_raw_tp;
1896         case BPF_FUNC_get_stack:
1897                 return &bpf_get_stack_proto_raw_tp;
1898         default:
1899                 return bpf_tracing_func_proto(func_id, prog);
1900         }
1901 }
1902
1903 const struct bpf_func_proto *
1904 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1905 {
1906         const struct bpf_func_proto *fn;
1907
1908         switch (func_id) {
1909 #ifdef CONFIG_NET
1910         case BPF_FUNC_skb_output:
1911                 return &bpf_skb_output_proto;
1912         case BPF_FUNC_xdp_output:
1913                 return &bpf_xdp_output_proto;
1914         case BPF_FUNC_skc_to_tcp6_sock:
1915                 return &bpf_skc_to_tcp6_sock_proto;
1916         case BPF_FUNC_skc_to_tcp_sock:
1917                 return &bpf_skc_to_tcp_sock_proto;
1918         case BPF_FUNC_skc_to_tcp_timewait_sock:
1919                 return &bpf_skc_to_tcp_timewait_sock_proto;
1920         case BPF_FUNC_skc_to_tcp_request_sock:
1921                 return &bpf_skc_to_tcp_request_sock_proto;
1922         case BPF_FUNC_skc_to_udp6_sock:
1923                 return &bpf_skc_to_udp6_sock_proto;
1924         case BPF_FUNC_skc_to_unix_sock:
1925                 return &bpf_skc_to_unix_sock_proto;
1926         case BPF_FUNC_skc_to_mptcp_sock:
1927                 return &bpf_skc_to_mptcp_sock_proto;
1928         case BPF_FUNC_sk_storage_get:
1929                 return &bpf_sk_storage_get_tracing_proto;
1930         case BPF_FUNC_sk_storage_delete:
1931                 return &bpf_sk_storage_delete_tracing_proto;
1932         case BPF_FUNC_sock_from_file:
1933                 return &bpf_sock_from_file_proto;
1934         case BPF_FUNC_get_socket_cookie:
1935                 return &bpf_get_socket_ptr_cookie_proto;
1936         case BPF_FUNC_xdp_get_buff_len:
1937                 return &bpf_xdp_get_buff_len_trace_proto;
1938 #endif
1939         case BPF_FUNC_seq_printf:
1940                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1941                        &bpf_seq_printf_proto :
1942                        NULL;
1943         case BPF_FUNC_seq_write:
1944                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1945                        &bpf_seq_write_proto :
1946                        NULL;
1947         case BPF_FUNC_seq_printf_btf:
1948                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1949                        &bpf_seq_printf_btf_proto :
1950                        NULL;
1951         case BPF_FUNC_d_path:
1952                 return &bpf_d_path_proto;
1953         case BPF_FUNC_get_func_arg:
1954                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1955         case BPF_FUNC_get_func_ret:
1956                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1957         case BPF_FUNC_get_func_arg_cnt:
1958                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1959         case BPF_FUNC_get_attach_cookie:
1960                 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
1961         default:
1962                 fn = raw_tp_prog_func_proto(func_id, prog);
1963                 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1964                         fn = bpf_iter_get_func_proto(func_id, prog);
1965                 return fn;
1966         }
1967 }
1968
1969 static bool raw_tp_prog_is_valid_access(int off, int size,
1970                                         enum bpf_access_type type,
1971                                         const struct bpf_prog *prog,
1972                                         struct bpf_insn_access_aux *info)
1973 {
1974         return bpf_tracing_ctx_access(off, size, type);
1975 }
1976
1977 static bool tracing_prog_is_valid_access(int off, int size,
1978                                          enum bpf_access_type type,
1979                                          const struct bpf_prog *prog,
1980                                          struct bpf_insn_access_aux *info)
1981 {
1982         return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1983 }
1984
1985 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1986                                      const union bpf_attr *kattr,
1987                                      union bpf_attr __user *uattr)
1988 {
1989         return -ENOTSUPP;
1990 }
1991
1992 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1993         .get_func_proto  = raw_tp_prog_func_proto,
1994         .is_valid_access = raw_tp_prog_is_valid_access,
1995 };
1996
1997 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1998 #ifdef CONFIG_NET
1999         .test_run = bpf_prog_test_run_raw_tp,
2000 #endif
2001 };
2002
2003 const struct bpf_verifier_ops tracing_verifier_ops = {
2004         .get_func_proto  = tracing_prog_func_proto,
2005         .is_valid_access = tracing_prog_is_valid_access,
2006 };
2007
2008 const struct bpf_prog_ops tracing_prog_ops = {
2009         .test_run = bpf_prog_test_run_tracing,
2010 };
2011
2012 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
2013                                                  enum bpf_access_type type,
2014                                                  const struct bpf_prog *prog,
2015                                                  struct bpf_insn_access_aux *info)
2016 {
2017         if (off == 0) {
2018                 if (size != sizeof(u64) || type != BPF_READ)
2019                         return false;
2020                 info->reg_type = PTR_TO_TP_BUFFER;
2021         }
2022         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2023 }
2024
2025 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2026         .get_func_proto  = raw_tp_prog_func_proto,
2027         .is_valid_access = raw_tp_writable_prog_is_valid_access,
2028 };
2029
2030 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2031 };
2032
2033 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2034                                     const struct bpf_prog *prog,
2035                                     struct bpf_insn_access_aux *info)
2036 {
2037         const int size_u64 = sizeof(u64);
2038
2039         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2040                 return false;
2041         if (type != BPF_READ)
2042                 return false;
2043         if (off % size != 0) {
2044                 if (sizeof(unsigned long) != 4)
2045                         return false;
2046                 if (size != 8)
2047                         return false;
2048                 if (off % size != 4)
2049                         return false;
2050         }
2051
2052         switch (off) {
2053         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2054                 bpf_ctx_record_field_size(info, size_u64);
2055                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2056                         return false;
2057                 break;
2058         case bpf_ctx_range(struct bpf_perf_event_data, addr):
2059                 bpf_ctx_record_field_size(info, size_u64);
2060                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2061                         return false;
2062                 break;
2063         default:
2064                 if (size != sizeof(long))
2065                         return false;
2066         }
2067
2068         return true;
2069 }
2070
2071 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2072                                       const struct bpf_insn *si,
2073                                       struct bpf_insn *insn_buf,
2074                                       struct bpf_prog *prog, u32 *target_size)
2075 {
2076         struct bpf_insn *insn = insn_buf;
2077
2078         switch (si->off) {
2079         case offsetof(struct bpf_perf_event_data, sample_period):
2080                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2081                                                        data), si->dst_reg, si->src_reg,
2082                                       offsetof(struct bpf_perf_event_data_kern, data));
2083                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2084                                       bpf_target_off(struct perf_sample_data, period, 8,
2085                                                      target_size));
2086                 break;
2087         case offsetof(struct bpf_perf_event_data, addr):
2088                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2089                                                        data), si->dst_reg, si->src_reg,
2090                                       offsetof(struct bpf_perf_event_data_kern, data));
2091                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2092                                       bpf_target_off(struct perf_sample_data, addr, 8,
2093                                                      target_size));
2094                 break;
2095         default:
2096                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2097                                                        regs), si->dst_reg, si->src_reg,
2098                                       offsetof(struct bpf_perf_event_data_kern, regs));
2099                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2100                                       si->off);
2101                 break;
2102         }
2103
2104         return insn - insn_buf;
2105 }
2106
2107 const struct bpf_verifier_ops perf_event_verifier_ops = {
2108         .get_func_proto         = pe_prog_func_proto,
2109         .is_valid_access        = pe_prog_is_valid_access,
2110         .convert_ctx_access     = pe_prog_convert_ctx_access,
2111 };
2112
2113 const struct bpf_prog_ops perf_event_prog_ops = {
2114 };
2115
2116 static DEFINE_MUTEX(bpf_event_mutex);
2117
2118 #define BPF_TRACE_MAX_PROGS 64
2119
2120 int perf_event_attach_bpf_prog(struct perf_event *event,
2121                                struct bpf_prog *prog,
2122                                u64 bpf_cookie)
2123 {
2124         struct bpf_prog_array *old_array;
2125         struct bpf_prog_array *new_array;
2126         int ret = -EEXIST;
2127
2128         /*
2129          * Kprobe override only works if they are on the function entry,
2130          * and only if they are on the opt-in list.
2131          */
2132         if (prog->kprobe_override &&
2133             (!trace_kprobe_on_func_entry(event->tp_event) ||
2134              !trace_kprobe_error_injectable(event->tp_event)))
2135                 return -EINVAL;
2136
2137         mutex_lock(&bpf_event_mutex);
2138
2139         if (event->prog)
2140                 goto unlock;
2141
2142         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2143         if (old_array &&
2144             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2145                 ret = -E2BIG;
2146                 goto unlock;
2147         }
2148
2149         ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2150         if (ret < 0)
2151                 goto unlock;
2152
2153         /* set the new array to event->tp_event and set event->prog */
2154         event->prog = prog;
2155         event->bpf_cookie = bpf_cookie;
2156         rcu_assign_pointer(event->tp_event->prog_array, new_array);
2157         bpf_prog_array_free_sleepable(old_array);
2158
2159 unlock:
2160         mutex_unlock(&bpf_event_mutex);
2161         return ret;
2162 }
2163
2164 void perf_event_detach_bpf_prog(struct perf_event *event)
2165 {
2166         struct bpf_prog_array *old_array;
2167         struct bpf_prog_array *new_array;
2168         int ret;
2169
2170         mutex_lock(&bpf_event_mutex);
2171
2172         if (!event->prog)
2173                 goto unlock;
2174
2175         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2176         ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2177         if (ret == -ENOENT)
2178                 goto unlock;
2179         if (ret < 0) {
2180                 bpf_prog_array_delete_safe(old_array, event->prog);
2181         } else {
2182                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2183                 bpf_prog_array_free_sleepable(old_array);
2184         }
2185
2186         bpf_prog_put(event->prog);
2187         event->prog = NULL;
2188
2189 unlock:
2190         mutex_unlock(&bpf_event_mutex);
2191 }
2192
2193 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2194 {
2195         struct perf_event_query_bpf __user *uquery = info;
2196         struct perf_event_query_bpf query = {};
2197         struct bpf_prog_array *progs;
2198         u32 *ids, prog_cnt, ids_len;
2199         int ret;
2200
2201         if (!perfmon_capable())
2202                 return -EPERM;
2203         if (event->attr.type != PERF_TYPE_TRACEPOINT)
2204                 return -EINVAL;
2205         if (copy_from_user(&query, uquery, sizeof(query)))
2206                 return -EFAULT;
2207
2208         ids_len = query.ids_len;
2209         if (ids_len > BPF_TRACE_MAX_PROGS)
2210                 return -E2BIG;
2211         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2212         if (!ids)
2213                 return -ENOMEM;
2214         /*
2215          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2216          * is required when user only wants to check for uquery->prog_cnt.
2217          * There is no need to check for it since the case is handled
2218          * gracefully in bpf_prog_array_copy_info.
2219          */
2220
2221         mutex_lock(&bpf_event_mutex);
2222         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2223         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2224         mutex_unlock(&bpf_event_mutex);
2225
2226         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2227             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2228                 ret = -EFAULT;
2229
2230         kfree(ids);
2231         return ret;
2232 }
2233
2234 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2235 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2236
2237 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2238 {
2239         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2240
2241         for (; btp < __stop__bpf_raw_tp; btp++) {
2242                 if (!strcmp(btp->tp->name, name))
2243                         return btp;
2244         }
2245
2246         return bpf_get_raw_tracepoint_module(name);
2247 }
2248
2249 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2250 {
2251         struct module *mod;
2252
2253         preempt_disable();
2254         mod = __module_address((unsigned long)btp);
2255         module_put(mod);
2256         preempt_enable();
2257 }
2258
2259 static __always_inline
2260 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2261 {
2262         cant_sleep();
2263         if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2264                 bpf_prog_inc_misses_counter(prog);
2265                 goto out;
2266         }
2267         rcu_read_lock();
2268         (void) bpf_prog_run(prog, args);
2269         rcu_read_unlock();
2270 out:
2271         this_cpu_dec(*(prog->active));
2272 }
2273
2274 #define UNPACK(...)                     __VA_ARGS__
2275 #define REPEAT_1(FN, DL, X, ...)        FN(X)
2276 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2277 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2278 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2279 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2280 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2281 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2282 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2283 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2284 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2285 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2286 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2287 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
2288
2289 #define SARG(X)         u64 arg##X
2290 #define COPY(X)         args[X] = arg##X
2291
2292 #define __DL_COM        (,)
2293 #define __DL_SEM        (;)
2294
2295 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2296
2297 #define BPF_TRACE_DEFN_x(x)                                             \
2298         void bpf_trace_run##x(struct bpf_prog *prog,                    \
2299                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
2300         {                                                               \
2301                 u64 args[x];                                            \
2302                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
2303                 __bpf_trace_run(prog, args);                            \
2304         }                                                               \
2305         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2306 BPF_TRACE_DEFN_x(1);
2307 BPF_TRACE_DEFN_x(2);
2308 BPF_TRACE_DEFN_x(3);
2309 BPF_TRACE_DEFN_x(4);
2310 BPF_TRACE_DEFN_x(5);
2311 BPF_TRACE_DEFN_x(6);
2312 BPF_TRACE_DEFN_x(7);
2313 BPF_TRACE_DEFN_x(8);
2314 BPF_TRACE_DEFN_x(9);
2315 BPF_TRACE_DEFN_x(10);
2316 BPF_TRACE_DEFN_x(11);
2317 BPF_TRACE_DEFN_x(12);
2318
2319 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2320 {
2321         struct tracepoint *tp = btp->tp;
2322
2323         /*
2324          * check that program doesn't access arguments beyond what's
2325          * available in this tracepoint
2326          */
2327         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2328                 return -EINVAL;
2329
2330         if (prog->aux->max_tp_access > btp->writable_size)
2331                 return -EINVAL;
2332
2333         return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2334                                                    prog);
2335 }
2336
2337 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2338 {
2339         return __bpf_probe_register(btp, prog);
2340 }
2341
2342 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2343 {
2344         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2345 }
2346
2347 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2348                             u32 *fd_type, const char **buf,
2349                             u64 *probe_offset, u64 *probe_addr)
2350 {
2351         bool is_tracepoint, is_syscall_tp;
2352         struct bpf_prog *prog;
2353         int flags, err = 0;
2354
2355         prog = event->prog;
2356         if (!prog)
2357                 return -ENOENT;
2358
2359         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2360         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2361                 return -EOPNOTSUPP;
2362
2363         *prog_id = prog->aux->id;
2364         flags = event->tp_event->flags;
2365         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2366         is_syscall_tp = is_syscall_trace_event(event->tp_event);
2367
2368         if (is_tracepoint || is_syscall_tp) {
2369                 *buf = is_tracepoint ? event->tp_event->tp->name
2370                                      : event->tp_event->name;
2371                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2372                 *probe_offset = 0x0;
2373                 *probe_addr = 0x0;
2374         } else {
2375                 /* kprobe/uprobe */
2376                 err = -EOPNOTSUPP;
2377 #ifdef CONFIG_KPROBE_EVENTS
2378                 if (flags & TRACE_EVENT_FL_KPROBE)
2379                         err = bpf_get_kprobe_info(event, fd_type, buf,
2380                                                   probe_offset, probe_addr,
2381                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2382 #endif
2383 #ifdef CONFIG_UPROBE_EVENTS
2384                 if (flags & TRACE_EVENT_FL_UPROBE)
2385                         err = bpf_get_uprobe_info(event, fd_type, buf,
2386                                                   probe_offset,
2387                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2388 #endif
2389         }
2390
2391         return err;
2392 }
2393
2394 static int __init send_signal_irq_work_init(void)
2395 {
2396         int cpu;
2397         struct send_signal_irq_work *work;
2398
2399         for_each_possible_cpu(cpu) {
2400                 work = per_cpu_ptr(&send_signal_work, cpu);
2401                 init_irq_work(&work->irq_work, do_bpf_send_signal);
2402         }
2403         return 0;
2404 }
2405
2406 subsys_initcall(send_signal_irq_work_init);
2407
2408 #ifdef CONFIG_MODULES
2409 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2410                             void *module)
2411 {
2412         struct bpf_trace_module *btm, *tmp;
2413         struct module *mod = module;
2414         int ret = 0;
2415
2416         if (mod->num_bpf_raw_events == 0 ||
2417             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2418                 goto out;
2419
2420         mutex_lock(&bpf_module_mutex);
2421
2422         switch (op) {
2423         case MODULE_STATE_COMING:
2424                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2425                 if (btm) {
2426                         btm->module = module;
2427                         list_add(&btm->list, &bpf_trace_modules);
2428                 } else {
2429                         ret = -ENOMEM;
2430                 }
2431                 break;
2432         case MODULE_STATE_GOING:
2433                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2434                         if (btm->module == module) {
2435                                 list_del(&btm->list);
2436                                 kfree(btm);
2437                                 break;
2438                         }
2439                 }
2440                 break;
2441         }
2442
2443         mutex_unlock(&bpf_module_mutex);
2444
2445 out:
2446         return notifier_from_errno(ret);
2447 }
2448
2449 static struct notifier_block bpf_module_nb = {
2450         .notifier_call = bpf_event_notify,
2451 };
2452
2453 static int __init bpf_event_init(void)
2454 {
2455         register_module_notifier(&bpf_module_nb);
2456         return 0;
2457 }
2458
2459 fs_initcall(bpf_event_init);
2460 #endif /* CONFIG_MODULES */
2461
2462 #ifdef CONFIG_FPROBE
2463 struct bpf_kprobe_multi_link {
2464         struct bpf_link link;
2465         struct fprobe fp;
2466         unsigned long *addrs;
2467         u64 *cookies;
2468         u32 cnt;
2469 };
2470
2471 struct bpf_kprobe_multi_run_ctx {
2472         struct bpf_run_ctx run_ctx;
2473         struct bpf_kprobe_multi_link *link;
2474         unsigned long entry_ip;
2475 };
2476
2477 struct user_syms {
2478         const char **syms;
2479         char *buf;
2480 };
2481
2482 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2483 {
2484         unsigned long __user usymbol;
2485         const char **syms = NULL;
2486         char *buf = NULL, *p;
2487         int err = -ENOMEM;
2488         unsigned int i;
2489
2490         syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2491         if (!syms)
2492                 goto error;
2493
2494         buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2495         if (!buf)
2496                 goto error;
2497
2498         for (p = buf, i = 0; i < cnt; i++) {
2499                 if (__get_user(usymbol, usyms + i)) {
2500                         err = -EFAULT;
2501                         goto error;
2502                 }
2503                 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2504                 if (err == KSYM_NAME_LEN)
2505                         err = -E2BIG;
2506                 if (err < 0)
2507                         goto error;
2508                 syms[i] = p;
2509                 p += err + 1;
2510         }
2511
2512         us->syms = syms;
2513         us->buf = buf;
2514         return 0;
2515
2516 error:
2517         if (err) {
2518                 kvfree(syms);
2519                 kvfree(buf);
2520         }
2521         return err;
2522 }
2523
2524 static void free_user_syms(struct user_syms *us)
2525 {
2526         kvfree(us->syms);
2527         kvfree(us->buf);
2528 }
2529
2530 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2531 {
2532         struct bpf_kprobe_multi_link *kmulti_link;
2533
2534         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2535         unregister_fprobe(&kmulti_link->fp);
2536 }
2537
2538 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2539 {
2540         struct bpf_kprobe_multi_link *kmulti_link;
2541
2542         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2543         kvfree(kmulti_link->addrs);
2544         kvfree(kmulti_link->cookies);
2545         kfree(kmulti_link);
2546 }
2547
2548 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2549         .release = bpf_kprobe_multi_link_release,
2550         .dealloc = bpf_kprobe_multi_link_dealloc,
2551 };
2552
2553 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2554 {
2555         const struct bpf_kprobe_multi_link *link = priv;
2556         unsigned long *addr_a = a, *addr_b = b;
2557         u64 *cookie_a, *cookie_b;
2558
2559         cookie_a = link->cookies + (addr_a - link->addrs);
2560         cookie_b = link->cookies + (addr_b - link->addrs);
2561
2562         /* swap addr_a/addr_b and cookie_a/cookie_b values */
2563         swap(*addr_a, *addr_b);
2564         swap(*cookie_a, *cookie_b);
2565 }
2566
2567 static int __bpf_kprobe_multi_cookie_cmp(const void *a, const void *b)
2568 {
2569         const unsigned long *addr_a = a, *addr_b = b;
2570
2571         if (*addr_a == *addr_b)
2572                 return 0;
2573         return *addr_a < *addr_b ? -1 : 1;
2574 }
2575
2576 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2577 {
2578         return __bpf_kprobe_multi_cookie_cmp(a, b);
2579 }
2580
2581 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2582 {
2583         struct bpf_kprobe_multi_run_ctx *run_ctx;
2584         struct bpf_kprobe_multi_link *link;
2585         u64 *cookie, entry_ip;
2586         unsigned long *addr;
2587
2588         if (WARN_ON_ONCE(!ctx))
2589                 return 0;
2590         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2591         link = run_ctx->link;
2592         if (!link->cookies)
2593                 return 0;
2594         entry_ip = run_ctx->entry_ip;
2595         addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2596                        __bpf_kprobe_multi_cookie_cmp);
2597         if (!addr)
2598                 return 0;
2599         cookie = link->cookies + (addr - link->addrs);
2600         return *cookie;
2601 }
2602
2603 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2604 {
2605         struct bpf_kprobe_multi_run_ctx *run_ctx;
2606
2607         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2608         return run_ctx->entry_ip;
2609 }
2610
2611 static int
2612 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2613                            unsigned long entry_ip, struct pt_regs *regs)
2614 {
2615         struct bpf_kprobe_multi_run_ctx run_ctx = {
2616                 .link = link,
2617                 .entry_ip = entry_ip,
2618         };
2619         struct bpf_run_ctx *old_run_ctx;
2620         int err;
2621
2622         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2623                 err = 0;
2624                 goto out;
2625         }
2626
2627         migrate_disable();
2628         rcu_read_lock();
2629         old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2630         err = bpf_prog_run(link->link.prog, regs);
2631         bpf_reset_run_ctx(old_run_ctx);
2632         rcu_read_unlock();
2633         migrate_enable();
2634
2635  out:
2636         __this_cpu_dec(bpf_prog_active);
2637         return err;
2638 }
2639
2640 static void
2641 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2642                           struct pt_regs *regs)
2643 {
2644         struct bpf_kprobe_multi_link *link;
2645
2646         link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2647         kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2648 }
2649
2650 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2651 {
2652         const char **str_a = (const char **) a;
2653         const char **str_b = (const char **) b;
2654
2655         return strcmp(*str_a, *str_b);
2656 }
2657
2658 struct multi_symbols_sort {
2659         const char **funcs;
2660         u64 *cookies;
2661 };
2662
2663 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2664 {
2665         const struct multi_symbols_sort *data = priv;
2666         const char **name_a = a, **name_b = b;
2667
2668         swap(*name_a, *name_b);
2669
2670         /* If defined, swap also related cookies. */
2671         if (data->cookies) {
2672                 u64 *cookie_a, *cookie_b;
2673
2674                 cookie_a = data->cookies + (name_a - data->funcs);
2675                 cookie_b = data->cookies + (name_b - data->funcs);
2676                 swap(*cookie_a, *cookie_b);
2677         }
2678 }
2679
2680 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2681 {
2682         struct bpf_kprobe_multi_link *link = NULL;
2683         struct bpf_link_primer link_primer;
2684         void __user *ucookies;
2685         unsigned long *addrs;
2686         u32 flags, cnt, size;
2687         void __user *uaddrs;
2688         u64 *cookies = NULL;
2689         void __user *usyms;
2690         int err;
2691
2692         /* no support for 32bit archs yet */
2693         if (sizeof(u64) != sizeof(void *))
2694                 return -EOPNOTSUPP;
2695
2696         if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2697                 return -EINVAL;
2698
2699         flags = attr->link_create.kprobe_multi.flags;
2700         if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2701                 return -EINVAL;
2702
2703         uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2704         usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2705         if (!!uaddrs == !!usyms)
2706                 return -EINVAL;
2707
2708         cnt = attr->link_create.kprobe_multi.cnt;
2709         if (!cnt)
2710                 return -EINVAL;
2711
2712         size = cnt * sizeof(*addrs);
2713         addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2714         if (!addrs)
2715                 return -ENOMEM;
2716
2717         ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2718         if (ucookies) {
2719                 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2720                 if (!cookies) {
2721                         err = -ENOMEM;
2722                         goto error;
2723                 }
2724                 if (copy_from_user(cookies, ucookies, size)) {
2725                         err = -EFAULT;
2726                         goto error;
2727                 }
2728         }
2729
2730         if (uaddrs) {
2731                 if (copy_from_user(addrs, uaddrs, size)) {
2732                         err = -EFAULT;
2733                         goto error;
2734                 }
2735         } else {
2736                 struct multi_symbols_sort data = {
2737                         .cookies = cookies,
2738                 };
2739                 struct user_syms us;
2740
2741                 err = copy_user_syms(&us, usyms, cnt);
2742                 if (err)
2743                         goto error;
2744
2745                 if (cookies)
2746                         data.funcs = us.syms;
2747
2748                 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2749                        symbols_swap_r, &data);
2750
2751                 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2752                 free_user_syms(&us);
2753                 if (err)
2754                         goto error;
2755         }
2756
2757         link = kzalloc(sizeof(*link), GFP_KERNEL);
2758         if (!link) {
2759                 err = -ENOMEM;
2760                 goto error;
2761         }
2762
2763         bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2764                       &bpf_kprobe_multi_link_lops, prog);
2765
2766         err = bpf_link_prime(&link->link, &link_primer);
2767         if (err)
2768                 goto error;
2769
2770         if (flags & BPF_F_KPROBE_MULTI_RETURN)
2771                 link->fp.exit_handler = kprobe_multi_link_handler;
2772         else
2773                 link->fp.entry_handler = kprobe_multi_link_handler;
2774
2775         link->addrs = addrs;
2776         link->cookies = cookies;
2777         link->cnt = cnt;
2778
2779         if (cookies) {
2780                 /*
2781                  * Sorting addresses will trigger sorting cookies as well
2782                  * (check bpf_kprobe_multi_cookie_swap). This way we can
2783                  * find cookie based on the address in bpf_get_attach_cookie
2784                  * helper.
2785                  */
2786                 sort_r(addrs, cnt, sizeof(*addrs),
2787                        bpf_kprobe_multi_cookie_cmp,
2788                        bpf_kprobe_multi_cookie_swap,
2789                        link);
2790         }
2791
2792         err = register_fprobe_ips(&link->fp, addrs, cnt);
2793         if (err) {
2794                 bpf_link_cleanup(&link_primer);
2795                 return err;
2796         }
2797
2798         return bpf_link_settle(&link_primer);
2799
2800 error:
2801         kfree(link);
2802         kvfree(addrs);
2803         kvfree(cookies);
2804         return err;
2805 }
2806 #else /* !CONFIG_FPROBE */
2807 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2808 {
2809         return -EOPNOTSUPP;
2810 }
2811 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2812 {
2813         return 0;
2814 }
2815 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2816 {
2817         return 0;
2818 }
2819 #endif