bpf_trace.c

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