kernel/trace/bpf_trace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   3  * Copyright (c) 2016 Facebook
   4  */
   5 #include <linux/kernel.h>
   6 #include <linux/types.h>
   7 #include <linux/slab.h>
   8 #include <linux/bpf.h>
   9 #include <linux/bpf_perf_event.h>
  10 #include <linux/btf.h>
  11 #include <linux/filter.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/ctype.h>
  14 #include <linux/kprobes.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/btf_ids.h>
  19 #include <linux/bpf_lsm.h>
  20 #include <linux/fprobe.h>
  21 #include <linux/bsearch.h>
  22 #include <linux/sort.h>
  23
  24 #include <net/bpf_sk_storage.h>
  25
  26 #include <uapi/linux/bpf.h>
  27 #include <uapi/linux/btf.h>
  28
  29 #include <asm/tlb.h>
  30
  31 #include "trace_probe.h"
  32 #include "trace.h"
  33
  34 #define CREATE_TRACE_POINTS
  35 #include "bpf_trace.h"
  36
  37 #define bpf_event_rcu_dereference(p)                                    \
  38         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  39
  40 #ifdef CONFIG_MODULES
  41 struct bpf_trace_module {
  42         struct module *module;
  43         struct list_head list;
  44 };
  45
  46 static LIST_HEAD(bpf_trace_modules);
  47 static DEFINE_MUTEX(bpf_module_mutex);
  48
  49 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  50 {
  51         struct bpf_raw_event_map *btp, *ret = NULL;
  52         struct bpf_trace_module *btm;
  53         unsigned int i;
  54
  55         mutex_lock(&bpf_module_mutex);
  56         list_for_each_entry(btm, &bpf_trace_modules, list) {
  57                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  58                         btp = &btm->module->bpf_raw_events[i];
  59                         if (!strcmp(btp->tp->name, name)) {
  60                                 if (try_module_get(btm->module))
  61                                         ret = btp;
  62                                 goto out;
  63                         }
  64                 }
  65         }
  66 out:
  67         mutex_unlock(&bpf_module_mutex);
  68         return ret;
  69 }
  70 #else
  71 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  72 {
  73         return NULL;
  74 }
  75 #endif /* CONFIG_MODULES */
  76
  77 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  78 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  79
  80 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
  81                                   u64 flags, const struct btf **btf,
  82                                   s32 *btf_id);
  83 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
  84 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
  85
  86 /**
  87  * trace_call_bpf - invoke BPF program
  88  * @call: tracepoint event
  89  * @ctx: opaque context pointer
  90  *
  91  * kprobe handlers execute BPF programs via this helper.
  92  * Can be used from static tracepoints in the future.
  93  *
  94  * Return: BPF programs always return an integer which is interpreted by
  95  * kprobe handler as:
  96  * 0 - return from kprobe (event is filtered out)
  97  * 1 - store kprobe event into ring buffer
  98  * Other values are reserved and currently alias to 1
  99  */
 100 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 101 {
 102         unsigned int ret;
 103
 104         cant_sleep();
 105
 106         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 107                 /*
 108                  * since some bpf program is already running on this cpu,
 109                  * don't call into another bpf program (same or different)
 110                  * and don't send kprobe event into ring-buffer,
 111                  * so return zero here
 112                  */
 113                 ret = 0;
 114                 goto out;
 115         }
 116
 117         /*
 118          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 119          * to all call sites, we did a bpf_prog_array_valid() there to check
 120          * whether call->prog_array is empty or not, which is
 121          * a heuristic to speed up execution.
 122          *
 123          * If bpf_prog_array_valid() fetched prog_array was
 124          * non-NULL, we go into trace_call_bpf() and do the actual
 125          * proper rcu_dereference() under RCU lock.
 126          * If it turns out that prog_array is NULL then, we bail out.
 127          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 128          * was NULL, you'll skip the prog_array with the risk of missing
 129          * out of events when it was updated in between this and the
 130          * rcu_dereference() which is accepted risk.
 131          */
 132         ret = BPF_PROG_RUN_ARRAY(call->prog_array, ctx, bpf_prog_run);
 133
 134  out:
 135         __this_cpu_dec(bpf_prog_active);
 136
 137         return ret;
 138 }
 139
 140 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 141 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 142 {
 143         regs_set_return_value(regs, rc);
 144         override_function_with_return(regs);
 145         return 0;
 146 }
 147
 148 static const struct bpf_func_proto bpf_override_return_proto = {
 149         .func           = bpf_override_return,
 150         .gpl_only       = true,
 151         .ret_type       = RET_INTEGER,
 152         .arg1_type      = ARG_PTR_TO_CTX,
 153         .arg2_type      = ARG_ANYTHING,
 154 };
 155 #endif
 156
 157 static __always_inline int
 158 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 159 {
 160         int ret;
 161
 162         ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 163         if (unlikely(ret < 0))
 164                 memset(dst, 0, size);
 165         return ret;
 166 }
 167
 168 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 169            const void __user *, unsafe_ptr)
 170 {
 171         return bpf_probe_read_user_common(dst, size, unsafe_ptr);
 172 }
 173
 174 const struct bpf_func_proto bpf_probe_read_user_proto = {
 175         .func           = bpf_probe_read_user,
 176         .gpl_only       = true,
 177         .ret_type       = RET_INTEGER,
 178         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 179         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 180         .arg3_type      = ARG_ANYTHING,
 181 };
 182
 183 static __always_inline int
 184 bpf_probe_read_user_str_common(void *dst, u32 size,
 185                                const void __user *unsafe_ptr)
 186 {
 187         int ret;
 188
 189         /*
 190          * NB: We rely on strncpy_from_user() not copying junk past the NUL
 191          * terminator into `dst`.
 192          *
 193          * strncpy_from_user() does long-sized strides in the fast path. If the
 194          * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
 195          * then there could be junk after the NUL in `dst`. If user takes `dst`
 196          * and keys a hash map with it, then semantically identical strings can
 197          * occupy multiple entries in the map.
 198          */
 199         ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 200         if (unlikely(ret < 0))
 201                 memset(dst, 0, size);
 202         return ret;
 203 }
 204
 205 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 206            const void __user *, unsafe_ptr)
 207 {
 208         return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
 209 }
 210
 211 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 212         .func           = bpf_probe_read_user_str,
 213         .gpl_only       = true,
 214         .ret_type       = RET_INTEGER,
 215         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 216         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 217         .arg3_type      = ARG_ANYTHING,
 218 };
 219
 220 static __always_inline int
 221 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
 222 {
 223         int ret;
 224
 225         ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
 226         if (unlikely(ret < 0))
 227                 memset(dst, 0, size);
 228         return ret;
 229 }
 230
 231 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 232            const void *, unsafe_ptr)
 233 {
 234         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 235 }
 236
 237 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 238         .func           = bpf_probe_read_kernel,
 239         .gpl_only       = true,
 240         .ret_type       = RET_INTEGER,
 241         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 242         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 243         .arg3_type      = ARG_ANYTHING,
 244 };
 245
 246 static __always_inline int
 247 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
 248 {
 249         int ret;
 250
 251         /*
 252          * The strncpy_from_kernel_nofault() call will likely not fill the
 253          * entire buffer, but that's okay in this circumstance as we're probing
 254          * arbitrary memory anyway similar to bpf_probe_read_*() and might
 255          * as well probe the stack. Thus, memory is explicitly cleared
 256          * only in error case, so that improper users ignoring return
 257          * code altogether don't copy garbage; otherwise length of string
 258          * is returned that can be used for bpf_perf_event_output() et al.
 259          */
 260         ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
 261         if (unlikely(ret < 0))
 262                 memset(dst, 0, size);
 263         return ret;
 264 }
 265
 266 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 267            const void *, unsafe_ptr)
 268 {
 269         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 270 }
 271
 272 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 273         .func           = bpf_probe_read_kernel_str,
 274         .gpl_only       = true,
 275         .ret_type       = RET_INTEGER,
 276         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 277         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 278         .arg3_type      = ARG_ANYTHING,
 279 };
 280
 281 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 282 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 283            const void *, unsafe_ptr)
 284 {
 285         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 286                 return bpf_probe_read_user_common(dst, size,
 287                                 (__force void __user *)unsafe_ptr);
 288         }
 289         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 290 }
 291
 292 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 293         .func           = bpf_probe_read_compat,
 294         .gpl_only       = true,
 295         .ret_type       = RET_INTEGER,
 296         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 297         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 298         .arg3_type      = ARG_ANYTHING,
 299 };
 300
 301 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 302            const void *, unsafe_ptr)
 303 {
 304         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 305                 return bpf_probe_read_user_str_common(dst, size,
 306                                 (__force void __user *)unsafe_ptr);
 307         }
 308         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 309 }
 310
 311 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 312         .func           = bpf_probe_read_compat_str,
 313         .gpl_only       = true,
 314         .ret_type       = RET_INTEGER,
 315         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 316         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 317         .arg3_type      = ARG_ANYTHING,
 318 };
 319 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 320
 321 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 322            u32, size)
 323 {
 324         /*
 325          * Ensure we're in user context which is safe for the helper to
 326          * run. This helper has no business in a kthread.
 327          *
 328          * access_ok() should prevent writing to non-user memory, but in
 329          * some situations (nommu, temporary switch, etc) access_ok() does
 330          * not provide enough validation, hence the check on KERNEL_DS.
 331          *
 332          * nmi_uaccess_okay() ensures the probe is not run in an interim
 333          * state, when the task or mm are switched. This is specifically
 334          * required to prevent the use of temporary mm.
 335          */
 336
 337         if (unlikely(in_interrupt() ||
 338                      current->flags & (PF_KTHREAD | PF_EXITING)))
 339                 return -EPERM;
 340         if (unlikely(!nmi_uaccess_okay()))
 341                 return -EPERM;
 342
 343         return copy_to_user_nofault(unsafe_ptr, src, size);
 344 }
 345
 346 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 347         .func           = bpf_probe_write_user,
 348         .gpl_only       = true,
 349         .ret_type       = RET_INTEGER,
 350         .arg1_type      = ARG_ANYTHING,
 351         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 352         .arg3_type      = ARG_CONST_SIZE,
 353 };
 354
 355 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 356 {
 357         if (!capable(CAP_SYS_ADMIN))
 358                 return NULL;
 359
 360         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 361                             current->comm, task_pid_nr(current));
 362
 363         return &bpf_probe_write_user_proto;
 364 }
 365
 366 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
 367
 368 #define MAX_TRACE_PRINTK_VARARGS        3
 369 #define BPF_TRACE_PRINTK_SIZE           1024
 370
 371 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 372            u64, arg2, u64, arg3)
 373 {
 374         u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
 375         u32 *bin_args;
 376         static char buf[BPF_TRACE_PRINTK_SIZE];
 377         unsigned long flags;
 378         int ret;
 379
 380         ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
 381                                   MAX_TRACE_PRINTK_VARARGS);
 382         if (ret < 0)
 383                 return ret;
 384
 385         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 386         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 387
 388         trace_bpf_trace_printk(buf);
 389         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 390
 391         bpf_bprintf_cleanup();
 392
 393         return ret;
 394 }
 395
 396 static const struct bpf_func_proto bpf_trace_printk_proto = {
 397         .func           = bpf_trace_printk,
 398         .gpl_only       = true,
 399         .ret_type       = RET_INTEGER,
 400         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 401         .arg2_type      = ARG_CONST_SIZE,
 402 };
 403
 404 static void __set_printk_clr_event(void)
 405 {
 406         /*
 407          * This program might be calling bpf_trace_printk,
 408          * so enable the associated bpf_trace/bpf_trace_printk event.
 409          * Repeat this each time as it is possible a user has
 410          * disabled bpf_trace_printk events.  By loading a program
 411          * calling bpf_trace_printk() however the user has expressed
 412          * the intent to see such events.
 413          */
 414         if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 415                 pr_warn_ratelimited("could not enable bpf_trace_printk events");
 416 }
 417
 418 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 419 {
 420         __set_printk_clr_event();
 421         return &bpf_trace_printk_proto;
 422 }
 423
 424 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
 425            u32, data_len)
 426 {
 427         static char buf[BPF_TRACE_PRINTK_SIZE];
 428         unsigned long flags;
 429         int ret, num_args;
 430         u32 *bin_args;
 431
 432         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 433             (data_len && !data))
 434                 return -EINVAL;
 435         num_args = data_len / 8;
 436
 437         ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 438         if (ret < 0)
 439                 return ret;
 440
 441         raw_spin_lock_irqsave(&trace_printk_lock, flags);
 442         ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 443
 444         trace_bpf_trace_printk(buf);
 445         raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 446
 447         bpf_bprintf_cleanup();
 448
 449         return ret;
 450 }
 451
 452 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
 453         .func           = bpf_trace_vprintk,
 454         .gpl_only       = true,
 455         .ret_type       = RET_INTEGER,
 456         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 457         .arg2_type      = ARG_CONST_SIZE,
 458         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 459         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
 460 };
 461
 462 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
 463 {
 464         __set_printk_clr_event();
 465         return &bpf_trace_vprintk_proto;
 466 }
 467
 468 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 469            const void *, data, u32, data_len)
 470 {
 471         int err, num_args;
 472         u32 *bin_args;
 473
 474         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 475             (data_len && !data))
 476                 return -EINVAL;
 477         num_args = data_len / 8;
 478
 479         err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 480         if (err < 0)
 481                 return err;
 482
 483         seq_bprintf(m, fmt, bin_args);
 484
 485         bpf_bprintf_cleanup();
 486
 487         return seq_has_overflowed(m) ? -EOVERFLOW : 0;
 488 }
 489
 490 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
 491
 492 static const struct bpf_func_proto bpf_seq_printf_proto = {
 493         .func           = bpf_seq_printf,
 494         .gpl_only       = true,
 495         .ret_type       = RET_INTEGER,
 496         .arg1_type      = ARG_PTR_TO_BTF_ID,
 497         .arg1_btf_id    = &btf_seq_file_ids[0],
 498         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 499         .arg3_type      = ARG_CONST_SIZE,
 500         .arg4_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 501         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 502 };
 503
 504 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
 505 {
 506         return seq_write(m, data, len) ? -EOVERFLOW : 0;
 507 }
 508
 509 static const struct bpf_func_proto bpf_seq_write_proto = {
 510         .func           = bpf_seq_write,
 511         .gpl_only       = true,
 512         .ret_type       = RET_INTEGER,
 513         .arg1_type      = ARG_PTR_TO_BTF_ID,
 514         .arg1_btf_id    = &btf_seq_file_ids[0],
 515         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 516         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 517 };
 518
 519 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
 520            u32, btf_ptr_size, u64, flags)
 521 {
 522         const struct btf *btf;
 523         s32 btf_id;
 524         int ret;
 525
 526         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 527         if (ret)
 528                 return ret;
 529
 530         return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
 531 }
 532
 533 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
 534         .func           = bpf_seq_printf_btf,
 535         .gpl_only       = true,
 536         .ret_type       = RET_INTEGER,
 537         .arg1_type      = ARG_PTR_TO_BTF_ID,
 538         .arg1_btf_id    = &btf_seq_file_ids[0],
 539         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 540         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 541         .arg4_type      = ARG_ANYTHING,
 542 };
 543
 544 static __always_inline int
 545 get_map_perf_counter(struct bpf_map *map, u64 flags,
 546                      u64 *value, u64 *enabled, u64 *running)
 547 {
 548         struct bpf_array *array = container_of(map, struct bpf_array, map);
 549         unsigned int cpu = smp_processor_id();
 550         u64 index = flags & BPF_F_INDEX_MASK;
 551         struct bpf_event_entry *ee;
 552
 553         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 554                 return -EINVAL;
 555         if (index == BPF_F_CURRENT_CPU)
 556                 index = cpu;
 557         if (unlikely(index >= array->map.max_entries))
 558                 return -E2BIG;
 559
 560         ee = READ_ONCE(array->ptrs[index]);
 561         if (!ee)
 562                 return -ENOENT;
 563
 564         return perf_event_read_local(ee->event, value, enabled, running);
 565 }
 566
 567 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 568 {
 569         u64 value = 0;
 570         int err;
 571
 572         err = get_map_perf_counter(map, flags, &value, NULL, NULL);
 573         /*
 574          * this api is ugly since we miss [-22..-2] range of valid
 575          * counter values, but that's uapi
 576          */
 577         if (err)
 578                 return err;
 579         return value;
 580 }
 581
 582 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 583         .func           = bpf_perf_event_read,
 584         .gpl_only       = true,
 585         .ret_type       = RET_INTEGER,
 586         .arg1_type      = ARG_CONST_MAP_PTR,
 587         .arg2_type      = ARG_ANYTHING,
 588 };
 589
 590 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
 591            struct bpf_perf_event_value *, buf, u32, size)
 592 {
 593         int err = -EINVAL;
 594
 595         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
 596                 goto clear;
 597         err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
 598                                    &buf->running);
 599         if (unlikely(err))
 600                 goto clear;
 601         return 0;
 602 clear:
 603         memset(buf, 0, size);
 604         return err;
 605 }
 606
 607 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
 608         .func           = bpf_perf_event_read_value,
 609         .gpl_only       = true,
 610         .ret_type       = RET_INTEGER,
 611         .arg1_type      = ARG_CONST_MAP_PTR,
 612         .arg2_type      = ARG_ANYTHING,
 613         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
 614         .arg4_type      = ARG_CONST_SIZE,
 615 };
 616
 617 static __always_inline u64
 618 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 619                         u64 flags, struct perf_sample_data *sd)
 620 {
 621         struct bpf_array *array = container_of(map, struct bpf_array, map);
 622         unsigned int cpu = smp_processor_id();
 623         u64 index = flags & BPF_F_INDEX_MASK;
 624         struct bpf_event_entry *ee;
 625         struct perf_event *event;
 626
 627         if (index == BPF_F_CURRENT_CPU)
 628                 index = cpu;
 629         if (unlikely(index >= array->map.max_entries))
 630                 return -E2BIG;
 631
 632         ee = READ_ONCE(array->ptrs[index]);
 633         if (!ee)
 634                 return -ENOENT;
 635
 636         event = ee->event;
 637         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 638                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 639                 return -EINVAL;
 640
 641         if (unlikely(event->oncpu != cpu))
 642                 return -EOPNOTSUPP;
 643
 644         return perf_event_output(event, sd, regs);
 645 }
 646
 647 /*
 648  * Support executing tracepoints in normal, irq, and nmi context that each call
 649  * bpf_perf_event_output
 650  */
 651 struct bpf_trace_sample_data {
 652         struct perf_sample_data sds[3];
 653 };
 654
 655 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 656 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 657 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 658            u64, flags, void *, data, u64, size)
 659 {
 660         struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
 661         int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 662         struct perf_raw_record raw = {
 663                 .frag = {
 664                         .size = size,
 665                         .data = data,
 666                 },
 667         };
 668         struct perf_sample_data *sd;
 669         int err;
 670
 671         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 672                 err = -EBUSY;
 673                 goto out;
 674         }
 675
 676         sd = &sds->sds[nest_level - 1];
 677
 678         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 679                 err = -EINVAL;
 680                 goto out;
 681         }
 682
 683         perf_sample_data_init(sd, 0, 0);
 684         sd->raw = &raw;
 685
 686         err = __bpf_perf_event_output(regs, map, flags, sd);
 687
 688 out:
 689         this_cpu_dec(bpf_trace_nest_level);
 690         return err;
 691 }
 692
 693 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 694         .func           = bpf_perf_event_output,
 695         .gpl_only       = true,
 696         .ret_type       = RET_INTEGER,
 697         .arg1_type      = ARG_PTR_TO_CTX,
 698         .arg2_type      = ARG_CONST_MAP_PTR,
 699         .arg3_type      = ARG_ANYTHING,
 700         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 701         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 702 };
 703
 704 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 705 struct bpf_nested_pt_regs {
 706         struct pt_regs regs[3];
 707 };
 708 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 709 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 710
 711 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 712                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 713 {
 714         int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 715         struct perf_raw_frag frag = {
 716                 .copy           = ctx_copy,
 717                 .size           = ctx_size,
 718                 .data           = ctx,
 719         };
 720         struct perf_raw_record raw = {
 721                 .frag = {
 722                         {
 723                                 .next   = ctx_size ? &frag : NULL,
 724                         },
 725                         .size   = meta_size,
 726                         .data   = meta,
 727                 },
 728         };
 729         struct perf_sample_data *sd;
 730         struct pt_regs *regs;
 731         u64 ret;
 732
 733         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 734                 ret = -EBUSY;
 735                 goto out;
 736         }
 737         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 738         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 739
 740         perf_fetch_caller_regs(regs);
 741         perf_sample_data_init(sd, 0, 0);
 742         sd->raw = &raw;
 743
 744         ret = __bpf_perf_event_output(regs, map, flags, sd);
 745 out:
 746         this_cpu_dec(bpf_event_output_nest_level);
 747         return ret;
 748 }
 749
 750 BPF_CALL_0(bpf_get_current_task)
 751 {
 752         return (long) current;
 753 }
 754
 755 const struct bpf_func_proto bpf_get_current_task_proto = {
 756         .func           = bpf_get_current_task,
 757         .gpl_only       = true,
 758         .ret_type       = RET_INTEGER,
 759 };
 760
 761 BPF_CALL_0(bpf_get_current_task_btf)
 762 {
 763         return (unsigned long) current;
 764 }
 765
 766 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
 767         .func           = bpf_get_current_task_btf,
 768         .gpl_only       = true,
 769         .ret_type       = RET_PTR_TO_BTF_ID,
 770         .ret_btf_id     = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 771 };
 772
 773 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
 774 {
 775         return (unsigned long) task_pt_regs(task);
 776 }
 777
 778 BTF_ID_LIST(bpf_task_pt_regs_ids)
 779 BTF_ID(struct, pt_regs)
 780
 781 const struct bpf_func_proto bpf_task_pt_regs_proto = {
 782         .func           = bpf_task_pt_regs,
 783         .gpl_only       = true,
 784         .arg1_type      = ARG_PTR_TO_BTF_ID,
 785         .arg1_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 786         .ret_type       = RET_PTR_TO_BTF_ID,
 787         .ret_btf_id     = &bpf_task_pt_regs_ids[0],
 788 };
 789
 790 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 791 {
 792         struct bpf_array *array = container_of(map, struct bpf_array, map);
 793         struct cgroup *cgrp;
 794
 795         if (unlikely(idx >= array->map.max_entries))
 796                 return -E2BIG;
 797
 798         cgrp = READ_ONCE(array->ptrs[idx]);
 799         if (unlikely(!cgrp))
 800                 return -EAGAIN;
 801
 802         return task_under_cgroup_hierarchy(current, cgrp);
 803 }
 804
 805 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 806         .func           = bpf_current_task_under_cgroup,
 807         .gpl_only       = false,
 808         .ret_type       = RET_INTEGER,
 809         .arg1_type      = ARG_CONST_MAP_PTR,
 810         .arg2_type      = ARG_ANYTHING,
 811 };
 812
 813 struct send_signal_irq_work {
 814         struct irq_work irq_work;
 815         struct task_struct *task;
 816         u32 sig;
 817         enum pid_type type;
 818 };
 819
 820 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 821
 822 static void do_bpf_send_signal(struct irq_work *entry)
 823 {
 824         struct send_signal_irq_work *work;
 825
 826         work = container_of(entry, struct send_signal_irq_work, irq_work);
 827         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 828 }
 829
 830 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 831 {
 832         struct send_signal_irq_work *work = NULL;
 833
 834         /* Similar to bpf_probe_write_user, task needs to be
 835          * in a sound condition and kernel memory access be
 836          * permitted in order to send signal to the current
 837          * task.
 838          */
 839         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 840                 return -EPERM;
 841         if (unlikely(!nmi_uaccess_okay()))
 842                 return -EPERM;
 843
 844         if (irqs_disabled()) {
 845                 /* Do an early check on signal validity. Otherwise,
 846                  * the error is lost in deferred irq_work.
 847                  */
 848                 if (unlikely(!valid_signal(sig)))
 849                         return -EINVAL;
 850
 851                 work = this_cpu_ptr(&send_signal_work);
 852                 if (irq_work_is_busy(&work->irq_work))
 853                         return -EBUSY;
 854
 855                 /* Add the current task, which is the target of sending signal,
 856                  * to the irq_work. The current task may change when queued
 857                  * irq works get executed.
 858                  */
 859                 work->task = current;
 860                 work->sig = sig;
 861                 work->type = type;
 862                 irq_work_queue(&work->irq_work);
 863                 return 0;
 864         }
 865
 866         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 867 }
 868
 869 BPF_CALL_1(bpf_send_signal, u32, sig)
 870 {
 871         return bpf_send_signal_common(sig, PIDTYPE_TGID);
 872 }
 873
 874 static const struct bpf_func_proto bpf_send_signal_proto = {
 875         .func           = bpf_send_signal,
 876         .gpl_only       = false,
 877         .ret_type       = RET_INTEGER,
 878         .arg1_type      = ARG_ANYTHING,
 879 };
 880
 881 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 882 {
 883         return bpf_send_signal_common(sig, PIDTYPE_PID);
 884 }
 885
 886 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 887         .func           = bpf_send_signal_thread,
 888         .gpl_only       = false,
 889         .ret_type       = RET_INTEGER,
 890         .arg1_type      = ARG_ANYTHING,
 891 };
 892
 893 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 894 {
 895         long len;
 896         char *p;
 897
 898         if (!sz)
 899                 return 0;
 900
 901         p = d_path(path, buf, sz);
 902         if (IS_ERR(p)) {
 903                 len = PTR_ERR(p);
 904         } else {
 905                 len = buf + sz - p;
 906                 memmove(buf, p, len);
 907         }
 908
 909         return len;
 910 }
 911
 912 BTF_SET_START(btf_allowlist_d_path)
 913 #ifdef CONFIG_SECURITY
 914 BTF_ID(func, security_file_permission)
 915 BTF_ID(func, security_inode_getattr)
 916 BTF_ID(func, security_file_open)
 917 #endif
 918 #ifdef CONFIG_SECURITY_PATH
 919 BTF_ID(func, security_path_truncate)
 920 #endif
 921 BTF_ID(func, vfs_truncate)
 922 BTF_ID(func, vfs_fallocate)
 923 BTF_ID(func, dentry_open)
 924 BTF_ID(func, vfs_getattr)
 925 BTF_ID(func, filp_close)
 926 BTF_SET_END(btf_allowlist_d_path)
 927
 928 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 929 {
 930         if (prog->type == BPF_PROG_TYPE_TRACING &&
 931             prog->expected_attach_type == BPF_TRACE_ITER)
 932                 return true;
 933
 934         if (prog->type == BPF_PROG_TYPE_LSM)
 935                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 936
 937         return btf_id_set_contains(&btf_allowlist_d_path,
 938                                    prog->aux->attach_btf_id);
 939 }
 940
 941 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 942
 943 static const struct bpf_func_proto bpf_d_path_proto = {
 944         .func           = bpf_d_path,
 945         .gpl_only       = false,
 946         .ret_type       = RET_INTEGER,
 947         .arg1_type      = ARG_PTR_TO_BTF_ID,
 948         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
 949         .arg2_type      = ARG_PTR_TO_MEM,
 950         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 951         .allowed        = bpf_d_path_allowed,
 952 };
 953
 954 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
 955                          BTF_F_PTR_RAW | BTF_F_ZERO)
 956
 957 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 958                                   u64 flags, const struct btf **btf,
 959                                   s32 *btf_id)
 960 {
 961         const struct btf_type *t;
 962
 963         if (unlikely(flags & ~(BTF_F_ALL)))
 964                 return -EINVAL;
 965
 966         if (btf_ptr_size != sizeof(struct btf_ptr))
 967                 return -EINVAL;
 968
 969         *btf = bpf_get_btf_vmlinux();
 970
 971         if (IS_ERR_OR_NULL(*btf))
 972                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 973
 974         if (ptr->type_id > 0)
 975                 *btf_id = ptr->type_id;
 976         else
 977                 return -EINVAL;
 978
 979         if (*btf_id > 0)
 980                 t = btf_type_by_id(*btf, *btf_id);
 981         if (*btf_id <= 0 || !t)
 982                 return -ENOENT;
 983
 984         return 0;
 985 }
 986
 987 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
 988            u32, btf_ptr_size, u64, flags)
 989 {
 990         const struct btf *btf;
 991         s32 btf_id;
 992         int ret;
 993
 994         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 995         if (ret)
 996                 return ret;
 997
 998         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
 999                                       flags);
1000 }
1001
1002 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1003         .func           = bpf_snprintf_btf,
1004         .gpl_only       = false,
1005         .ret_type       = RET_INTEGER,
1006         .arg1_type      = ARG_PTR_TO_MEM,
1007         .arg2_type      = ARG_CONST_SIZE,
1008         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1009         .arg4_type      = ARG_CONST_SIZE,
1010         .arg5_type      = ARG_ANYTHING,
1011 };
1012
1013 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1014 {
1015         /* This helper call is inlined by verifier. */
1016         return ((u64 *)ctx)[-2];
1017 }
1018
1019 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1020         .func           = bpf_get_func_ip_tracing,
1021         .gpl_only       = true,
1022         .ret_type       = RET_INTEGER,
1023         .arg1_type      = ARG_PTR_TO_CTX,
1024 };
1025
1026 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1027 {
1028         struct kprobe *kp = kprobe_running();
1029
1030         return kp ? (uintptr_t)kp->addr : 0;
1031 }
1032
1033 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1034         .func           = bpf_get_func_ip_kprobe,
1035         .gpl_only       = true,
1036         .ret_type       = RET_INTEGER,
1037         .arg1_type      = ARG_PTR_TO_CTX,
1038 };
1039
1040 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1041 {
1042         return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1043 }
1044
1045 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1046         .func           = bpf_get_func_ip_kprobe_multi,
1047         .gpl_only       = false,
1048         .ret_type       = RET_INTEGER,
1049         .arg1_type      = ARG_PTR_TO_CTX,
1050 };
1051
1052 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1053 {
1054         return bpf_kprobe_multi_cookie(current->bpf_ctx);
1055 }
1056
1057 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1058         .func           = bpf_get_attach_cookie_kprobe_multi,
1059         .gpl_only       = false,
1060         .ret_type       = RET_INTEGER,
1061         .arg1_type      = ARG_PTR_TO_CTX,
1062 };
1063
1064 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1065 {
1066         struct bpf_trace_run_ctx *run_ctx;
1067
1068         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1069         return run_ctx->bpf_cookie;
1070 }
1071
1072 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1073         .func           = bpf_get_attach_cookie_trace,
1074         .gpl_only       = false,
1075         .ret_type       = RET_INTEGER,
1076         .arg1_type      = ARG_PTR_TO_CTX,
1077 };
1078
1079 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1080 {
1081         return ctx->event->bpf_cookie;
1082 }
1083
1084 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1085         .func           = bpf_get_attach_cookie_pe,
1086         .gpl_only       = false,
1087         .ret_type       = RET_INTEGER,
1088         .arg1_type      = ARG_PTR_TO_CTX,
1089 };
1090
1091 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1092 {
1093 #ifndef CONFIG_X86
1094         return -ENOENT;
1095 #else
1096         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1097         u32 entry_cnt = size / br_entry_size;
1098
1099         entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1100
1101         if (unlikely(flags))
1102                 return -EINVAL;
1103
1104         if (!entry_cnt)
1105                 return -ENOENT;
1106
1107         return entry_cnt * br_entry_size;
1108 #endif
1109 }
1110
1111 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1112         .func           = bpf_get_branch_snapshot,
1113         .gpl_only       = true,
1114         .ret_type       = RET_INTEGER,
1115         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1116         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1117 };
1118
1119 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1120 {
1121         /* This helper call is inlined by verifier. */
1122         u64 nr_args = ((u64 *)ctx)[-1];
1123
1124         if ((u64) n >= nr_args)
1125                 return -EINVAL;
1126         *value = ((u64 *)ctx)[n];
1127         return 0;
1128 }
1129
1130 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1131         .func           = get_func_arg,
1132         .ret_type       = RET_INTEGER,
1133         .arg1_type      = ARG_PTR_TO_CTX,
1134         .arg2_type      = ARG_ANYTHING,
1135         .arg3_type      = ARG_PTR_TO_LONG,
1136 };
1137
1138 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1139 {
1140         /* This helper call is inlined by verifier. */
1141         u64 nr_args = ((u64 *)ctx)[-1];
1142
1143         *value = ((u64 *)ctx)[nr_args];
1144         return 0;
1145 }
1146
1147 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1148         .func           = get_func_ret,
1149         .ret_type       = RET_INTEGER,
1150         .arg1_type      = ARG_PTR_TO_CTX,
1151         .arg2_type      = ARG_PTR_TO_LONG,
1152 };
1153
1154 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1155 {
1156         /* This helper call is inlined by verifier. */
1157         return ((u64 *)ctx)[-1];
1158 }
1159
1160 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1161         .func           = get_func_arg_cnt,
1162         .ret_type       = RET_INTEGER,
1163         .arg1_type      = ARG_PTR_TO_CTX,
1164 };
1165
1166 static const struct bpf_func_proto *
1167 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1168 {
1169         switch (func_id) {
1170         case BPF_FUNC_map_lookup_elem:
1171                 return &bpf_map_lookup_elem_proto;
1172         case BPF_FUNC_map_update_elem:
1173                 return &bpf_map_update_elem_proto;
1174         case BPF_FUNC_map_delete_elem:
1175                 return &bpf_map_delete_elem_proto;
1176         case BPF_FUNC_map_push_elem:
1177                 return &bpf_map_push_elem_proto;
1178         case BPF_FUNC_map_pop_elem:
1179                 return &bpf_map_pop_elem_proto;
1180         case BPF_FUNC_map_peek_elem:
1181                 return &bpf_map_peek_elem_proto;
1182         case BPF_FUNC_ktime_get_ns:
1183                 return &bpf_ktime_get_ns_proto;
1184         case BPF_FUNC_ktime_get_boot_ns:
1185                 return &bpf_ktime_get_boot_ns_proto;
1186         case BPF_FUNC_tail_call:
1187                 return &bpf_tail_call_proto;
1188         case BPF_FUNC_get_current_pid_tgid:
1189                 return &bpf_get_current_pid_tgid_proto;
1190         case BPF_FUNC_get_current_task:
1191                 return &bpf_get_current_task_proto;
1192         case BPF_FUNC_get_current_task_btf:
1193                 return &bpf_get_current_task_btf_proto;
1194         case BPF_FUNC_task_pt_regs:
1195                 return &bpf_task_pt_regs_proto;
1196         case BPF_FUNC_get_current_uid_gid:
1197                 return &bpf_get_current_uid_gid_proto;
1198         case BPF_FUNC_get_current_comm:
1199                 return &bpf_get_current_comm_proto;
1200         case BPF_FUNC_trace_printk:
1201                 return bpf_get_trace_printk_proto();
1202         case BPF_FUNC_get_smp_processor_id:
1203                 return &bpf_get_smp_processor_id_proto;
1204         case BPF_FUNC_get_numa_node_id:
1205                 return &bpf_get_numa_node_id_proto;
1206         case BPF_FUNC_perf_event_read:
1207                 return &bpf_perf_event_read_proto;
1208         case BPF_FUNC_current_task_under_cgroup:
1209                 return &bpf_current_task_under_cgroup_proto;
1210         case BPF_FUNC_get_prandom_u32:
1211                 return &bpf_get_prandom_u32_proto;
1212         case BPF_FUNC_probe_write_user:
1213                 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1214                        NULL : bpf_get_probe_write_proto();
1215         case BPF_FUNC_probe_read_user:
1216                 return &bpf_probe_read_user_proto;
1217         case BPF_FUNC_probe_read_kernel:
1218                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1219                        NULL : &bpf_probe_read_kernel_proto;
1220         case BPF_FUNC_probe_read_user_str:
1221                 return &bpf_probe_read_user_str_proto;
1222         case BPF_FUNC_probe_read_kernel_str:
1223                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1224                        NULL : &bpf_probe_read_kernel_str_proto;
1225 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1226         case BPF_FUNC_probe_read:
1227                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1228                        NULL : &bpf_probe_read_compat_proto;
1229         case BPF_FUNC_probe_read_str:
1230                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1231                        NULL : &bpf_probe_read_compat_str_proto;
1232 #endif
1233 #ifdef CONFIG_CGROUPS
1234         case BPF_FUNC_get_current_cgroup_id:
1235                 return &bpf_get_current_cgroup_id_proto;
1236         case BPF_FUNC_get_current_ancestor_cgroup_id:
1237                 return &bpf_get_current_ancestor_cgroup_id_proto;
1238 #endif
1239         case BPF_FUNC_send_signal:
1240                 return &bpf_send_signal_proto;
1241         case BPF_FUNC_send_signal_thread:
1242                 return &bpf_send_signal_thread_proto;
1243         case BPF_FUNC_perf_event_read_value:
1244                 return &bpf_perf_event_read_value_proto;
1245         case BPF_FUNC_get_ns_current_pid_tgid:
1246                 return &bpf_get_ns_current_pid_tgid_proto;
1247         case BPF_FUNC_ringbuf_output:
1248                 return &bpf_ringbuf_output_proto;
1249         case BPF_FUNC_ringbuf_reserve:
1250                 return &bpf_ringbuf_reserve_proto;
1251         case BPF_FUNC_ringbuf_submit:
1252                 return &bpf_ringbuf_submit_proto;
1253         case BPF_FUNC_ringbuf_discard:
1254                 return &bpf_ringbuf_discard_proto;
1255         case BPF_FUNC_ringbuf_query:
1256                 return &bpf_ringbuf_query_proto;
1257         case BPF_FUNC_jiffies64:
1258                 return &bpf_jiffies64_proto;
1259         case BPF_FUNC_get_task_stack:
1260                 return &bpf_get_task_stack_proto;
1261         case BPF_FUNC_copy_from_user:
1262                 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1263         case BPF_FUNC_copy_from_user_task:
1264                 return prog->aux->sleepable ? &bpf_copy_from_user_task_proto : NULL;
1265         case BPF_FUNC_snprintf_btf:
1266                 return &bpf_snprintf_btf_proto;
1267         case BPF_FUNC_per_cpu_ptr:
1268                 return &bpf_per_cpu_ptr_proto;
1269         case BPF_FUNC_this_cpu_ptr:
1270                 return &bpf_this_cpu_ptr_proto;
1271         case BPF_FUNC_task_storage_get:
1272                 return &bpf_task_storage_get_proto;
1273         case BPF_FUNC_task_storage_delete:
1274                 return &bpf_task_storage_delete_proto;
1275         case BPF_FUNC_for_each_map_elem:
1276                 return &bpf_for_each_map_elem_proto;
1277         case BPF_FUNC_snprintf:
1278                 return &bpf_snprintf_proto;
1279         case BPF_FUNC_get_func_ip:
1280                 return &bpf_get_func_ip_proto_tracing;
1281         case BPF_FUNC_get_branch_snapshot:
1282                 return &bpf_get_branch_snapshot_proto;
1283         case BPF_FUNC_find_vma:
1284                 return &bpf_find_vma_proto;
1285         case BPF_FUNC_trace_vprintk:
1286                 return bpf_get_trace_vprintk_proto();
1287         default:
1288                 return bpf_base_func_proto(func_id);
1289         }
1290 }
1291
1292 static const struct bpf_func_proto *
1293 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1294 {
1295         switch (func_id) {
1296         case BPF_FUNC_perf_event_output:
1297                 return &bpf_perf_event_output_proto;
1298         case BPF_FUNC_get_stackid:
1299                 return &bpf_get_stackid_proto;
1300         case BPF_FUNC_get_stack:
1301                 return &bpf_get_stack_proto;
1302 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1303         case BPF_FUNC_override_return:
1304                 return &bpf_override_return_proto;
1305 #endif
1306         case BPF_FUNC_get_func_ip:
1307                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1308                         &bpf_get_func_ip_proto_kprobe_multi :
1309                         &bpf_get_func_ip_proto_kprobe;
1310         case BPF_FUNC_get_attach_cookie:
1311                 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ?
1312                         &bpf_get_attach_cookie_proto_kmulti :
1313                         &bpf_get_attach_cookie_proto_trace;
1314         default:
1315                 return bpf_tracing_func_proto(func_id, prog);
1316         }
1317 }
1318
1319 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1320 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1321                                         const struct bpf_prog *prog,
1322                                         struct bpf_insn_access_aux *info)
1323 {
1324         if (off < 0 || off >= sizeof(struct pt_regs))
1325                 return false;
1326         if (type != BPF_READ)
1327                 return false;
1328         if (off % size != 0)
1329                 return false;
1330         /*
1331          * Assertion for 32 bit to make sure last 8 byte access
1332          * (BPF_DW) to the last 4 byte member is disallowed.
1333          */
1334         if (off + size > sizeof(struct pt_regs))
1335                 return false;
1336
1337         return true;
1338 }
1339
1340 const struct bpf_verifier_ops kprobe_verifier_ops = {
1341         .get_func_proto  = kprobe_prog_func_proto,
1342         .is_valid_access = kprobe_prog_is_valid_access,
1343 };
1344
1345 const struct bpf_prog_ops kprobe_prog_ops = {
1346 };
1347
1348 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1349            u64, flags, void *, data, u64, size)
1350 {
1351         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1352
1353         /*
1354          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1355          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1356          * from there and call the same bpf_perf_event_output() helper inline.
1357          */
1358         return ____bpf_perf_event_output(regs, map, flags, data, size);
1359 }
1360
1361 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1362         .func           = bpf_perf_event_output_tp,
1363         .gpl_only       = true,
1364         .ret_type       = RET_INTEGER,
1365         .arg1_type      = ARG_PTR_TO_CTX,
1366         .arg2_type      = ARG_CONST_MAP_PTR,
1367         .arg3_type      = ARG_ANYTHING,
1368         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1369         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1370 };
1371
1372 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1373            u64, flags)
1374 {
1375         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1376
1377         /*
1378          * Same comment as in bpf_perf_event_output_tp(), only that this time
1379          * the other helper's function body cannot be inlined due to being
1380          * external, thus we need to call raw helper function.
1381          */
1382         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1383                                flags, 0, 0);
1384 }
1385
1386 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1387         .func           = bpf_get_stackid_tp,
1388         .gpl_only       = true,
1389         .ret_type       = RET_INTEGER,
1390         .arg1_type      = ARG_PTR_TO_CTX,
1391         .arg2_type      = ARG_CONST_MAP_PTR,
1392         .arg3_type      = ARG_ANYTHING,
1393 };
1394
1395 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1396            u64, flags)
1397 {
1398         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1399
1400         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1401                              (unsigned long) size, flags, 0);
1402 }
1403
1404 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1405         .func           = bpf_get_stack_tp,
1406         .gpl_only       = true,
1407         .ret_type       = RET_INTEGER,
1408         .arg1_type      = ARG_PTR_TO_CTX,
1409         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1410         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1411         .arg4_type      = ARG_ANYTHING,
1412 };
1413
1414 static const struct bpf_func_proto *
1415 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1416 {
1417         switch (func_id) {
1418         case BPF_FUNC_perf_event_output:
1419                 return &bpf_perf_event_output_proto_tp;
1420         case BPF_FUNC_get_stackid:
1421                 return &bpf_get_stackid_proto_tp;
1422         case BPF_FUNC_get_stack:
1423                 return &bpf_get_stack_proto_tp;
1424         case BPF_FUNC_get_attach_cookie:
1425                 return &bpf_get_attach_cookie_proto_trace;
1426         default:
1427                 return bpf_tracing_func_proto(func_id, prog);
1428         }
1429 }
1430
1431 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1432                                     const struct bpf_prog *prog,
1433                                     struct bpf_insn_access_aux *info)
1434 {
1435         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1436                 return false;
1437         if (type != BPF_READ)
1438                 return false;
1439         if (off % size != 0)
1440                 return false;
1441
1442         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1443         return true;
1444 }
1445
1446 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1447         .get_func_proto  = tp_prog_func_proto,
1448         .is_valid_access = tp_prog_is_valid_access,
1449 };
1450
1451 const struct bpf_prog_ops tracepoint_prog_ops = {
1452 };
1453
1454 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1455            struct bpf_perf_event_value *, buf, u32, size)
1456 {
1457         int err = -EINVAL;
1458
1459         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1460                 goto clear;
1461         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1462                                     &buf->running);
1463         if (unlikely(err))
1464                 goto clear;
1465         return 0;
1466 clear:
1467         memset(buf, 0, size);
1468         return err;
1469 }
1470
1471 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1472          .func           = bpf_perf_prog_read_value,
1473          .gpl_only       = true,
1474          .ret_type       = RET_INTEGER,
1475          .arg1_type      = ARG_PTR_TO_CTX,
1476          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1477          .arg3_type      = ARG_CONST_SIZE,
1478 };
1479
1480 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1481            void *, buf, u32, size, u64, flags)
1482 {
1483         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1484         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1485         u32 to_copy;
1486
1487         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1488                 return -EINVAL;
1489
1490         if (unlikely(!br_stack))
1491                 return -ENOENT;
1492
1493         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1494                 return br_stack->nr * br_entry_size;
1495
1496         if (!buf || (size % br_entry_size != 0))
1497                 return -EINVAL;
1498
1499         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1500         memcpy(buf, br_stack->entries, to_copy);
1501
1502         return to_copy;
1503 }
1504
1505 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1506         .func           = bpf_read_branch_records,
1507         .gpl_only       = true,
1508         .ret_type       = RET_INTEGER,
1509         .arg1_type      = ARG_PTR_TO_CTX,
1510         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1511         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1512         .arg4_type      = ARG_ANYTHING,
1513 };
1514
1515 static const struct bpf_func_proto *
1516 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1517 {
1518         switch (func_id) {
1519         case BPF_FUNC_perf_event_output:
1520                 return &bpf_perf_event_output_proto_tp;
1521         case BPF_FUNC_get_stackid:
1522                 return &bpf_get_stackid_proto_pe;
1523         case BPF_FUNC_get_stack:
1524                 return &bpf_get_stack_proto_pe;
1525         case BPF_FUNC_perf_prog_read_value:
1526                 return &bpf_perf_prog_read_value_proto;
1527         case BPF_FUNC_read_branch_records:
1528                 return &bpf_read_branch_records_proto;
1529         case BPF_FUNC_get_attach_cookie:
1530                 return &bpf_get_attach_cookie_proto_pe;
1531         default:
1532                 return bpf_tracing_func_proto(func_id, prog);
1533         }
1534 }
1535
1536 /*
1537  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1538  * to avoid potential recursive reuse issue when/if tracepoints are added
1539  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1540  *
1541  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1542  * in normal, irq, and nmi context.
1543  */
1544 struct bpf_raw_tp_regs {
1545         struct pt_regs regs[3];
1546 };
1547 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1548 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1549 static struct pt_regs *get_bpf_raw_tp_regs(void)
1550 {
1551         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1552         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1553
1554         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1555                 this_cpu_dec(bpf_raw_tp_nest_level);
1556                 return ERR_PTR(-EBUSY);
1557         }
1558
1559         return &tp_regs->regs[nest_level - 1];
1560 }
1561
1562 static void put_bpf_raw_tp_regs(void)
1563 {
1564         this_cpu_dec(bpf_raw_tp_nest_level);
1565 }
1566
1567 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1568            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1569 {
1570         struct pt_regs *regs = get_bpf_raw_tp_regs();
1571         int ret;
1572
1573         if (IS_ERR(regs))
1574                 return PTR_ERR(regs);
1575
1576         perf_fetch_caller_regs(regs);
1577         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1578
1579         put_bpf_raw_tp_regs();
1580         return ret;
1581 }
1582
1583 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1584         .func           = bpf_perf_event_output_raw_tp,
1585         .gpl_only       = true,
1586         .ret_type       = RET_INTEGER,
1587         .arg1_type      = ARG_PTR_TO_CTX,
1588         .arg2_type      = ARG_CONST_MAP_PTR,
1589         .arg3_type      = ARG_ANYTHING,
1590         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1591         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1592 };
1593
1594 extern const struct bpf_func_proto bpf_skb_output_proto;
1595 extern const struct bpf_func_proto bpf_xdp_output_proto;
1596 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1597
1598 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1599            struct bpf_map *, map, u64, flags)
1600 {
1601         struct pt_regs *regs = get_bpf_raw_tp_regs();
1602         int ret;
1603
1604         if (IS_ERR(regs))
1605                 return PTR_ERR(regs);
1606
1607         perf_fetch_caller_regs(regs);
1608         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1609         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1610                               flags, 0, 0);
1611         put_bpf_raw_tp_regs();
1612         return ret;
1613 }
1614
1615 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1616         .func           = bpf_get_stackid_raw_tp,
1617         .gpl_only       = true,
1618         .ret_type       = RET_INTEGER,
1619         .arg1_type      = ARG_PTR_TO_CTX,
1620         .arg2_type      = ARG_CONST_MAP_PTR,
1621         .arg3_type      = ARG_ANYTHING,
1622 };
1623
1624 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1625            void *, buf, u32, size, u64, flags)
1626 {
1627         struct pt_regs *regs = get_bpf_raw_tp_regs();
1628         int ret;
1629
1630         if (IS_ERR(regs))
1631                 return PTR_ERR(regs);
1632
1633         perf_fetch_caller_regs(regs);
1634         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1635                             (unsigned long) size, flags, 0);
1636         put_bpf_raw_tp_regs();
1637         return ret;
1638 }
1639
1640 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1641         .func           = bpf_get_stack_raw_tp,
1642         .gpl_only       = true,
1643         .ret_type       = RET_INTEGER,
1644         .arg1_type      = ARG_PTR_TO_CTX,
1645         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1646         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1647         .arg4_type      = ARG_ANYTHING,
1648 };
1649
1650 static const struct bpf_func_proto *
1651 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1652 {
1653         switch (func_id) {
1654         case BPF_FUNC_perf_event_output:
1655                 return &bpf_perf_event_output_proto_raw_tp;
1656         case BPF_FUNC_get_stackid:
1657                 return &bpf_get_stackid_proto_raw_tp;
1658         case BPF_FUNC_get_stack:
1659                 return &bpf_get_stack_proto_raw_tp;
1660         default:
1661                 return bpf_tracing_func_proto(func_id, prog);
1662         }
1663 }
1664
1665 const struct bpf_func_proto *
1666 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1667 {
1668         const struct bpf_func_proto *fn;
1669
1670         switch (func_id) {
1671 #ifdef CONFIG_NET
1672         case BPF_FUNC_skb_output:
1673                 return &bpf_skb_output_proto;
1674         case BPF_FUNC_xdp_output:
1675                 return &bpf_xdp_output_proto;
1676         case BPF_FUNC_skc_to_tcp6_sock:
1677                 return &bpf_skc_to_tcp6_sock_proto;
1678         case BPF_FUNC_skc_to_tcp_sock:
1679                 return &bpf_skc_to_tcp_sock_proto;
1680         case BPF_FUNC_skc_to_tcp_timewait_sock:
1681                 return &bpf_skc_to_tcp_timewait_sock_proto;
1682         case BPF_FUNC_skc_to_tcp_request_sock:
1683                 return &bpf_skc_to_tcp_request_sock_proto;
1684         case BPF_FUNC_skc_to_udp6_sock:
1685                 return &bpf_skc_to_udp6_sock_proto;
1686         case BPF_FUNC_skc_to_unix_sock:
1687                 return &bpf_skc_to_unix_sock_proto;
1688         case BPF_FUNC_sk_storage_get:
1689                 return &bpf_sk_storage_get_tracing_proto;
1690         case BPF_FUNC_sk_storage_delete:
1691                 return &bpf_sk_storage_delete_tracing_proto;
1692         case BPF_FUNC_sock_from_file:
1693                 return &bpf_sock_from_file_proto;
1694         case BPF_FUNC_get_socket_cookie:
1695                 return &bpf_get_socket_ptr_cookie_proto;
1696         case BPF_FUNC_xdp_get_buff_len:
1697                 return &bpf_xdp_get_buff_len_trace_proto;
1698 #endif
1699         case BPF_FUNC_seq_printf:
1700                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1701                        &bpf_seq_printf_proto :
1702                        NULL;
1703         case BPF_FUNC_seq_write:
1704                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1705                        &bpf_seq_write_proto :
1706                        NULL;
1707         case BPF_FUNC_seq_printf_btf:
1708                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1709                        &bpf_seq_printf_btf_proto :
1710                        NULL;
1711         case BPF_FUNC_d_path:
1712                 return &bpf_d_path_proto;
1713         case BPF_FUNC_get_func_arg:
1714                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1715         case BPF_FUNC_get_func_ret:
1716                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1717         case BPF_FUNC_get_func_arg_cnt:
1718                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1719         default:
1720                 fn = raw_tp_prog_func_proto(func_id, prog);
1721                 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1722                         fn = bpf_iter_get_func_proto(func_id, prog);
1723                 return fn;
1724         }
1725 }
1726
1727 static bool raw_tp_prog_is_valid_access(int off, int size,
1728                                         enum bpf_access_type type,
1729                                         const struct bpf_prog *prog,
1730                                         struct bpf_insn_access_aux *info)
1731 {
1732         return bpf_tracing_ctx_access(off, size, type);
1733 }
1734
1735 static bool tracing_prog_is_valid_access(int off, int size,
1736                                          enum bpf_access_type type,
1737                                          const struct bpf_prog *prog,
1738                                          struct bpf_insn_access_aux *info)
1739 {
1740         return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1741 }
1742
1743 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1744                                      const union bpf_attr *kattr,
1745                                      union bpf_attr __user *uattr)
1746 {
1747         return -ENOTSUPP;
1748 }
1749
1750 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1751         .get_func_proto  = raw_tp_prog_func_proto,
1752         .is_valid_access = raw_tp_prog_is_valid_access,
1753 };
1754
1755 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1756 #ifdef CONFIG_NET
1757         .test_run = bpf_prog_test_run_raw_tp,
1758 #endif
1759 };
1760
1761 const struct bpf_verifier_ops tracing_verifier_ops = {
1762         .get_func_proto  = tracing_prog_func_proto,
1763         .is_valid_access = tracing_prog_is_valid_access,
1764 };
1765
1766 const struct bpf_prog_ops tracing_prog_ops = {
1767         .test_run = bpf_prog_test_run_tracing,
1768 };
1769
1770 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1771                                                  enum bpf_access_type type,
1772                                                  const struct bpf_prog *prog,
1773                                                  struct bpf_insn_access_aux *info)
1774 {
1775         if (off == 0) {
1776                 if (size != sizeof(u64) || type != BPF_READ)
1777                         return false;
1778                 info->reg_type = PTR_TO_TP_BUFFER;
1779         }
1780         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1781 }
1782
1783 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1784         .get_func_proto  = raw_tp_prog_func_proto,
1785         .is_valid_access = raw_tp_writable_prog_is_valid_access,
1786 };
1787
1788 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1789 };
1790
1791 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1792                                     const struct bpf_prog *prog,
1793                                     struct bpf_insn_access_aux *info)
1794 {
1795         const int size_u64 = sizeof(u64);
1796
1797         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1798                 return false;
1799         if (type != BPF_READ)
1800                 return false;
1801         if (off % size != 0) {
1802                 if (sizeof(unsigned long) != 4)
1803                         return false;
1804                 if (size != 8)
1805                         return false;
1806                 if (off % size != 4)
1807                         return false;
1808         }
1809
1810         switch (off) {
1811         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1812                 bpf_ctx_record_field_size(info, size_u64);
1813                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1814                         return false;
1815                 break;
1816         case bpf_ctx_range(struct bpf_perf_event_data, addr):
1817                 bpf_ctx_record_field_size(info, size_u64);
1818                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1819                         return false;
1820                 break;
1821         default:
1822                 if (size != sizeof(long))
1823                         return false;
1824         }
1825
1826         return true;
1827 }
1828
1829 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1830                                       const struct bpf_insn *si,
1831                                       struct bpf_insn *insn_buf,
1832                                       struct bpf_prog *prog, u32 *target_size)
1833 {
1834         struct bpf_insn *insn = insn_buf;
1835
1836         switch (si->off) {
1837         case offsetof(struct bpf_perf_event_data, sample_period):
1838                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1839                                                        data), si->dst_reg, si->src_reg,
1840                                       offsetof(struct bpf_perf_event_data_kern, data));
1841                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1842                                       bpf_target_off(struct perf_sample_data, period, 8,
1843                                                      target_size));
1844                 break;
1845         case offsetof(struct bpf_perf_event_data, addr):
1846                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1847                                                        data), si->dst_reg, si->src_reg,
1848                                       offsetof(struct bpf_perf_event_data_kern, data));
1849                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1850                                       bpf_target_off(struct perf_sample_data, addr, 8,
1851                                                      target_size));
1852                 break;
1853         default:
1854                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1855                                                        regs), si->dst_reg, si->src_reg,
1856                                       offsetof(struct bpf_perf_event_data_kern, regs));
1857                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1858                                       si->off);
1859                 break;
1860         }
1861
1862         return insn - insn_buf;
1863 }
1864
1865 const struct bpf_verifier_ops perf_event_verifier_ops = {
1866         .get_func_proto         = pe_prog_func_proto,
1867         .is_valid_access        = pe_prog_is_valid_access,
1868         .convert_ctx_access     = pe_prog_convert_ctx_access,
1869 };
1870
1871 const struct bpf_prog_ops perf_event_prog_ops = {
1872 };
1873
1874 static DEFINE_MUTEX(bpf_event_mutex);
1875
1876 #define BPF_TRACE_MAX_PROGS 64
1877
1878 int perf_event_attach_bpf_prog(struct perf_event *event,
1879                                struct bpf_prog *prog,
1880                                u64 bpf_cookie)
1881 {
1882         struct bpf_prog_array *old_array;
1883         struct bpf_prog_array *new_array;
1884         int ret = -EEXIST;
1885
1886         /*
1887          * Kprobe override only works if they are on the function entry,
1888          * and only if they are on the opt-in list.
1889          */
1890         if (prog->kprobe_override &&
1891             (!trace_kprobe_on_func_entry(event->tp_event) ||
1892              !trace_kprobe_error_injectable(event->tp_event)))
1893                 return -EINVAL;
1894
1895         mutex_lock(&bpf_event_mutex);
1896
1897         if (event->prog)
1898                 goto unlock;
1899
1900         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1901         if (old_array &&
1902             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1903                 ret = -E2BIG;
1904                 goto unlock;
1905         }
1906
1907         ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
1908         if (ret < 0)
1909                 goto unlock;
1910
1911         /* set the new array to event->tp_event and set event->prog */
1912         event->prog = prog;
1913         event->bpf_cookie = bpf_cookie;
1914         rcu_assign_pointer(event->tp_event->prog_array, new_array);
1915         bpf_prog_array_free(old_array);
1916
1917 unlock:
1918         mutex_unlock(&bpf_event_mutex);
1919         return ret;
1920 }
1921
1922 void perf_event_detach_bpf_prog(struct perf_event *event)
1923 {
1924         struct bpf_prog_array *old_array;
1925         struct bpf_prog_array *new_array;
1926         int ret;
1927
1928         mutex_lock(&bpf_event_mutex);
1929
1930         if (!event->prog)
1931                 goto unlock;
1932
1933         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1934         ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
1935         if (ret == -ENOENT)
1936                 goto unlock;
1937         if (ret < 0) {
1938                 bpf_prog_array_delete_safe(old_array, event->prog);
1939         } else {
1940                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1941                 bpf_prog_array_free(old_array);
1942         }
1943
1944         bpf_prog_put(event->prog);
1945         event->prog = NULL;
1946
1947 unlock:
1948         mutex_unlock(&bpf_event_mutex);
1949 }
1950
1951 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1952 {
1953         struct perf_event_query_bpf __user *uquery = info;
1954         struct perf_event_query_bpf query = {};
1955         struct bpf_prog_array *progs;
1956         u32 *ids, prog_cnt, ids_len;
1957         int ret;
1958
1959         if (!perfmon_capable())
1960                 return -EPERM;
1961         if (event->attr.type != PERF_TYPE_TRACEPOINT)
1962                 return -EINVAL;
1963         if (copy_from_user(&query, uquery, sizeof(query)))
1964                 return -EFAULT;
1965
1966         ids_len = query.ids_len;
1967         if (ids_len > BPF_TRACE_MAX_PROGS)
1968                 return -E2BIG;
1969         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1970         if (!ids)
1971                 return -ENOMEM;
1972         /*
1973          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1974          * is required when user only wants to check for uquery->prog_cnt.
1975          * There is no need to check for it since the case is handled
1976          * gracefully in bpf_prog_array_copy_info.
1977          */
1978
1979         mutex_lock(&bpf_event_mutex);
1980         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1981         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1982         mutex_unlock(&bpf_event_mutex);
1983
1984         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1985             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1986                 ret = -EFAULT;
1987
1988         kfree(ids);
1989         return ret;
1990 }
1991
1992 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1993 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1994
1995 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1996 {
1997         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1998
1999         for (; btp < __stop__bpf_raw_tp; btp++) {
2000                 if (!strcmp(btp->tp->name, name))
2001                         return btp;
2002         }
2003
2004         return bpf_get_raw_tracepoint_module(name);
2005 }
2006
2007 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2008 {
2009         struct module *mod;
2010
2011         preempt_disable();
2012         mod = __module_address((unsigned long)btp);
2013         module_put(mod);
2014         preempt_enable();
2015 }
2016
2017 static __always_inline
2018 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2019 {
2020         cant_sleep();
2021         rcu_read_lock();
2022         (void) bpf_prog_run(prog, args);
2023         rcu_read_unlock();
2024 }
2025
2026 #define UNPACK(...)                     __VA_ARGS__
2027 #define REPEAT_1(FN, DL, X, ...)        FN(X)
2028 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2029 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2030 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2031 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2032 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2033 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2034 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2035 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2036 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2037 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2038 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2039 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
2040
2041 #define SARG(X)         u64 arg##X
2042 #define COPY(X)         args[X] = arg##X
2043
2044 #define __DL_COM        (,)
2045 #define __DL_SEM        (;)
2046
2047 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2048
2049 #define BPF_TRACE_DEFN_x(x)                                             \
2050         void bpf_trace_run##x(struct bpf_prog *prog,                    \
2051                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
2052         {                                                               \
2053                 u64 args[x];                                            \
2054                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
2055                 __bpf_trace_run(prog, args);                            \
2056         }                                                               \
2057         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2058 BPF_TRACE_DEFN_x(1);
2059 BPF_TRACE_DEFN_x(2);
2060 BPF_TRACE_DEFN_x(3);
2061 BPF_TRACE_DEFN_x(4);
2062 BPF_TRACE_DEFN_x(5);
2063 BPF_TRACE_DEFN_x(6);
2064 BPF_TRACE_DEFN_x(7);
2065 BPF_TRACE_DEFN_x(8);
2066 BPF_TRACE_DEFN_x(9);
2067 BPF_TRACE_DEFN_x(10);
2068 BPF_TRACE_DEFN_x(11);
2069 BPF_TRACE_DEFN_x(12);
2070
2071 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2072 {
2073         struct tracepoint *tp = btp->tp;
2074
2075         /*
2076          * check that program doesn't access arguments beyond what's
2077          * available in this tracepoint
2078          */
2079         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2080                 return -EINVAL;
2081
2082         if (prog->aux->max_tp_access > btp->writable_size)
2083                 return -EINVAL;
2084
2085         return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2086                                                    prog);
2087 }
2088
2089 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2090 {
2091         return __bpf_probe_register(btp, prog);
2092 }
2093
2094 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2095 {
2096         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2097 }
2098
2099 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2100                             u32 *fd_type, const char **buf,
2101                             u64 *probe_offset, u64 *probe_addr)
2102 {
2103         bool is_tracepoint, is_syscall_tp;
2104         struct bpf_prog *prog;
2105         int flags, err = 0;
2106
2107         prog = event->prog;
2108         if (!prog)
2109                 return -ENOENT;
2110
2111         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2112         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2113                 return -EOPNOTSUPP;
2114
2115         *prog_id = prog->aux->id;
2116         flags = event->tp_event->flags;
2117         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2118         is_syscall_tp = is_syscall_trace_event(event->tp_event);
2119
2120         if (is_tracepoint || is_syscall_tp) {
2121                 *buf = is_tracepoint ? event->tp_event->tp->name
2122                                      : event->tp_event->name;
2123                 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2124                 *probe_offset = 0x0;
2125                 *probe_addr = 0x0;
2126         } else {
2127                 /* kprobe/uprobe */
2128                 err = -EOPNOTSUPP;
2129 #ifdef CONFIG_KPROBE_EVENTS
2130                 if (flags & TRACE_EVENT_FL_KPROBE)
2131                         err = bpf_get_kprobe_info(event, fd_type, buf,
2132                                                   probe_offset, probe_addr,
2133                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2134 #endif
2135 #ifdef CONFIG_UPROBE_EVENTS
2136                 if (flags & TRACE_EVENT_FL_UPROBE)
2137                         err = bpf_get_uprobe_info(event, fd_type, buf,
2138                                                   probe_offset,
2139                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2140 #endif
2141         }
2142
2143         return err;
2144 }
2145
2146 static int __init send_signal_irq_work_init(void)
2147 {
2148         int cpu;
2149         struct send_signal_irq_work *work;
2150
2151         for_each_possible_cpu(cpu) {
2152                 work = per_cpu_ptr(&send_signal_work, cpu);
2153                 init_irq_work(&work->irq_work, do_bpf_send_signal);
2154         }
2155         return 0;
2156 }
2157
2158 subsys_initcall(send_signal_irq_work_init);
2159
2160 #ifdef CONFIG_MODULES
2161 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2162                             void *module)
2163 {
2164         struct bpf_trace_module *btm, *tmp;
2165         struct module *mod = module;
2166         int ret = 0;
2167
2168         if (mod->num_bpf_raw_events == 0 ||
2169             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2170                 goto out;
2171
2172         mutex_lock(&bpf_module_mutex);
2173
2174         switch (op) {
2175         case MODULE_STATE_COMING:
2176                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2177                 if (btm) {
2178                         btm->module = module;
2179                         list_add(&btm->list, &bpf_trace_modules);
2180                 } else {
2181                         ret = -ENOMEM;
2182                 }
2183                 break;
2184         case MODULE_STATE_GOING:
2185                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2186                         if (btm->module == module) {
2187                                 list_del(&btm->list);
2188                                 kfree(btm);
2189                                 break;
2190                         }
2191                 }
2192                 break;
2193         }
2194
2195         mutex_unlock(&bpf_module_mutex);
2196
2197 out:
2198         return notifier_from_errno(ret);
2199 }
2200
2201 static struct notifier_block bpf_module_nb = {
2202         .notifier_call = bpf_event_notify,
2203 };
2204
2205 static int __init bpf_event_init(void)
2206 {
2207         register_module_notifier(&bpf_module_nb);
2208         return 0;
2209 }
2210
2211 fs_initcall(bpf_event_init);
2212 #endif /* CONFIG_MODULES */
2213
2214 #ifdef CONFIG_FPROBE
2215 struct bpf_kprobe_multi_link {
2216         struct bpf_link link;
2217         struct fprobe fp;
2218         unsigned long *addrs;
2219         u64 *cookies;
2220         u32 cnt;
2221 };
2222
2223 struct bpf_kprobe_multi_run_ctx {
2224         struct bpf_run_ctx run_ctx;
2225         struct bpf_kprobe_multi_link *link;
2226         unsigned long entry_ip;
2227 };
2228
2229 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2230 {
2231         struct bpf_kprobe_multi_link *kmulti_link;
2232
2233         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2234         unregister_fprobe(&kmulti_link->fp);
2235 }
2236
2237 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2238 {
2239         struct bpf_kprobe_multi_link *kmulti_link;
2240
2241         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2242         kvfree(kmulti_link->addrs);
2243         kvfree(kmulti_link->cookies);
2244         kfree(kmulti_link);
2245 }
2246
2247 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2248         .release = bpf_kprobe_multi_link_release,
2249         .dealloc = bpf_kprobe_multi_link_dealloc,
2250 };
2251
2252 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2253 {
2254         const struct bpf_kprobe_multi_link *link = priv;
2255         unsigned long *addr_a = a, *addr_b = b;
2256         u64 *cookie_a, *cookie_b;
2257         unsigned long tmp1;
2258         u64 tmp2;
2259
2260         cookie_a = link->cookies + (addr_a - link->addrs);
2261         cookie_b = link->cookies + (addr_b - link->addrs);
2262
2263         /* swap addr_a/addr_b and cookie_a/cookie_b values */
2264         tmp1 = *addr_a; *addr_a = *addr_b; *addr_b = tmp1;
2265         tmp2 = *cookie_a; *cookie_a = *cookie_b; *cookie_b = tmp2;
2266 }
2267
2268 static int __bpf_kprobe_multi_cookie_cmp(const void *a, const void *b)
2269 {
2270         const unsigned long *addr_a = a, *addr_b = b;
2271
2272         if (*addr_a == *addr_b)
2273                 return 0;
2274         return *addr_a < *addr_b ? -1 : 1;
2275 }
2276
2277 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2278 {
2279         return __bpf_kprobe_multi_cookie_cmp(a, b);
2280 }
2281
2282 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2283 {
2284         struct bpf_kprobe_multi_run_ctx *run_ctx;
2285         struct bpf_kprobe_multi_link *link;
2286         u64 *cookie, entry_ip;
2287         unsigned long *addr;
2288
2289         if (WARN_ON_ONCE(!ctx))
2290                 return 0;
2291         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2292         link = run_ctx->link;
2293         if (!link->cookies)
2294                 return 0;
2295         entry_ip = run_ctx->entry_ip;
2296         addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2297                        __bpf_kprobe_multi_cookie_cmp);
2298         if (!addr)
2299                 return 0;
2300         cookie = link->cookies + (addr - link->addrs);
2301         return *cookie;
2302 }
2303
2304 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2305 {
2306         struct bpf_kprobe_multi_run_ctx *run_ctx;
2307
2308         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2309         return run_ctx->entry_ip;
2310 }
2311
2312 static int
2313 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2314                            unsigned long entry_ip, struct pt_regs *regs)
2315 {
2316         struct bpf_kprobe_multi_run_ctx run_ctx = {
2317                 .link = link,
2318                 .entry_ip = entry_ip,
2319         };
2320         struct bpf_run_ctx *old_run_ctx;
2321         int err;
2322
2323         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2324                 err = 0;
2325                 goto out;
2326         }
2327
2328         migrate_disable();
2329         rcu_read_lock();
2330         old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2331         err = bpf_prog_run(link->link.prog, regs);
2332         bpf_reset_run_ctx(old_run_ctx);
2333         rcu_read_unlock();
2334         migrate_enable();
2335
2336  out:
2337         __this_cpu_dec(bpf_prog_active);
2338         return err;
2339 }
2340
2341 static void
2342 kprobe_multi_link_handler(struct fprobe *fp, unsigned long entry_ip,
2343                           struct pt_regs *regs)
2344 {
2345         struct bpf_kprobe_multi_link *link;
2346
2347         link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2348         kprobe_multi_link_prog_run(link, entry_ip, regs);
2349 }
2350
2351 static int
2352 kprobe_multi_resolve_syms(const void __user *usyms, u32 cnt,
2353                           unsigned long *addrs)
2354 {
2355         unsigned long addr, size;
2356         const char __user **syms;
2357         int err = -ENOMEM;
2358         unsigned int i;
2359         char *func;
2360
2361         size = cnt * sizeof(*syms);
2362         syms = kvzalloc(size, GFP_KERNEL);
2363         if (!syms)
2364                 return -ENOMEM;
2365
2366         func = kmalloc(KSYM_NAME_LEN, GFP_KERNEL);
2367         if (!func)
2368                 goto error;
2369
2370         if (copy_from_user(syms, usyms, size)) {
2371                 err = -EFAULT;
2372                 goto error;
2373         }
2374
2375         for (i = 0; i < cnt; i++) {
2376                 err = strncpy_from_user(func, syms[i], KSYM_NAME_LEN);
2377                 if (err == KSYM_NAME_LEN)
2378                         err = -E2BIG;
2379                 if (err < 0)
2380                         goto error;
2381                 err = -EINVAL;
2382                 addr = kallsyms_lookup_name(func);
2383                 if (!addr)
2384                         goto error;
2385                 if (!kallsyms_lookup_size_offset(addr, &size, NULL))
2386                         goto error;
2387                 addr = ftrace_location_range(addr, addr + size - 1);
2388                 if (!addr)
2389                         goto error;
2390                 addrs[i] = addr;
2391         }
2392
2393         err = 0;
2394 error:
2395         kvfree(syms);
2396         kfree(func);
2397         return err;
2398 }
2399
2400 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2401 {
2402         struct bpf_kprobe_multi_link *link = NULL;
2403         struct bpf_link_primer link_primer;
2404         void __user *ucookies;
2405         unsigned long *addrs;
2406         u32 flags, cnt, size;
2407         void __user *uaddrs;
2408         u64 *cookies = NULL;
2409         void __user *usyms;
2410         int err;
2411
2412         /* no support for 32bit archs yet */
2413         if (sizeof(u64) != sizeof(void *))
2414                 return -EOPNOTSUPP;
2415
2416         if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2417                 return -EINVAL;
2418
2419         flags = attr->link_create.kprobe_multi.flags;
2420         if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2421                 return -EINVAL;
2422
2423         uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2424         usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2425         if (!!uaddrs == !!usyms)
2426                 return -EINVAL;
2427
2428         cnt = attr->link_create.kprobe_multi.cnt;
2429         if (!cnt)
2430                 return -EINVAL;
2431
2432         size = cnt * sizeof(*addrs);
2433         addrs = kvmalloc(size, GFP_KERNEL);
2434         if (!addrs)
2435                 return -ENOMEM;
2436
2437         if (uaddrs) {
2438                 if (copy_from_user(addrs, uaddrs, size)) {
2439                         err = -EFAULT;
2440                         goto error;
2441                 }
2442         } else {
2443                 err = kprobe_multi_resolve_syms(usyms, cnt, addrs);
2444                 if (err)
2445                         goto error;
2446         }
2447
2448         ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2449         if (ucookies) {
2450                 cookies = kvmalloc(size, GFP_KERNEL);
2451                 if (!cookies) {
2452                         err = -ENOMEM;
2453                         goto error;
2454                 }
2455                 if (copy_from_user(cookies, ucookies, size)) {
2456                         err = -EFAULT;
2457                         goto error;
2458                 }
2459         }
2460
2461         link = kzalloc(sizeof(*link), GFP_KERNEL);
2462         if (!link) {
2463                 err = -ENOMEM;
2464                 goto error;
2465         }
2466
2467         bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2468                       &bpf_kprobe_multi_link_lops, prog);
2469
2470         err = bpf_link_prime(&link->link, &link_primer);
2471         if (err)
2472                 goto error;
2473
2474         if (flags & BPF_F_KPROBE_MULTI_RETURN)
2475                 link->fp.exit_handler = kprobe_multi_link_handler;
2476         else
2477                 link->fp.entry_handler = kprobe_multi_link_handler;
2478
2479         link->addrs = addrs;
2480         link->cookies = cookies;
2481         link->cnt = cnt;
2482
2483         if (cookies) {
2484                 /*
2485                  * Sorting addresses will trigger sorting cookies as well
2486                  * (check bpf_kprobe_multi_cookie_swap). This way we can
2487                  * find cookie based on the address in bpf_get_attach_cookie
2488                  * helper.
2489                  */
2490                 sort_r(addrs, cnt, sizeof(*addrs),
2491                        bpf_kprobe_multi_cookie_cmp,
2492                        bpf_kprobe_multi_cookie_swap,
2493                        link);
2494         }
2495
2496         err = register_fprobe_ips(&link->fp, addrs, cnt);
2497         if (err) {
2498                 bpf_link_cleanup(&link_primer);
2499                 return err;
2500         }
2501
2502         return bpf_link_settle(&link_primer);
2503
2504 error:
2505         kfree(link);
2506         kvfree(addrs);
2507         kvfree(cookies);
2508         return err;
2509 }
2510 #else /* !CONFIG_FPROBE */
2511 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2512 {
2513         return -EOPNOTSUPP;
2514 }
2515 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2516 {
2517         return 0;
2518 }
2519 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2520 {
2521         return 0;
2522 }
2523 #endif