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_verifier.h>
  10 #include <linux/bpf_perf_event.h>
  11 #include <linux/btf.h>
  12 #include <linux/filter.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/ctype.h>
  15 #include <linux/kprobes.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/syscalls.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/btf_ids.h>
  20 #include <linux/bpf_lsm.h>
  21 #include <linux/fprobe.h>
  22 #include <linux/bsearch.h>
  23 #include <linux/sort.h>
  24 #include <linux/key.h>
  25 #include <linux/verification.h>
  26 #include <linux/namei.h>
  27
  28 #include <net/bpf_sk_storage.h>
  29
  30 #include <uapi/linux/bpf.h>
  31 #include <uapi/linux/btf.h>
  32
  33 #include <asm/tlb.h>
  34
  35 #include "trace_probe.h"
  36 #include "trace.h"
  37
  38 #define CREATE_TRACE_POINTS
  39 #include "bpf_trace.h"
  40
  41 #define bpf_event_rcu_dereference(p)                                    \
  42         rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  43
  44 #ifdef CONFIG_MODULES
  45 struct bpf_trace_module {
  46         struct module *module;
  47         struct list_head list;
  48 };
  49
  50 static LIST_HEAD(bpf_trace_modules);
  51 static DEFINE_MUTEX(bpf_module_mutex);
  52
  53 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  54 {
  55         struct bpf_raw_event_map *btp, *ret = NULL;
  56         struct bpf_trace_module *btm;
  57         unsigned int i;
  58
  59         mutex_lock(&bpf_module_mutex);
  60         list_for_each_entry(btm, &bpf_trace_modules, list) {
  61                 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
  62                         btp = &btm->module->bpf_raw_events[i];
  63                         if (!strcmp(btp->tp->name, name)) {
  64                                 if (try_module_get(btm->module))
  65                                         ret = btp;
  66                                 goto out;
  67                         }
  68                 }
  69         }
  70 out:
  71         mutex_unlock(&bpf_module_mutex);
  72         return ret;
  73 }
  74 #else
  75 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  76 {
  77         return NULL;
  78 }
  79 #endif /* CONFIG_MODULES */
  80
  81 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  82 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  83
  84 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
  85                                   u64 flags, const struct btf **btf,
  86                                   s32 *btf_id);
  87 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
  88 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
  89
  90 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx);
  91 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
  92
  93 /**
  94  * trace_call_bpf - invoke BPF program
  95  * @call: tracepoint event
  96  * @ctx: opaque context pointer
  97  *
  98  * kprobe handlers execute BPF programs via this helper.
  99  * Can be used from static tracepoints in the future.
 100  *
 101  * Return: BPF programs always return an integer which is interpreted by
 102  * kprobe handler as:
 103  * 0 - return from kprobe (event is filtered out)
 104  * 1 - store kprobe event into ring buffer
 105  * Other values are reserved and currently alias to 1
 106  */
 107 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 108 {
 109         unsigned int ret;
 110
 111         cant_sleep();
 112
 113         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 114                 /*
 115                  * since some bpf program is already running on this cpu,
 116                  * don't call into another bpf program (same or different)
 117                  * and don't send kprobe event into ring-buffer,
 118                  * so return zero here
 119                  */
 120                 rcu_read_lock();
 121                 bpf_prog_inc_misses_counters(rcu_dereference(call->prog_array));
 122                 rcu_read_unlock();
 123                 ret = 0;
 124                 goto out;
 125         }
 126
 127         /*
 128          * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
 129          * to all call sites, we did a bpf_prog_array_valid() there to check
 130          * whether call->prog_array is empty or not, which is
 131          * a heuristic to speed up execution.
 132          *
 133          * If bpf_prog_array_valid() fetched prog_array was
 134          * non-NULL, we go into trace_call_bpf() and do the actual
 135          * proper rcu_dereference() under RCU lock.
 136          * If it turns out that prog_array is NULL then, we bail out.
 137          * For the opposite, if the bpf_prog_array_valid() fetched pointer
 138          * was NULL, you'll skip the prog_array with the risk of missing
 139          * out of events when it was updated in between this and the
 140          * rcu_dereference() which is accepted risk.
 141          */
 142         rcu_read_lock();
 143         ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
 144                                  ctx, bpf_prog_run);
 145         rcu_read_unlock();
 146
 147  out:
 148         __this_cpu_dec(bpf_prog_active);
 149
 150         return ret;
 151 }
 152
 153 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 154 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 155 {
 156         regs_set_return_value(regs, rc);
 157         override_function_with_return(regs);
 158         return 0;
 159 }
 160
 161 static const struct bpf_func_proto bpf_override_return_proto = {
 162         .func           = bpf_override_return,
 163         .gpl_only       = true,
 164         .ret_type       = RET_INTEGER,
 165         .arg1_type      = ARG_PTR_TO_CTX,
 166         .arg2_type      = ARG_ANYTHING,
 167 };
 168 #endif
 169
 170 static __always_inline int
 171 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 172 {
 173         int ret;
 174
 175         ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 176         if (unlikely(ret < 0))
 177                 memset(dst, 0, size);
 178         return ret;
 179 }
 180
 181 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
 182            const void __user *, unsafe_ptr)
 183 {
 184         return bpf_probe_read_user_common(dst, size, unsafe_ptr);
 185 }
 186
 187 const struct bpf_func_proto bpf_probe_read_user_proto = {
 188         .func           = bpf_probe_read_user,
 189         .gpl_only       = true,
 190         .ret_type       = RET_INTEGER,
 191         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 192         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 193         .arg3_type      = ARG_ANYTHING,
 194 };
 195
 196 static __always_inline int
 197 bpf_probe_read_user_str_common(void *dst, u32 size,
 198                                const void __user *unsafe_ptr)
 199 {
 200         int ret;
 201
 202         /*
 203          * NB: We rely on strncpy_from_user() not copying junk past the NUL
 204          * terminator into `dst`.
 205          *
 206          * strncpy_from_user() does long-sized strides in the fast path. If the
 207          * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
 208          * then there could be junk after the NUL in `dst`. If user takes `dst`
 209          * and keys a hash map with it, then semantically identical strings can
 210          * occupy multiple entries in the map.
 211          */
 212         ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 213         if (unlikely(ret < 0))
 214                 memset(dst, 0, size);
 215         return ret;
 216 }
 217
 218 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
 219            const void __user *, unsafe_ptr)
 220 {
 221         return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
 222 }
 223
 224 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 225         .func           = bpf_probe_read_user_str,
 226         .gpl_only       = true,
 227         .ret_type       = RET_INTEGER,
 228         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 229         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 230         .arg3_type      = ARG_ANYTHING,
 231 };
 232
 233 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 234            const void *, unsafe_ptr)
 235 {
 236         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 237 }
 238
 239 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 240         .func           = bpf_probe_read_kernel,
 241         .gpl_only       = true,
 242         .ret_type       = RET_INTEGER,
 243         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 244         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 245         .arg3_type      = ARG_ANYTHING,
 246 };
 247
 248 static __always_inline int
 249 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
 250 {
 251         int ret;
 252
 253         /*
 254          * The strncpy_from_kernel_nofault() call will likely not fill the
 255          * entire buffer, but that's okay in this circumstance as we're probing
 256          * arbitrary memory anyway similar to bpf_probe_read_*() and might
 257          * as well probe the stack. Thus, memory is explicitly cleared
 258          * only in error case, so that improper users ignoring return
 259          * code altogether don't copy garbage; otherwise length of string
 260          * is returned that can be used for bpf_perf_event_output() et al.
 261          */
 262         ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
 263         if (unlikely(ret < 0))
 264                 memset(dst, 0, size);
 265         return ret;
 266 }
 267
 268 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 269            const void *, unsafe_ptr)
 270 {
 271         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 272 }
 273
 274 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 275         .func           = bpf_probe_read_kernel_str,
 276         .gpl_only       = true,
 277         .ret_type       = RET_INTEGER,
 278         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 279         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 280         .arg3_type      = ARG_ANYTHING,
 281 };
 282
 283 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 284 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
 285            const void *, unsafe_ptr)
 286 {
 287         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 288                 return bpf_probe_read_user_common(dst, size,
 289                                 (__force void __user *)unsafe_ptr);
 290         }
 291         return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 292 }
 293
 294 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
 295         .func           = bpf_probe_read_compat,
 296         .gpl_only       = true,
 297         .ret_type       = RET_INTEGER,
 298         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 299         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 300         .arg3_type      = ARG_ANYTHING,
 301 };
 302
 303 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 304            const void *, unsafe_ptr)
 305 {
 306         if ((unsigned long)unsafe_ptr < TASK_SIZE) {
 307                 return bpf_probe_read_user_str_common(dst, size,
 308                                 (__force void __user *)unsafe_ptr);
 309         }
 310         return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 311 }
 312
 313 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 314         .func           = bpf_probe_read_compat_str,
 315         .gpl_only       = true,
 316         .ret_type       = RET_INTEGER,
 317         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
 318         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
 319         .arg3_type      = ARG_ANYTHING,
 320 };
 321 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 322
 323 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 324            u32, size)
 325 {
 326         /*
 327          * Ensure we're in user context which is safe for the helper to
 328          * run. This helper has no business in a kthread.
 329          *
 330          * access_ok() should prevent writing to non-user memory, but in
 331          * some situations (nommu, temporary switch, etc) access_ok() does
 332          * not provide enough validation, hence the check on KERNEL_DS.
 333          *
 334          * nmi_uaccess_okay() ensures the probe is not run in an interim
 335          * state, when the task or mm are switched. This is specifically
 336          * required to prevent the use of temporary mm.
 337          */
 338
 339         if (unlikely(in_interrupt() ||
 340                      current->flags & (PF_KTHREAD | PF_EXITING)))
 341                 return -EPERM;
 342         if (unlikely(!nmi_uaccess_okay()))
 343                 return -EPERM;
 344
 345         return copy_to_user_nofault(unsafe_ptr, src, size);
 346 }
 347
 348 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 349         .func           = bpf_probe_write_user,
 350         .gpl_only       = true,
 351         .ret_type       = RET_INTEGER,
 352         .arg1_type      = ARG_ANYTHING,
 353         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 354         .arg3_type      = ARG_CONST_SIZE,
 355 };
 356
 357 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 358 {
 359         if (!capable(CAP_SYS_ADMIN))
 360                 return NULL;
 361
 362         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 363                             current->comm, task_pid_nr(current));
 364
 365         return &bpf_probe_write_user_proto;
 366 }
 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         struct bpf_bprintf_data data = {
 376                 .get_bin_args   = true,
 377                 .get_buf        = true,
 378         };
 379         int ret;
 380
 381         ret = bpf_bprintf_prepare(fmt, fmt_size, args,
 382                                   MAX_TRACE_PRINTK_VARARGS, &data);
 383         if (ret < 0)
 384                 return ret;
 385
 386         ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
 387
 388         trace_bpf_trace_printk(data.buf);
 389
 390         bpf_bprintf_cleanup(&data);
 391
 392         return ret;
 393 }
 394
 395 static const struct bpf_func_proto bpf_trace_printk_proto = {
 396         .func           = bpf_trace_printk,
 397         .gpl_only       = true,
 398         .ret_type       = RET_INTEGER,
 399         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 400         .arg2_type      = ARG_CONST_SIZE,
 401 };
 402
 403 static void __set_printk_clr_event(void)
 404 {
 405         /*
 406          * This program might be calling bpf_trace_printk,
 407          * so enable the associated bpf_trace/bpf_trace_printk event.
 408          * Repeat this each time as it is possible a user has
 409          * disabled bpf_trace_printk events.  By loading a program
 410          * calling bpf_trace_printk() however the user has expressed
 411          * the intent to see such events.
 412          */
 413         if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 414                 pr_warn_ratelimited("could not enable bpf_trace_printk events");
 415 }
 416
 417 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 418 {
 419         __set_printk_clr_event();
 420         return &bpf_trace_printk_proto;
 421 }
 422
 423 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, args,
 424            u32, data_len)
 425 {
 426         struct bpf_bprintf_data data = {
 427                 .get_bin_args   = true,
 428                 .get_buf        = true,
 429         };
 430         int ret, num_args;
 431
 432         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 433             (data_len && !args))
 434                 return -EINVAL;
 435         num_args = data_len / 8;
 436
 437         ret = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
 438         if (ret < 0)
 439                 return ret;
 440
 441         ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
 442
 443         trace_bpf_trace_printk(data.buf);
 444
 445         bpf_bprintf_cleanup(&data);
 446
 447         return ret;
 448 }
 449
 450 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
 451         .func           = bpf_trace_vprintk,
 452         .gpl_only       = true,
 453         .ret_type       = RET_INTEGER,
 454         .arg1_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 455         .arg2_type      = ARG_CONST_SIZE,
 456         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
 457         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
 458 };
 459
 460 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
 461 {
 462         __set_printk_clr_event();
 463         return &bpf_trace_vprintk_proto;
 464 }
 465
 466 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 467            const void *, args, u32, data_len)
 468 {
 469         struct bpf_bprintf_data data = {
 470                 .get_bin_args   = true,
 471         };
 472         int err, num_args;
 473
 474         if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
 475             (data_len && !args))
 476                 return -EINVAL;
 477         num_args = data_len / 8;
 478
 479         err = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
 480         if (err < 0)
 481                 return err;
 482
 483         seq_bprintf(m, fmt, data.bin_args);
 484
 485         bpf_bprintf_cleanup(&data);
 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;
 661         struct perf_raw_record raw = {
 662                 .frag = {
 663                         .size = size,
 664                         .data = data,
 665                 },
 666         };
 667         struct perf_sample_data *sd;
 668         int nest_level, err;
 669
 670         preempt_disable();
 671         sds = this_cpu_ptr(&bpf_trace_sds);
 672         nest_level = this_cpu_inc_return(bpf_trace_nest_level);
 673
 674         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
 675                 err = -EBUSY;
 676                 goto out;
 677         }
 678
 679         sd = &sds->sds[nest_level - 1];
 680
 681         if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
 682                 err = -EINVAL;
 683                 goto out;
 684         }
 685
 686         perf_sample_data_init(sd, 0, 0);
 687         perf_sample_save_raw_data(sd, &raw);
 688
 689         err = __bpf_perf_event_output(regs, map, flags, sd);
 690 out:
 691         this_cpu_dec(bpf_trace_nest_level);
 692         preempt_enable();
 693         return err;
 694 }
 695
 696 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 697         .func           = bpf_perf_event_output,
 698         .gpl_only       = true,
 699         .ret_type       = RET_INTEGER,
 700         .arg1_type      = ARG_PTR_TO_CTX,
 701         .arg2_type      = ARG_CONST_MAP_PTR,
 702         .arg3_type      = ARG_ANYTHING,
 703         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
 704         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
 705 };
 706
 707 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 708 struct bpf_nested_pt_regs {
 709         struct pt_regs regs[3];
 710 };
 711 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 712 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 713
 714 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 715                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 716 {
 717         struct perf_raw_frag frag = {
 718                 .copy           = ctx_copy,
 719                 .size           = ctx_size,
 720                 .data           = ctx,
 721         };
 722         struct perf_raw_record raw = {
 723                 .frag = {
 724                         {
 725                                 .next   = ctx_size ? &frag : NULL,
 726                         },
 727                         .size   = meta_size,
 728                         .data   = meta,
 729                 },
 730         };
 731         struct perf_sample_data *sd;
 732         struct pt_regs *regs;
 733         int nest_level;
 734         u64 ret;
 735
 736         preempt_disable();
 737         nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 738
 739         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
 740                 ret = -EBUSY;
 741                 goto out;
 742         }
 743         sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
 744         regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
 745
 746         perf_fetch_caller_regs(regs);
 747         perf_sample_data_init(sd, 0, 0);
 748         perf_sample_save_raw_data(sd, &raw);
 749
 750         ret = __bpf_perf_event_output(regs, map, flags, sd);
 751 out:
 752         this_cpu_dec(bpf_event_output_nest_level);
 753         preempt_enable();
 754         return ret;
 755 }
 756
 757 BPF_CALL_0(bpf_get_current_task)
 758 {
 759         return (long) current;
 760 }
 761
 762 const struct bpf_func_proto bpf_get_current_task_proto = {
 763         .func           = bpf_get_current_task,
 764         .gpl_only       = true,
 765         .ret_type       = RET_INTEGER,
 766 };
 767
 768 BPF_CALL_0(bpf_get_current_task_btf)
 769 {
 770         return (unsigned long) current;
 771 }
 772
 773 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
 774         .func           = bpf_get_current_task_btf,
 775         .gpl_only       = true,
 776         .ret_type       = RET_PTR_TO_BTF_ID_TRUSTED,
 777         .ret_btf_id     = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 778 };
 779
 780 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
 781 {
 782         return (unsigned long) task_pt_regs(task);
 783 }
 784
 785 BTF_ID_LIST(bpf_task_pt_regs_ids)
 786 BTF_ID(struct, pt_regs)
 787
 788 const struct bpf_func_proto bpf_task_pt_regs_proto = {
 789         .func           = bpf_task_pt_regs,
 790         .gpl_only       = true,
 791         .arg1_type      = ARG_PTR_TO_BTF_ID,
 792         .arg1_btf_id    = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
 793         .ret_type       = RET_PTR_TO_BTF_ID,
 794         .ret_btf_id     = &bpf_task_pt_regs_ids[0],
 795 };
 796
 797 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 798 {
 799         struct bpf_array *array = container_of(map, struct bpf_array, map);
 800         struct cgroup *cgrp;
 801
 802         if (unlikely(idx >= array->map.max_entries))
 803                 return -E2BIG;
 804
 805         cgrp = READ_ONCE(array->ptrs[idx]);
 806         if (unlikely(!cgrp))
 807                 return -EAGAIN;
 808
 809         return task_under_cgroup_hierarchy(current, cgrp);
 810 }
 811
 812 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 813         .func           = bpf_current_task_under_cgroup,
 814         .gpl_only       = false,
 815         .ret_type       = RET_INTEGER,
 816         .arg1_type      = ARG_CONST_MAP_PTR,
 817         .arg2_type      = ARG_ANYTHING,
 818 };
 819
 820 struct send_signal_irq_work {
 821         struct irq_work irq_work;
 822         struct task_struct *task;
 823         u32 sig;
 824         enum pid_type type;
 825 };
 826
 827 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 828
 829 static void do_bpf_send_signal(struct irq_work *entry)
 830 {
 831         struct send_signal_irq_work *work;
 832
 833         work = container_of(entry, struct send_signal_irq_work, irq_work);
 834         group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 835         put_task_struct(work->task);
 836 }
 837
 838 static int bpf_send_signal_common(u32 sig, enum pid_type type)
 839 {
 840         struct send_signal_irq_work *work = NULL;
 841
 842         /* Similar to bpf_probe_write_user, task needs to be
 843          * in a sound condition and kernel memory access be
 844          * permitted in order to send signal to the current
 845          * task.
 846          */
 847         if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 848                 return -EPERM;
 849         if (unlikely(!nmi_uaccess_okay()))
 850                 return -EPERM;
 851         /* Task should not be pid=1 to avoid kernel panic. */
 852         if (unlikely(is_global_init(current)))
 853                 return -EPERM;
 854
 855         if (irqs_disabled()) {
 856                 /* Do an early check on signal validity. Otherwise,
 857                  * the error is lost in deferred irq_work.
 858                  */
 859                 if (unlikely(!valid_signal(sig)))
 860                         return -EINVAL;
 861
 862                 work = this_cpu_ptr(&send_signal_work);
 863                 if (irq_work_is_busy(&work->irq_work))
 864                         return -EBUSY;
 865
 866                 /* Add the current task, which is the target of sending signal,
 867                  * to the irq_work. The current task may change when queued
 868                  * irq works get executed.
 869                  */
 870                 work->task = get_task_struct(current);
 871                 work->sig = sig;
 872                 work->type = type;
 873                 irq_work_queue(&work->irq_work);
 874                 return 0;
 875         }
 876
 877         return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 878 }
 879
 880 BPF_CALL_1(bpf_send_signal, u32, sig)
 881 {
 882         return bpf_send_signal_common(sig, PIDTYPE_TGID);
 883 }
 884
 885 static const struct bpf_func_proto bpf_send_signal_proto = {
 886         .func           = bpf_send_signal,
 887         .gpl_only       = false,
 888         .ret_type       = RET_INTEGER,
 889         .arg1_type      = ARG_ANYTHING,
 890 };
 891
 892 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 893 {
 894         return bpf_send_signal_common(sig, PIDTYPE_PID);
 895 }
 896
 897 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 898         .func           = bpf_send_signal_thread,
 899         .gpl_only       = false,
 900         .ret_type       = RET_INTEGER,
 901         .arg1_type      = ARG_ANYTHING,
 902 };
 903
 904 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 905 {
 906         struct path copy;
 907         long len;
 908         char *p;
 909
 910         if (!sz)
 911                 return 0;
 912
 913         /*
 914          * The path pointer is verified as trusted and safe to use,
 915          * but let's double check it's valid anyway to workaround
 916          * potentially broken verifier.
 917          */
 918         len = copy_from_kernel_nofault(&copy, path, sizeof(*path));
 919         if (len < 0)
 920                 return len;
 921
 922         p = d_path(&copy, buf, sz);
 923         if (IS_ERR(p)) {
 924                 len = PTR_ERR(p);
 925         } else {
 926                 len = buf + sz - p;
 927                 memmove(buf, p, len);
 928         }
 929
 930         return len;
 931 }
 932
 933 BTF_SET_START(btf_allowlist_d_path)
 934 #ifdef CONFIG_SECURITY
 935 BTF_ID(func, security_file_permission)
 936 BTF_ID(func, security_inode_getattr)
 937 BTF_ID(func, security_file_open)
 938 #endif
 939 #ifdef CONFIG_SECURITY_PATH
 940 BTF_ID(func, security_path_truncate)
 941 #endif
 942 BTF_ID(func, vfs_truncate)
 943 BTF_ID(func, vfs_fallocate)
 944 BTF_ID(func, dentry_open)
 945 BTF_ID(func, vfs_getattr)
 946 BTF_ID(func, filp_close)
 947 BTF_SET_END(btf_allowlist_d_path)
 948
 949 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 950 {
 951         if (prog->type == BPF_PROG_TYPE_TRACING &&
 952             prog->expected_attach_type == BPF_TRACE_ITER)
 953                 return true;
 954
 955         if (prog->type == BPF_PROG_TYPE_LSM)
 956                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 957
 958         return btf_id_set_contains(&btf_allowlist_d_path,
 959                                    prog->aux->attach_btf_id);
 960 }
 961
 962 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 963
 964 static const struct bpf_func_proto bpf_d_path_proto = {
 965         .func           = bpf_d_path,
 966         .gpl_only       = false,
 967         .ret_type       = RET_INTEGER,
 968         .arg1_type      = ARG_PTR_TO_BTF_ID,
 969         .arg1_btf_id    = &bpf_d_path_btf_ids[0],
 970         .arg2_type      = ARG_PTR_TO_MEM,
 971         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
 972         .allowed        = bpf_d_path_allowed,
 973 };
 974
 975 #define BTF_F_ALL       (BTF_F_COMPACT  | BTF_F_NONAME | \
 976                          BTF_F_PTR_RAW | BTF_F_ZERO)
 977
 978 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 979                                   u64 flags, const struct btf **btf,
 980                                   s32 *btf_id)
 981 {
 982         const struct btf_type *t;
 983
 984         if (unlikely(flags & ~(BTF_F_ALL)))
 985                 return -EINVAL;
 986
 987         if (btf_ptr_size != sizeof(struct btf_ptr))
 988                 return -EINVAL;
 989
 990         *btf = bpf_get_btf_vmlinux();
 991
 992         if (IS_ERR_OR_NULL(*btf))
 993                 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 994
 995         if (ptr->type_id > 0)
 996                 *btf_id = ptr->type_id;
 997         else
 998                 return -EINVAL;
 999
1000         if (*btf_id > 0)
1001                 t = btf_type_by_id(*btf, *btf_id);
1002         if (*btf_id <= 0 || !t)
1003                 return -ENOENT;
1004
1005         return 0;
1006 }
1007
1008 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1009            u32, btf_ptr_size, u64, flags)
1010 {
1011         const struct btf *btf;
1012         s32 btf_id;
1013         int ret;
1014
1015         ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1016         if (ret)
1017                 return ret;
1018
1019         return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1020                                       flags);
1021 }
1022
1023 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1024         .func           = bpf_snprintf_btf,
1025         .gpl_only       = false,
1026         .ret_type       = RET_INTEGER,
1027         .arg1_type      = ARG_PTR_TO_MEM,
1028         .arg2_type      = ARG_CONST_SIZE,
1029         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1030         .arg4_type      = ARG_CONST_SIZE,
1031         .arg5_type      = ARG_ANYTHING,
1032 };
1033
1034 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1035 {
1036         /* This helper call is inlined by verifier. */
1037         return ((u64 *)ctx)[-2];
1038 }
1039
1040 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1041         .func           = bpf_get_func_ip_tracing,
1042         .gpl_only       = true,
1043         .ret_type       = RET_INTEGER,
1044         .arg1_type      = ARG_PTR_TO_CTX,
1045 };
1046
1047 #ifdef CONFIG_X86_KERNEL_IBT
1048 static unsigned long get_entry_ip(unsigned long fentry_ip)
1049 {
1050         u32 instr;
1051
1052         /* Being extra safe in here in case entry ip is on the page-edge. */
1053         if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1))
1054                 return fentry_ip;
1055         if (is_endbr(instr))
1056                 fentry_ip -= ENDBR_INSN_SIZE;
1057         return fentry_ip;
1058 }
1059 #else
1060 #define get_entry_ip(fentry_ip) fentry_ip
1061 #endif
1062
1063 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1064 {
1065         struct bpf_trace_run_ctx *run_ctx __maybe_unused;
1066         struct kprobe *kp;
1067
1068 #ifdef CONFIG_UPROBES
1069         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1070         if (run_ctx->is_uprobe)
1071                 return ((struct uprobe_dispatch_data *)current->utask->vaddr)->bp_addr;
1072 #endif
1073
1074         kp = kprobe_running();
1075
1076         if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1077                 return 0;
1078
1079         return get_entry_ip((uintptr_t)kp->addr);
1080 }
1081
1082 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1083         .func           = bpf_get_func_ip_kprobe,
1084         .gpl_only       = true,
1085         .ret_type       = RET_INTEGER,
1086         .arg1_type      = ARG_PTR_TO_CTX,
1087 };
1088
1089 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1090 {
1091         return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1092 }
1093
1094 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1095         .func           = bpf_get_func_ip_kprobe_multi,
1096         .gpl_only       = false,
1097         .ret_type       = RET_INTEGER,
1098         .arg1_type      = ARG_PTR_TO_CTX,
1099 };
1100
1101 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1102 {
1103         return bpf_kprobe_multi_cookie(current->bpf_ctx);
1104 }
1105
1106 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1107         .func           = bpf_get_attach_cookie_kprobe_multi,
1108         .gpl_only       = false,
1109         .ret_type       = RET_INTEGER,
1110         .arg1_type      = ARG_PTR_TO_CTX,
1111 };
1112
1113 BPF_CALL_1(bpf_get_func_ip_uprobe_multi, struct pt_regs *, regs)
1114 {
1115         return bpf_uprobe_multi_entry_ip(current->bpf_ctx);
1116 }
1117
1118 static const struct bpf_func_proto bpf_get_func_ip_proto_uprobe_multi = {
1119         .func           = bpf_get_func_ip_uprobe_multi,
1120         .gpl_only       = false,
1121         .ret_type       = RET_INTEGER,
1122         .arg1_type      = ARG_PTR_TO_CTX,
1123 };
1124
1125 BPF_CALL_1(bpf_get_attach_cookie_uprobe_multi, struct pt_regs *, regs)
1126 {
1127         return bpf_uprobe_multi_cookie(current->bpf_ctx);
1128 }
1129
1130 static const struct bpf_func_proto bpf_get_attach_cookie_proto_umulti = {
1131         .func           = bpf_get_attach_cookie_uprobe_multi,
1132         .gpl_only       = false,
1133         .ret_type       = RET_INTEGER,
1134         .arg1_type      = ARG_PTR_TO_CTX,
1135 };
1136
1137 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1138 {
1139         struct bpf_trace_run_ctx *run_ctx;
1140
1141         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1142         return run_ctx->bpf_cookie;
1143 }
1144
1145 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1146         .func           = bpf_get_attach_cookie_trace,
1147         .gpl_only       = false,
1148         .ret_type       = RET_INTEGER,
1149         .arg1_type      = ARG_PTR_TO_CTX,
1150 };
1151
1152 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1153 {
1154         return ctx->event->bpf_cookie;
1155 }
1156
1157 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1158         .func           = bpf_get_attach_cookie_pe,
1159         .gpl_only       = false,
1160         .ret_type       = RET_INTEGER,
1161         .arg1_type      = ARG_PTR_TO_CTX,
1162 };
1163
1164 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1165 {
1166         struct bpf_trace_run_ctx *run_ctx;
1167
1168         run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1169         return run_ctx->bpf_cookie;
1170 }
1171
1172 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1173         .func           = bpf_get_attach_cookie_tracing,
1174         .gpl_only       = false,
1175         .ret_type       = RET_INTEGER,
1176         .arg1_type      = ARG_PTR_TO_CTX,
1177 };
1178
1179 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1180 {
1181 #ifndef CONFIG_X86
1182         return -ENOENT;
1183 #else
1184         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1185         u32 entry_cnt = size / br_entry_size;
1186
1187         entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1188
1189         if (unlikely(flags))
1190                 return -EINVAL;
1191
1192         if (!entry_cnt)
1193                 return -ENOENT;
1194
1195         return entry_cnt * br_entry_size;
1196 #endif
1197 }
1198
1199 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1200         .func           = bpf_get_branch_snapshot,
1201         .gpl_only       = true,
1202         .ret_type       = RET_INTEGER,
1203         .arg1_type      = ARG_PTR_TO_UNINIT_MEM,
1204         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1205 };
1206
1207 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1208 {
1209         /* This helper call is inlined by verifier. */
1210         u64 nr_args = ((u64 *)ctx)[-1];
1211
1212         if ((u64) n >= nr_args)
1213                 return -EINVAL;
1214         *value = ((u64 *)ctx)[n];
1215         return 0;
1216 }
1217
1218 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1219         .func           = get_func_arg,
1220         .ret_type       = RET_INTEGER,
1221         .arg1_type      = ARG_PTR_TO_CTX,
1222         .arg2_type      = ARG_ANYTHING,
1223         .arg3_type      = ARG_PTR_TO_LONG,
1224 };
1225
1226 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1227 {
1228         /* This helper call is inlined by verifier. */
1229         u64 nr_args = ((u64 *)ctx)[-1];
1230
1231         *value = ((u64 *)ctx)[nr_args];
1232         return 0;
1233 }
1234
1235 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1236         .func           = get_func_ret,
1237         .ret_type       = RET_INTEGER,
1238         .arg1_type      = ARG_PTR_TO_CTX,
1239         .arg2_type      = ARG_PTR_TO_LONG,
1240 };
1241
1242 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1243 {
1244         /* This helper call is inlined by verifier. */
1245         return ((u64 *)ctx)[-1];
1246 }
1247
1248 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1249         .func           = get_func_arg_cnt,
1250         .ret_type       = RET_INTEGER,
1251         .arg1_type      = ARG_PTR_TO_CTX,
1252 };
1253
1254 #ifdef CONFIG_KEYS
1255 __bpf_kfunc_start_defs();
1256
1257 /**
1258  * bpf_lookup_user_key - lookup a key by its serial
1259  * @serial: key handle serial number
1260  * @flags: lookup-specific flags
1261  *
1262  * Search a key with a given *serial* and the provided *flags*.
1263  * If found, increment the reference count of the key by one, and
1264  * return it in the bpf_key structure.
1265  *
1266  * The bpf_key structure must be passed to bpf_key_put() when done
1267  * with it, so that the key reference count is decremented and the
1268  * bpf_key structure is freed.
1269  *
1270  * Permission checks are deferred to the time the key is used by
1271  * one of the available key-specific kfuncs.
1272  *
1273  * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1274  * special keyring (e.g. session keyring), if it doesn't yet exist.
1275  * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1276  * for the key construction, and to retrieve uninstantiated keys (keys
1277  * without data attached to them).
1278  *
1279  * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1280  *         NULL pointer otherwise.
1281  */
1282 __bpf_kfunc struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1283 {
1284         key_ref_t key_ref;
1285         struct bpf_key *bkey;
1286
1287         if (flags & ~KEY_LOOKUP_ALL)
1288                 return NULL;
1289
1290         /*
1291          * Permission check is deferred until the key is used, as the
1292          * intent of the caller is unknown here.
1293          */
1294         key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1295         if (IS_ERR(key_ref))
1296                 return NULL;
1297
1298         bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1299         if (!bkey) {
1300                 key_put(key_ref_to_ptr(key_ref));
1301                 return NULL;
1302         }
1303
1304         bkey->key = key_ref_to_ptr(key_ref);
1305         bkey->has_ref = true;
1306
1307         return bkey;
1308 }
1309
1310 /**
1311  * bpf_lookup_system_key - lookup a key by a system-defined ID
1312  * @id: key ID
1313  *
1314  * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1315  * The key pointer is marked as invalid, to prevent bpf_key_put() from
1316  * attempting to decrement the key reference count on that pointer. The key
1317  * pointer set in such way is currently understood only by
1318  * verify_pkcs7_signature().
1319  *
1320  * Set *id* to one of the values defined in include/linux/verification.h:
1321  * 0 for the primary keyring (immutable keyring of system keys);
1322  * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1323  * (where keys can be added only if they are vouched for by existing keys
1324  * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1325  * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1326  * kerned image and, possibly, the initramfs signature).
1327  *
1328  * Return: a bpf_key pointer with an invalid key pointer set from the
1329  *         pre-determined ID on success, a NULL pointer otherwise
1330  */
1331 __bpf_kfunc struct bpf_key *bpf_lookup_system_key(u64 id)
1332 {
1333         struct bpf_key *bkey;
1334
1335         if (system_keyring_id_check(id) < 0)
1336                 return NULL;
1337
1338         bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1339         if (!bkey)
1340                 return NULL;
1341
1342         bkey->key = (struct key *)(unsigned long)id;
1343         bkey->has_ref = false;
1344
1345         return bkey;
1346 }
1347
1348 /**
1349  * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1350  * @bkey: bpf_key structure
1351  *
1352  * Decrement the reference count of the key inside *bkey*, if the pointer
1353  * is valid, and free *bkey*.
1354  */
1355 __bpf_kfunc void bpf_key_put(struct bpf_key *bkey)
1356 {
1357         if (bkey->has_ref)
1358                 key_put(bkey->key);
1359
1360         kfree(bkey);
1361 }
1362
1363 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1364 /**
1365  * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1366  * @data_ptr: data to verify
1367  * @sig_ptr: signature of the data
1368  * @trusted_keyring: keyring with keys trusted for signature verification
1369  *
1370  * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1371  * with keys in a keyring referenced by *trusted_keyring*.
1372  *
1373  * Return: 0 on success, a negative value on error.
1374  */
1375 __bpf_kfunc int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr,
1376                                struct bpf_dynptr_kern *sig_ptr,
1377                                struct bpf_key *trusted_keyring)
1378 {
1379         int ret;
1380
1381         if (trusted_keyring->has_ref) {
1382                 /*
1383                  * Do the permission check deferred in bpf_lookup_user_key().
1384                  * See bpf_lookup_user_key() for more details.
1385                  *
1386                  * A call to key_task_permission() here would be redundant, as
1387                  * it is already done by keyring_search() called by
1388                  * find_asymmetric_key().
1389                  */
1390                 ret = key_validate(trusted_keyring->key);
1391                 if (ret < 0)
1392                         return ret;
1393         }
1394
1395         return verify_pkcs7_signature(data_ptr->data,
1396                                       __bpf_dynptr_size(data_ptr),
1397                                       sig_ptr->data,
1398                                       __bpf_dynptr_size(sig_ptr),
1399                                       trusted_keyring->key,
1400                                       VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1401                                       NULL);
1402 }
1403 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1404
1405 __bpf_kfunc_end_defs();
1406
1407 BTF_SET8_START(key_sig_kfunc_set)
1408 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1409 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1410 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1411 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1412 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1413 #endif
1414 BTF_SET8_END(key_sig_kfunc_set)
1415
1416 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1417         .owner = THIS_MODULE,
1418         .set = &key_sig_kfunc_set,
1419 };
1420
1421 static int __init bpf_key_sig_kfuncs_init(void)
1422 {
1423         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1424                                          &bpf_key_sig_kfunc_set);
1425 }
1426
1427 late_initcall(bpf_key_sig_kfuncs_init);
1428 #endif /* CONFIG_KEYS */
1429
1430 static const struct bpf_func_proto *
1431 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1432 {
1433         switch (func_id) {
1434         case BPF_FUNC_map_lookup_elem:
1435                 return &bpf_map_lookup_elem_proto;
1436         case BPF_FUNC_map_update_elem:
1437                 return &bpf_map_update_elem_proto;
1438         case BPF_FUNC_map_delete_elem:
1439                 return &bpf_map_delete_elem_proto;
1440         case BPF_FUNC_map_push_elem:
1441                 return &bpf_map_push_elem_proto;
1442         case BPF_FUNC_map_pop_elem:
1443                 return &bpf_map_pop_elem_proto;
1444         case BPF_FUNC_map_peek_elem:
1445                 return &bpf_map_peek_elem_proto;
1446         case BPF_FUNC_map_lookup_percpu_elem:
1447                 return &bpf_map_lookup_percpu_elem_proto;
1448         case BPF_FUNC_ktime_get_ns:
1449                 return &bpf_ktime_get_ns_proto;
1450         case BPF_FUNC_ktime_get_boot_ns:
1451                 return &bpf_ktime_get_boot_ns_proto;
1452         case BPF_FUNC_tail_call:
1453                 return &bpf_tail_call_proto;
1454         case BPF_FUNC_get_current_pid_tgid:
1455                 return &bpf_get_current_pid_tgid_proto;
1456         case BPF_FUNC_get_current_task:
1457                 return &bpf_get_current_task_proto;
1458         case BPF_FUNC_get_current_task_btf:
1459                 return &bpf_get_current_task_btf_proto;
1460         case BPF_FUNC_task_pt_regs:
1461                 return &bpf_task_pt_regs_proto;
1462         case BPF_FUNC_get_current_uid_gid:
1463                 return &bpf_get_current_uid_gid_proto;
1464         case BPF_FUNC_get_current_comm:
1465                 return &bpf_get_current_comm_proto;
1466         case BPF_FUNC_trace_printk:
1467                 return bpf_get_trace_printk_proto();
1468         case BPF_FUNC_get_smp_processor_id:
1469                 return &bpf_get_smp_processor_id_proto;
1470         case BPF_FUNC_get_numa_node_id:
1471                 return &bpf_get_numa_node_id_proto;
1472         case BPF_FUNC_perf_event_read:
1473                 return &bpf_perf_event_read_proto;
1474         case BPF_FUNC_current_task_under_cgroup:
1475                 return &bpf_current_task_under_cgroup_proto;
1476         case BPF_FUNC_get_prandom_u32:
1477                 return &bpf_get_prandom_u32_proto;
1478         case BPF_FUNC_probe_write_user:
1479                 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1480                        NULL : bpf_get_probe_write_proto();
1481         case BPF_FUNC_probe_read_user:
1482                 return &bpf_probe_read_user_proto;
1483         case BPF_FUNC_probe_read_kernel:
1484                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1485                        NULL : &bpf_probe_read_kernel_proto;
1486         case BPF_FUNC_probe_read_user_str:
1487                 return &bpf_probe_read_user_str_proto;
1488         case BPF_FUNC_probe_read_kernel_str:
1489                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1490                        NULL : &bpf_probe_read_kernel_str_proto;
1491 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1492         case BPF_FUNC_probe_read:
1493                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1494                        NULL : &bpf_probe_read_compat_proto;
1495         case BPF_FUNC_probe_read_str:
1496                 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1497                        NULL : &bpf_probe_read_compat_str_proto;
1498 #endif
1499 #ifdef CONFIG_CGROUPS
1500         case BPF_FUNC_cgrp_storage_get:
1501                 return &bpf_cgrp_storage_get_proto;
1502         case BPF_FUNC_cgrp_storage_delete:
1503                 return &bpf_cgrp_storage_delete_proto;
1504 #endif
1505         case BPF_FUNC_send_signal:
1506                 return &bpf_send_signal_proto;
1507         case BPF_FUNC_send_signal_thread:
1508                 return &bpf_send_signal_thread_proto;
1509         case BPF_FUNC_perf_event_read_value:
1510                 return &bpf_perf_event_read_value_proto;
1511         case BPF_FUNC_get_ns_current_pid_tgid:
1512                 return &bpf_get_ns_current_pid_tgid_proto;
1513         case BPF_FUNC_ringbuf_output:
1514                 return &bpf_ringbuf_output_proto;
1515         case BPF_FUNC_ringbuf_reserve:
1516                 return &bpf_ringbuf_reserve_proto;
1517         case BPF_FUNC_ringbuf_submit:
1518                 return &bpf_ringbuf_submit_proto;
1519         case BPF_FUNC_ringbuf_discard:
1520                 return &bpf_ringbuf_discard_proto;
1521         case BPF_FUNC_ringbuf_query:
1522                 return &bpf_ringbuf_query_proto;
1523         case BPF_FUNC_jiffies64:
1524                 return &bpf_jiffies64_proto;
1525         case BPF_FUNC_get_task_stack:
1526                 return &bpf_get_task_stack_proto;
1527         case BPF_FUNC_copy_from_user:
1528                 return &bpf_copy_from_user_proto;
1529         case BPF_FUNC_copy_from_user_task:
1530                 return &bpf_copy_from_user_task_proto;
1531         case BPF_FUNC_snprintf_btf:
1532                 return &bpf_snprintf_btf_proto;
1533         case BPF_FUNC_per_cpu_ptr:
1534                 return &bpf_per_cpu_ptr_proto;
1535         case BPF_FUNC_this_cpu_ptr:
1536                 return &bpf_this_cpu_ptr_proto;
1537         case BPF_FUNC_task_storage_get:
1538                 if (bpf_prog_check_recur(prog))
1539                         return &bpf_task_storage_get_recur_proto;
1540                 return &bpf_task_storage_get_proto;
1541         case BPF_FUNC_task_storage_delete:
1542                 if (bpf_prog_check_recur(prog))
1543                         return &bpf_task_storage_delete_recur_proto;
1544                 return &bpf_task_storage_delete_proto;
1545         case BPF_FUNC_for_each_map_elem:
1546                 return &bpf_for_each_map_elem_proto;
1547         case BPF_FUNC_snprintf:
1548                 return &bpf_snprintf_proto;
1549         case BPF_FUNC_get_func_ip:
1550                 return &bpf_get_func_ip_proto_tracing;
1551         case BPF_FUNC_get_branch_snapshot:
1552                 return &bpf_get_branch_snapshot_proto;
1553         case BPF_FUNC_find_vma:
1554                 return &bpf_find_vma_proto;
1555         case BPF_FUNC_trace_vprintk:
1556                 return bpf_get_trace_vprintk_proto();
1557         default:
1558                 return bpf_base_func_proto(func_id);
1559         }
1560 }
1561
1562 static const struct bpf_func_proto *
1563 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1564 {
1565         switch (func_id) {
1566         case BPF_FUNC_perf_event_output:
1567                 return &bpf_perf_event_output_proto;
1568         case BPF_FUNC_get_stackid:
1569                 return &bpf_get_stackid_proto;
1570         case BPF_FUNC_get_stack:
1571                 return &bpf_get_stack_proto;
1572 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1573         case BPF_FUNC_override_return:
1574                 return &bpf_override_return_proto;
1575 #endif
1576         case BPF_FUNC_get_func_ip:
1577                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI)
1578                         return &bpf_get_func_ip_proto_kprobe_multi;
1579                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI)
1580                         return &bpf_get_func_ip_proto_uprobe_multi;
1581                 return &bpf_get_func_ip_proto_kprobe;
1582         case BPF_FUNC_get_attach_cookie:
1583                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI)
1584                         return &bpf_get_attach_cookie_proto_kmulti;
1585                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI)
1586                         return &bpf_get_attach_cookie_proto_umulti;
1587                 return &bpf_get_attach_cookie_proto_trace;
1588         default:
1589                 return bpf_tracing_func_proto(func_id, prog);
1590         }
1591 }
1592
1593 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1594 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1595                                         const struct bpf_prog *prog,
1596                                         struct bpf_insn_access_aux *info)
1597 {
1598         if (off < 0 || off >= sizeof(struct pt_regs))
1599                 return false;
1600         if (type != BPF_READ)
1601                 return false;
1602         if (off % size != 0)
1603                 return false;
1604         /*
1605          * Assertion for 32 bit to make sure last 8 byte access
1606          * (BPF_DW) to the last 4 byte member is disallowed.
1607          */
1608         if (off + size > sizeof(struct pt_regs))
1609                 return false;
1610
1611         return true;
1612 }
1613
1614 const struct bpf_verifier_ops kprobe_verifier_ops = {
1615         .get_func_proto  = kprobe_prog_func_proto,
1616         .is_valid_access = kprobe_prog_is_valid_access,
1617 };
1618
1619 const struct bpf_prog_ops kprobe_prog_ops = {
1620 };
1621
1622 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1623            u64, flags, void *, data, u64, size)
1624 {
1625         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1626
1627         /*
1628          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1629          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1630          * from there and call the same bpf_perf_event_output() helper inline.
1631          */
1632         return ____bpf_perf_event_output(regs, map, flags, data, size);
1633 }
1634
1635 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1636         .func           = bpf_perf_event_output_tp,
1637         .gpl_only       = true,
1638         .ret_type       = RET_INTEGER,
1639         .arg1_type      = ARG_PTR_TO_CTX,
1640         .arg2_type      = ARG_CONST_MAP_PTR,
1641         .arg3_type      = ARG_ANYTHING,
1642         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1643         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1644 };
1645
1646 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1647            u64, flags)
1648 {
1649         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1650
1651         /*
1652          * Same comment as in bpf_perf_event_output_tp(), only that this time
1653          * the other helper's function body cannot be inlined due to being
1654          * external, thus we need to call raw helper function.
1655          */
1656         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1657                                flags, 0, 0);
1658 }
1659
1660 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1661         .func           = bpf_get_stackid_tp,
1662         .gpl_only       = true,
1663         .ret_type       = RET_INTEGER,
1664         .arg1_type      = ARG_PTR_TO_CTX,
1665         .arg2_type      = ARG_CONST_MAP_PTR,
1666         .arg3_type      = ARG_ANYTHING,
1667 };
1668
1669 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1670            u64, flags)
1671 {
1672         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1673
1674         return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1675                              (unsigned long) size, flags, 0);
1676 }
1677
1678 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1679         .func           = bpf_get_stack_tp,
1680         .gpl_only       = true,
1681         .ret_type       = RET_INTEGER,
1682         .arg1_type      = ARG_PTR_TO_CTX,
1683         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1684         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1685         .arg4_type      = ARG_ANYTHING,
1686 };
1687
1688 static const struct bpf_func_proto *
1689 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1690 {
1691         switch (func_id) {
1692         case BPF_FUNC_perf_event_output:
1693                 return &bpf_perf_event_output_proto_tp;
1694         case BPF_FUNC_get_stackid:
1695                 return &bpf_get_stackid_proto_tp;
1696         case BPF_FUNC_get_stack:
1697                 return &bpf_get_stack_proto_tp;
1698         case BPF_FUNC_get_attach_cookie:
1699                 return &bpf_get_attach_cookie_proto_trace;
1700         default:
1701                 return bpf_tracing_func_proto(func_id, prog);
1702         }
1703 }
1704
1705 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1706                                     const struct bpf_prog *prog,
1707                                     struct bpf_insn_access_aux *info)
1708 {
1709         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1710                 return false;
1711         if (type != BPF_READ)
1712                 return false;
1713         if (off % size != 0)
1714                 return false;
1715
1716         BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1717         return true;
1718 }
1719
1720 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1721         .get_func_proto  = tp_prog_func_proto,
1722         .is_valid_access = tp_prog_is_valid_access,
1723 };
1724
1725 const struct bpf_prog_ops tracepoint_prog_ops = {
1726 };
1727
1728 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1729            struct bpf_perf_event_value *, buf, u32, size)
1730 {
1731         int err = -EINVAL;
1732
1733         if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1734                 goto clear;
1735         err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1736                                     &buf->running);
1737         if (unlikely(err))
1738                 goto clear;
1739         return 0;
1740 clear:
1741         memset(buf, 0, size);
1742         return err;
1743 }
1744
1745 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1746          .func           = bpf_perf_prog_read_value,
1747          .gpl_only       = true,
1748          .ret_type       = RET_INTEGER,
1749          .arg1_type      = ARG_PTR_TO_CTX,
1750          .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1751          .arg3_type      = ARG_CONST_SIZE,
1752 };
1753
1754 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1755            void *, buf, u32, size, u64, flags)
1756 {
1757         static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1758         struct perf_branch_stack *br_stack = ctx->data->br_stack;
1759         u32 to_copy;
1760
1761         if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1762                 return -EINVAL;
1763
1764         if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1765                 return -ENOENT;
1766
1767         if (unlikely(!br_stack))
1768                 return -ENOENT;
1769
1770         if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1771                 return br_stack->nr * br_entry_size;
1772
1773         if (!buf || (size % br_entry_size != 0))
1774                 return -EINVAL;
1775
1776         to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1777         memcpy(buf, br_stack->entries, to_copy);
1778
1779         return to_copy;
1780 }
1781
1782 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1783         .func           = bpf_read_branch_records,
1784         .gpl_only       = true,
1785         .ret_type       = RET_INTEGER,
1786         .arg1_type      = ARG_PTR_TO_CTX,
1787         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1788         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1789         .arg4_type      = ARG_ANYTHING,
1790 };
1791
1792 static const struct bpf_func_proto *
1793 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1794 {
1795         switch (func_id) {
1796         case BPF_FUNC_perf_event_output:
1797                 return &bpf_perf_event_output_proto_tp;
1798         case BPF_FUNC_get_stackid:
1799                 return &bpf_get_stackid_proto_pe;
1800         case BPF_FUNC_get_stack:
1801                 return &bpf_get_stack_proto_pe;
1802         case BPF_FUNC_perf_prog_read_value:
1803                 return &bpf_perf_prog_read_value_proto;
1804         case BPF_FUNC_read_branch_records:
1805                 return &bpf_read_branch_records_proto;
1806         case BPF_FUNC_get_attach_cookie:
1807                 return &bpf_get_attach_cookie_proto_pe;
1808         default:
1809                 return bpf_tracing_func_proto(func_id, prog);
1810         }
1811 }
1812
1813 /*
1814  * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1815  * to avoid potential recursive reuse issue when/if tracepoints are added
1816  * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1817  *
1818  * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1819  * in normal, irq, and nmi context.
1820  */
1821 struct bpf_raw_tp_regs {
1822         struct pt_regs regs[3];
1823 };
1824 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1825 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1826 static struct pt_regs *get_bpf_raw_tp_regs(void)
1827 {
1828         struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1829         int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1830
1831         if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1832                 this_cpu_dec(bpf_raw_tp_nest_level);
1833                 return ERR_PTR(-EBUSY);
1834         }
1835
1836         return &tp_regs->regs[nest_level - 1];
1837 }
1838
1839 static void put_bpf_raw_tp_regs(void)
1840 {
1841         this_cpu_dec(bpf_raw_tp_nest_level);
1842 }
1843
1844 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1845            struct bpf_map *, map, u64, flags, void *, data, u64, size)
1846 {
1847         struct pt_regs *regs = get_bpf_raw_tp_regs();
1848         int ret;
1849
1850         if (IS_ERR(regs))
1851                 return PTR_ERR(regs);
1852
1853         perf_fetch_caller_regs(regs);
1854         ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1855
1856         put_bpf_raw_tp_regs();
1857         return ret;
1858 }
1859
1860 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1861         .func           = bpf_perf_event_output_raw_tp,
1862         .gpl_only       = true,
1863         .ret_type       = RET_INTEGER,
1864         .arg1_type      = ARG_PTR_TO_CTX,
1865         .arg2_type      = ARG_CONST_MAP_PTR,
1866         .arg3_type      = ARG_ANYTHING,
1867         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1868         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
1869 };
1870
1871 extern const struct bpf_func_proto bpf_skb_output_proto;
1872 extern const struct bpf_func_proto bpf_xdp_output_proto;
1873 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1874
1875 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1876            struct bpf_map *, map, u64, flags)
1877 {
1878         struct pt_regs *regs = get_bpf_raw_tp_regs();
1879         int ret;
1880
1881         if (IS_ERR(regs))
1882                 return PTR_ERR(regs);
1883
1884         perf_fetch_caller_regs(regs);
1885         /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1886         ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1887                               flags, 0, 0);
1888         put_bpf_raw_tp_regs();
1889         return ret;
1890 }
1891
1892 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1893         .func           = bpf_get_stackid_raw_tp,
1894         .gpl_only       = true,
1895         .ret_type       = RET_INTEGER,
1896         .arg1_type      = ARG_PTR_TO_CTX,
1897         .arg2_type      = ARG_CONST_MAP_PTR,
1898         .arg3_type      = ARG_ANYTHING,
1899 };
1900
1901 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1902            void *, buf, u32, size, u64, flags)
1903 {
1904         struct pt_regs *regs = get_bpf_raw_tp_regs();
1905         int ret;
1906
1907         if (IS_ERR(regs))
1908                 return PTR_ERR(regs);
1909
1910         perf_fetch_caller_regs(regs);
1911         ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1912                             (unsigned long) size, flags, 0);
1913         put_bpf_raw_tp_regs();
1914         return ret;
1915 }
1916
1917 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1918         .func           = bpf_get_stack_raw_tp,
1919         .gpl_only       = true,
1920         .ret_type       = RET_INTEGER,
1921         .arg1_type      = ARG_PTR_TO_CTX,
1922         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1923         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1924         .arg4_type      = ARG_ANYTHING,
1925 };
1926
1927 static const struct bpf_func_proto *
1928 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1929 {
1930         switch (func_id) {
1931         case BPF_FUNC_perf_event_output:
1932                 return &bpf_perf_event_output_proto_raw_tp;
1933         case BPF_FUNC_get_stackid:
1934                 return &bpf_get_stackid_proto_raw_tp;
1935         case BPF_FUNC_get_stack:
1936                 return &bpf_get_stack_proto_raw_tp;
1937         default:
1938                 return bpf_tracing_func_proto(func_id, prog);
1939         }
1940 }
1941
1942 const struct bpf_func_proto *
1943 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1944 {
1945         const struct bpf_func_proto *fn;
1946
1947         switch (func_id) {
1948 #ifdef CONFIG_NET
1949         case BPF_FUNC_skb_output:
1950                 return &bpf_skb_output_proto;
1951         case BPF_FUNC_xdp_output:
1952                 return &bpf_xdp_output_proto;
1953         case BPF_FUNC_skc_to_tcp6_sock:
1954                 return &bpf_skc_to_tcp6_sock_proto;
1955         case BPF_FUNC_skc_to_tcp_sock:
1956                 return &bpf_skc_to_tcp_sock_proto;
1957         case BPF_FUNC_skc_to_tcp_timewait_sock:
1958                 return &bpf_skc_to_tcp_timewait_sock_proto;
1959         case BPF_FUNC_skc_to_tcp_request_sock:
1960                 return &bpf_skc_to_tcp_request_sock_proto;
1961         case BPF_FUNC_skc_to_udp6_sock:
1962                 return &bpf_skc_to_udp6_sock_proto;
1963         case BPF_FUNC_skc_to_unix_sock:
1964                 return &bpf_skc_to_unix_sock_proto;
1965         case BPF_FUNC_skc_to_mptcp_sock:
1966                 return &bpf_skc_to_mptcp_sock_proto;
1967         case BPF_FUNC_sk_storage_get:
1968                 return &bpf_sk_storage_get_tracing_proto;
1969         case BPF_FUNC_sk_storage_delete:
1970                 return &bpf_sk_storage_delete_tracing_proto;
1971         case BPF_FUNC_sock_from_file:
1972                 return &bpf_sock_from_file_proto;
1973         case BPF_FUNC_get_socket_cookie:
1974                 return &bpf_get_socket_ptr_cookie_proto;
1975         case BPF_FUNC_xdp_get_buff_len:
1976                 return &bpf_xdp_get_buff_len_trace_proto;
1977 #endif
1978         case BPF_FUNC_seq_printf:
1979                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1980                        &bpf_seq_printf_proto :
1981                        NULL;
1982         case BPF_FUNC_seq_write:
1983                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1984                        &bpf_seq_write_proto :
1985                        NULL;
1986         case BPF_FUNC_seq_printf_btf:
1987                 return prog->expected_attach_type == BPF_TRACE_ITER ?
1988                        &bpf_seq_printf_btf_proto :
1989                        NULL;
1990         case BPF_FUNC_d_path:
1991                 return &bpf_d_path_proto;
1992         case BPF_FUNC_get_func_arg:
1993                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1994         case BPF_FUNC_get_func_ret:
1995                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1996         case BPF_FUNC_get_func_arg_cnt:
1997                 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1998         case BPF_FUNC_get_attach_cookie:
1999                 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
2000         default:
2001                 fn = raw_tp_prog_func_proto(func_id, prog);
2002                 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
2003                         fn = bpf_iter_get_func_proto(func_id, prog);
2004                 return fn;
2005         }
2006 }
2007
2008 static bool raw_tp_prog_is_valid_access(int off, int size,
2009                                         enum bpf_access_type type,
2010                                         const struct bpf_prog *prog,
2011                                         struct bpf_insn_access_aux *info)
2012 {
2013         return bpf_tracing_ctx_access(off, size, type);
2014 }
2015
2016 static bool tracing_prog_is_valid_access(int off, int size,
2017                                          enum bpf_access_type type,
2018                                          const struct bpf_prog *prog,
2019                                          struct bpf_insn_access_aux *info)
2020 {
2021         return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
2022 }
2023
2024 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
2025                                      const union bpf_attr *kattr,
2026                                      union bpf_attr __user *uattr)
2027 {
2028         return -ENOTSUPP;
2029 }
2030
2031 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
2032         .get_func_proto  = raw_tp_prog_func_proto,
2033         .is_valid_access = raw_tp_prog_is_valid_access,
2034 };
2035
2036 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
2037 #ifdef CONFIG_NET
2038         .test_run = bpf_prog_test_run_raw_tp,
2039 #endif
2040 };
2041
2042 const struct bpf_verifier_ops tracing_verifier_ops = {
2043         .get_func_proto  = tracing_prog_func_proto,
2044         .is_valid_access = tracing_prog_is_valid_access,
2045 };
2046
2047 const struct bpf_prog_ops tracing_prog_ops = {
2048         .test_run = bpf_prog_test_run_tracing,
2049 };
2050
2051 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
2052                                                  enum bpf_access_type type,
2053                                                  const struct bpf_prog *prog,
2054                                                  struct bpf_insn_access_aux *info)
2055 {
2056         if (off == 0) {
2057                 if (size != sizeof(u64) || type != BPF_READ)
2058                         return false;
2059                 info->reg_type = PTR_TO_TP_BUFFER;
2060         }
2061         return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2062 }
2063
2064 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2065         .get_func_proto  = raw_tp_prog_func_proto,
2066         .is_valid_access = raw_tp_writable_prog_is_valid_access,
2067 };
2068
2069 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2070 };
2071
2072 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2073                                     const struct bpf_prog *prog,
2074                                     struct bpf_insn_access_aux *info)
2075 {
2076         const int size_u64 = sizeof(u64);
2077
2078         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2079                 return false;
2080         if (type != BPF_READ)
2081                 return false;
2082         if (off % size != 0) {
2083                 if (sizeof(unsigned long) != 4)
2084                         return false;
2085                 if (size != 8)
2086                         return false;
2087                 if (off % size != 4)
2088                         return false;
2089         }
2090
2091         switch (off) {
2092         case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2093                 bpf_ctx_record_field_size(info, size_u64);
2094                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2095                         return false;
2096                 break;
2097         case bpf_ctx_range(struct bpf_perf_event_data, addr):
2098                 bpf_ctx_record_field_size(info, size_u64);
2099                 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2100                         return false;
2101                 break;
2102         default:
2103                 if (size != sizeof(long))
2104                         return false;
2105         }
2106
2107         return true;
2108 }
2109
2110 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2111                                       const struct bpf_insn *si,
2112                                       struct bpf_insn *insn_buf,
2113                                       struct bpf_prog *prog, u32 *target_size)
2114 {
2115         struct bpf_insn *insn = insn_buf;
2116
2117         switch (si->off) {
2118         case offsetof(struct bpf_perf_event_data, sample_period):
2119                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2120                                                        data), si->dst_reg, si->src_reg,
2121                                       offsetof(struct bpf_perf_event_data_kern, data));
2122                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2123                                       bpf_target_off(struct perf_sample_data, period, 8,
2124                                                      target_size));
2125                 break;
2126         case offsetof(struct bpf_perf_event_data, addr):
2127                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2128                                                        data), si->dst_reg, si->src_reg,
2129                                       offsetof(struct bpf_perf_event_data_kern, data));
2130                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2131                                       bpf_target_off(struct perf_sample_data, addr, 8,
2132                                                      target_size));
2133                 break;
2134         default:
2135                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2136                                                        regs), si->dst_reg, si->src_reg,
2137                                       offsetof(struct bpf_perf_event_data_kern, regs));
2138                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2139                                       si->off);
2140                 break;
2141         }
2142
2143         return insn - insn_buf;
2144 }
2145
2146 const struct bpf_verifier_ops perf_event_verifier_ops = {
2147         .get_func_proto         = pe_prog_func_proto,
2148         .is_valid_access        = pe_prog_is_valid_access,
2149         .convert_ctx_access     = pe_prog_convert_ctx_access,
2150 };
2151
2152 const struct bpf_prog_ops perf_event_prog_ops = {
2153 };
2154
2155 static DEFINE_MUTEX(bpf_event_mutex);
2156
2157 #define BPF_TRACE_MAX_PROGS 64
2158
2159 int perf_event_attach_bpf_prog(struct perf_event *event,
2160                                struct bpf_prog *prog,
2161                                u64 bpf_cookie)
2162 {
2163         struct bpf_prog_array *old_array;
2164         struct bpf_prog_array *new_array;
2165         int ret = -EEXIST;
2166
2167         /*
2168          * Kprobe override only works if they are on the function entry,
2169          * and only if they are on the opt-in list.
2170          */
2171         if (prog->kprobe_override &&
2172             (!trace_kprobe_on_func_entry(event->tp_event) ||
2173              !trace_kprobe_error_injectable(event->tp_event)))
2174                 return -EINVAL;
2175
2176         mutex_lock(&bpf_event_mutex);
2177
2178         if (event->prog)
2179                 goto unlock;
2180
2181         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2182         if (old_array &&
2183             bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2184                 ret = -E2BIG;
2185                 goto unlock;
2186         }
2187
2188         ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2189         if (ret < 0)
2190                 goto unlock;
2191
2192         /* set the new array to event->tp_event and set event->prog */
2193         event->prog = prog;
2194         event->bpf_cookie = bpf_cookie;
2195         rcu_assign_pointer(event->tp_event->prog_array, new_array);
2196         bpf_prog_array_free_sleepable(old_array);
2197
2198 unlock:
2199         mutex_unlock(&bpf_event_mutex);
2200         return ret;
2201 }
2202
2203 void perf_event_detach_bpf_prog(struct perf_event *event)
2204 {
2205         struct bpf_prog_array *old_array;
2206         struct bpf_prog_array *new_array;
2207         int ret;
2208
2209         mutex_lock(&bpf_event_mutex);
2210
2211         if (!event->prog)
2212                 goto unlock;
2213
2214         old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2215         ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2216         if (ret == -ENOENT)
2217                 goto unlock;
2218         if (ret < 0) {
2219                 bpf_prog_array_delete_safe(old_array, event->prog);
2220         } else {
2221                 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2222                 bpf_prog_array_free_sleepable(old_array);
2223         }
2224
2225         bpf_prog_put(event->prog);
2226         event->prog = NULL;
2227
2228 unlock:
2229         mutex_unlock(&bpf_event_mutex);
2230 }
2231
2232 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2233 {
2234         struct perf_event_query_bpf __user *uquery = info;
2235         struct perf_event_query_bpf query = {};
2236         struct bpf_prog_array *progs;
2237         u32 *ids, prog_cnt, ids_len;
2238         int ret;
2239
2240         if (!perfmon_capable())
2241                 return -EPERM;
2242         if (event->attr.type != PERF_TYPE_TRACEPOINT)
2243                 return -EINVAL;
2244         if (copy_from_user(&query, uquery, sizeof(query)))
2245                 return -EFAULT;
2246
2247         ids_len = query.ids_len;
2248         if (ids_len > BPF_TRACE_MAX_PROGS)
2249                 return -E2BIG;
2250         ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2251         if (!ids)
2252                 return -ENOMEM;
2253         /*
2254          * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2255          * is required when user only wants to check for uquery->prog_cnt.
2256          * There is no need to check for it since the case is handled
2257          * gracefully in bpf_prog_array_copy_info.
2258          */
2259
2260         mutex_lock(&bpf_event_mutex);
2261         progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2262         ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2263         mutex_unlock(&bpf_event_mutex);
2264
2265         if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2266             copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2267                 ret = -EFAULT;
2268
2269         kfree(ids);
2270         return ret;
2271 }
2272
2273 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2274 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2275
2276 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2277 {
2278         struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2279
2280         for (; btp < __stop__bpf_raw_tp; btp++) {
2281                 if (!strcmp(btp->tp->name, name))
2282                         return btp;
2283         }
2284
2285         return bpf_get_raw_tracepoint_module(name);
2286 }
2287
2288 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2289 {
2290         struct module *mod;
2291
2292         preempt_disable();
2293         mod = __module_address((unsigned long)btp);
2294         module_put(mod);
2295         preempt_enable();
2296 }
2297
2298 static __always_inline
2299 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2300 {
2301         cant_sleep();
2302         if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2303                 bpf_prog_inc_misses_counter(prog);
2304                 goto out;
2305         }
2306         rcu_read_lock();
2307         (void) bpf_prog_run(prog, args);
2308         rcu_read_unlock();
2309 out:
2310         this_cpu_dec(*(prog->active));
2311 }
2312
2313 #define UNPACK(...)                     __VA_ARGS__
2314 #define REPEAT_1(FN, DL, X, ...)        FN(X)
2315 #define REPEAT_2(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2316 #define REPEAT_3(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2317 #define REPEAT_4(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2318 #define REPEAT_5(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2319 #define REPEAT_6(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2320 #define REPEAT_7(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2321 #define REPEAT_8(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2322 #define REPEAT_9(FN, DL, X, ...)        FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2323 #define REPEAT_10(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2324 #define REPEAT_11(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2325 #define REPEAT_12(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2326 #define REPEAT(X, FN, DL, ...)          REPEAT_##X(FN, DL, __VA_ARGS__)
2327
2328 #define SARG(X)         u64 arg##X
2329 #define COPY(X)         args[X] = arg##X
2330
2331 #define __DL_COM        (,)
2332 #define __DL_SEM        (;)
2333
2334 #define __SEQ_0_11      0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2335
2336 #define BPF_TRACE_DEFN_x(x)                                             \
2337         void bpf_trace_run##x(struct bpf_prog *prog,                    \
2338                               REPEAT(x, SARG, __DL_COM, __SEQ_0_11))    \
2339         {                                                               \
2340                 u64 args[x];                                            \
2341                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
2342                 __bpf_trace_run(prog, args);                            \
2343         }                                                               \
2344         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2345 BPF_TRACE_DEFN_x(1);
2346 BPF_TRACE_DEFN_x(2);
2347 BPF_TRACE_DEFN_x(3);
2348 BPF_TRACE_DEFN_x(4);
2349 BPF_TRACE_DEFN_x(5);
2350 BPF_TRACE_DEFN_x(6);
2351 BPF_TRACE_DEFN_x(7);
2352 BPF_TRACE_DEFN_x(8);
2353 BPF_TRACE_DEFN_x(9);
2354 BPF_TRACE_DEFN_x(10);
2355 BPF_TRACE_DEFN_x(11);
2356 BPF_TRACE_DEFN_x(12);
2357
2358 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2359 {
2360         struct tracepoint *tp = btp->tp;
2361
2362         /*
2363          * check that program doesn't access arguments beyond what's
2364          * available in this tracepoint
2365          */
2366         if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2367                 return -EINVAL;
2368
2369         if (prog->aux->max_tp_access > btp->writable_size)
2370                 return -EINVAL;
2371
2372         return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2373                                                    prog);
2374 }
2375
2376 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2377 {
2378         return __bpf_probe_register(btp, prog);
2379 }
2380
2381 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2382 {
2383         return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2384 }
2385
2386 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2387                             u32 *fd_type, const char **buf,
2388                             u64 *probe_offset, u64 *probe_addr,
2389                             unsigned long *missed)
2390 {
2391         bool is_tracepoint, is_syscall_tp;
2392         struct bpf_prog *prog;
2393         int flags, err = 0;
2394
2395         prog = event->prog;
2396         if (!prog)
2397                 return -ENOENT;
2398
2399         /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2400         if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2401                 return -EOPNOTSUPP;
2402
2403         *prog_id = prog->aux->id;
2404         flags = event->tp_event->flags;
2405         is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2406         is_syscall_tp = is_syscall_trace_event(event->tp_event);
2407
2408         if (is_tracepoint || is_syscall_tp) {
2409                 *buf = is_tracepoint ? event->tp_event->tp->name
2410                                      : event->tp_event->name;
2411                 /* We allow NULL pointer for tracepoint */
2412                 if (fd_type)
2413                         *fd_type = BPF_FD_TYPE_TRACEPOINT;
2414                 if (probe_offset)
2415                         *probe_offset = 0x0;
2416                 if (probe_addr)
2417                         *probe_addr = 0x0;
2418         } else {
2419                 /* kprobe/uprobe */
2420                 err = -EOPNOTSUPP;
2421 #ifdef CONFIG_KPROBE_EVENTS
2422                 if (flags & TRACE_EVENT_FL_KPROBE)
2423                         err = bpf_get_kprobe_info(event, fd_type, buf,
2424                                                   probe_offset, probe_addr, missed,
2425                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2426 #endif
2427 #ifdef CONFIG_UPROBE_EVENTS
2428                 if (flags & TRACE_EVENT_FL_UPROBE)
2429                         err = bpf_get_uprobe_info(event, fd_type, buf,
2430                                                   probe_offset, probe_addr,
2431                                                   event->attr.type == PERF_TYPE_TRACEPOINT);
2432 #endif
2433         }
2434
2435         return err;
2436 }
2437
2438 static int __init send_signal_irq_work_init(void)
2439 {
2440         int cpu;
2441         struct send_signal_irq_work *work;
2442
2443         for_each_possible_cpu(cpu) {
2444                 work = per_cpu_ptr(&send_signal_work, cpu);
2445                 init_irq_work(&work->irq_work, do_bpf_send_signal);
2446         }
2447         return 0;
2448 }
2449
2450 subsys_initcall(send_signal_irq_work_init);
2451
2452 #ifdef CONFIG_MODULES
2453 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2454                             void *module)
2455 {
2456         struct bpf_trace_module *btm, *tmp;
2457         struct module *mod = module;
2458         int ret = 0;
2459
2460         if (mod->num_bpf_raw_events == 0 ||
2461             (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2462                 goto out;
2463
2464         mutex_lock(&bpf_module_mutex);
2465
2466         switch (op) {
2467         case MODULE_STATE_COMING:
2468                 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2469                 if (btm) {
2470                         btm->module = module;
2471                         list_add(&btm->list, &bpf_trace_modules);
2472                 } else {
2473                         ret = -ENOMEM;
2474                 }
2475                 break;
2476         case MODULE_STATE_GOING:
2477                 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2478                         if (btm->module == module) {
2479                                 list_del(&btm->list);
2480                                 kfree(btm);
2481                                 break;
2482                         }
2483                 }
2484                 break;
2485         }
2486
2487         mutex_unlock(&bpf_module_mutex);
2488
2489 out:
2490         return notifier_from_errno(ret);
2491 }
2492
2493 static struct notifier_block bpf_module_nb = {
2494         .notifier_call = bpf_event_notify,
2495 };
2496
2497 static int __init bpf_event_init(void)
2498 {
2499         register_module_notifier(&bpf_module_nb);
2500         return 0;
2501 }
2502
2503 fs_initcall(bpf_event_init);
2504 #endif /* CONFIG_MODULES */
2505
2506 #ifdef CONFIG_FPROBE
2507 struct bpf_kprobe_multi_link {
2508         struct bpf_link link;
2509         struct fprobe fp;
2510         unsigned long *addrs;
2511         u64 *cookies;
2512         u32 cnt;
2513         u32 mods_cnt;
2514         struct module **mods;
2515         u32 flags;
2516 };
2517
2518 struct bpf_kprobe_multi_run_ctx {
2519         struct bpf_run_ctx run_ctx;
2520         struct bpf_kprobe_multi_link *link;
2521         unsigned long entry_ip;
2522 };
2523
2524 struct user_syms {
2525         const char **syms;
2526         char *buf;
2527 };
2528
2529 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2530 {
2531         unsigned long __user usymbol;
2532         const char **syms = NULL;
2533         char *buf = NULL, *p;
2534         int err = -ENOMEM;
2535         unsigned int i;
2536
2537         syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2538         if (!syms)
2539                 goto error;
2540
2541         buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2542         if (!buf)
2543                 goto error;
2544
2545         for (p = buf, i = 0; i < cnt; i++) {
2546                 if (__get_user(usymbol, usyms + i)) {
2547                         err = -EFAULT;
2548                         goto error;
2549                 }
2550                 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2551                 if (err == KSYM_NAME_LEN)
2552                         err = -E2BIG;
2553                 if (err < 0)
2554                         goto error;
2555                 syms[i] = p;
2556                 p += err + 1;
2557         }
2558
2559         us->syms = syms;
2560         us->buf = buf;
2561         return 0;
2562
2563 error:
2564         if (err) {
2565                 kvfree(syms);
2566                 kvfree(buf);
2567         }
2568         return err;
2569 }
2570
2571 static void kprobe_multi_put_modules(struct module **mods, u32 cnt)
2572 {
2573         u32 i;
2574
2575         for (i = 0; i < cnt; i++)
2576                 module_put(mods[i]);
2577 }
2578
2579 static void free_user_syms(struct user_syms *us)
2580 {
2581         kvfree(us->syms);
2582         kvfree(us->buf);
2583 }
2584
2585 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2586 {
2587         struct bpf_kprobe_multi_link *kmulti_link;
2588
2589         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2590         unregister_fprobe(&kmulti_link->fp);
2591         kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt);
2592 }
2593
2594 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2595 {
2596         struct bpf_kprobe_multi_link *kmulti_link;
2597
2598         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2599         kvfree(kmulti_link->addrs);
2600         kvfree(kmulti_link->cookies);
2601         kfree(kmulti_link->mods);
2602         kfree(kmulti_link);
2603 }
2604
2605 static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,
2606                                                 struct bpf_link_info *info)
2607 {
2608         u64 __user *uaddrs = u64_to_user_ptr(info->kprobe_multi.addrs);
2609         struct bpf_kprobe_multi_link *kmulti_link;
2610         u32 ucount = info->kprobe_multi.count;
2611         int err = 0, i;
2612
2613         if (!uaddrs ^ !ucount)
2614                 return -EINVAL;
2615
2616         kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2617         info->kprobe_multi.count = kmulti_link->cnt;
2618         info->kprobe_multi.flags = kmulti_link->flags;
2619         info->kprobe_multi.missed = kmulti_link->fp.nmissed;
2620
2621         if (!uaddrs)
2622                 return 0;
2623         if (ucount < kmulti_link->cnt)
2624                 err = -ENOSPC;
2625         else
2626                 ucount = kmulti_link->cnt;
2627
2628         if (kallsyms_show_value(current_cred())) {
2629                 if (copy_to_user(uaddrs, kmulti_link->addrs, ucount * sizeof(u64)))
2630                         return -EFAULT;
2631         } else {
2632                 for (i = 0; i < ucount; i++) {
2633                         if (put_user(0, uaddrs + i))
2634                                 return -EFAULT;
2635                 }
2636         }
2637         return err;
2638 }
2639
2640 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2641         .release = bpf_kprobe_multi_link_release,
2642         .dealloc = bpf_kprobe_multi_link_dealloc,
2643         .fill_link_info = bpf_kprobe_multi_link_fill_link_info,
2644 };
2645
2646 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2647 {
2648         const struct bpf_kprobe_multi_link *link = priv;
2649         unsigned long *addr_a = a, *addr_b = b;
2650         u64 *cookie_a, *cookie_b;
2651
2652         cookie_a = link->cookies + (addr_a - link->addrs);
2653         cookie_b = link->cookies + (addr_b - link->addrs);
2654
2655         /* swap addr_a/addr_b and cookie_a/cookie_b values */
2656         swap(*addr_a, *addr_b);
2657         swap(*cookie_a, *cookie_b);
2658 }
2659
2660 static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b)
2661 {
2662         const unsigned long *addr_a = a, *addr_b = b;
2663
2664         if (*addr_a == *addr_b)
2665                 return 0;
2666         return *addr_a < *addr_b ? -1 : 1;
2667 }
2668
2669 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2670 {
2671         return bpf_kprobe_multi_addrs_cmp(a, b);
2672 }
2673
2674 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2675 {
2676         struct bpf_kprobe_multi_run_ctx *run_ctx;
2677         struct bpf_kprobe_multi_link *link;
2678         u64 *cookie, entry_ip;
2679         unsigned long *addr;
2680
2681         if (WARN_ON_ONCE(!ctx))
2682                 return 0;
2683         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2684         link = run_ctx->link;
2685         if (!link->cookies)
2686                 return 0;
2687         entry_ip = run_ctx->entry_ip;
2688         addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2689                        bpf_kprobe_multi_addrs_cmp);
2690         if (!addr)
2691                 return 0;
2692         cookie = link->cookies + (addr - link->addrs);
2693         return *cookie;
2694 }
2695
2696 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2697 {
2698         struct bpf_kprobe_multi_run_ctx *run_ctx;
2699
2700         run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2701         return run_ctx->entry_ip;
2702 }
2703
2704 static int
2705 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2706                            unsigned long entry_ip, struct pt_regs *regs)
2707 {
2708         struct bpf_kprobe_multi_run_ctx run_ctx = {
2709                 .link = link,
2710                 .entry_ip = entry_ip,
2711         };
2712         struct bpf_run_ctx *old_run_ctx;
2713         int err;
2714
2715         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2716                 bpf_prog_inc_misses_counter(link->link.prog);
2717                 err = 0;
2718                 goto out;
2719         }
2720
2721         migrate_disable();
2722         rcu_read_lock();
2723         old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2724         err = bpf_prog_run(link->link.prog, regs);
2725         bpf_reset_run_ctx(old_run_ctx);
2726         rcu_read_unlock();
2727         migrate_enable();
2728
2729  out:
2730         __this_cpu_dec(bpf_prog_active);
2731         return err;
2732 }
2733
2734 static int
2735 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2736                           unsigned long ret_ip, struct pt_regs *regs,
2737                           void *data)
2738 {
2739         struct bpf_kprobe_multi_link *link;
2740
2741         link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2742         kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2743         return 0;
2744 }
2745
2746 static void
2747 kprobe_multi_link_exit_handler(struct fprobe *fp, unsigned long fentry_ip,
2748                                unsigned long ret_ip, struct pt_regs *regs,
2749                                void *data)
2750 {
2751         struct bpf_kprobe_multi_link *link;
2752
2753         link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2754         kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2755 }
2756
2757 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2758 {
2759         const char **str_a = (const char **) a;
2760         const char **str_b = (const char **) b;
2761
2762         return strcmp(*str_a, *str_b);
2763 }
2764
2765 struct multi_symbols_sort {
2766         const char **funcs;
2767         u64 *cookies;
2768 };
2769
2770 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2771 {
2772         const struct multi_symbols_sort *data = priv;
2773         const char **name_a = a, **name_b = b;
2774
2775         swap(*name_a, *name_b);
2776
2777         /* If defined, swap also related cookies. */
2778         if (data->cookies) {
2779                 u64 *cookie_a, *cookie_b;
2780
2781                 cookie_a = data->cookies + (name_a - data->funcs);
2782                 cookie_b = data->cookies + (name_b - data->funcs);
2783                 swap(*cookie_a, *cookie_b);
2784         }
2785 }
2786
2787 struct modules_array {
2788         struct module **mods;
2789         int mods_cnt;
2790         int mods_cap;
2791 };
2792
2793 static int add_module(struct modules_array *arr, struct module *mod)
2794 {
2795         struct module **mods;
2796
2797         if (arr->mods_cnt == arr->mods_cap) {
2798                 arr->mods_cap = max(16, arr->mods_cap * 3 / 2);
2799                 mods = krealloc_array(arr->mods, arr->mods_cap, sizeof(*mods), GFP_KERNEL);
2800                 if (!mods)
2801                         return -ENOMEM;
2802                 arr->mods = mods;
2803         }
2804
2805         arr->mods[arr->mods_cnt] = mod;
2806         arr->mods_cnt++;
2807         return 0;
2808 }
2809
2810 static bool has_module(struct modules_array *arr, struct module *mod)
2811 {
2812         int i;
2813
2814         for (i = arr->mods_cnt - 1; i >= 0; i--) {
2815                 if (arr->mods[i] == mod)
2816                         return true;
2817         }
2818         return false;
2819 }
2820
2821 static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt)
2822 {
2823         struct modules_array arr = {};
2824         u32 i, err = 0;
2825
2826         for (i = 0; i < addrs_cnt; i++) {
2827                 struct module *mod;
2828
2829                 preempt_disable();
2830                 mod = __module_address(addrs[i]);
2831                 /* Either no module or we it's already stored  */
2832                 if (!mod || has_module(&arr, mod)) {
2833                         preempt_enable();
2834                         continue;
2835                 }
2836                 if (!try_module_get(mod))
2837                         err = -EINVAL;
2838                 preempt_enable();
2839                 if (err)
2840                         break;
2841                 err = add_module(&arr, mod);
2842                 if (err) {
2843                         module_put(mod);
2844                         break;
2845                 }
2846         }
2847
2848         /* We return either err < 0 in case of error, ... */
2849         if (err) {
2850                 kprobe_multi_put_modules(arr.mods, arr.mods_cnt);
2851                 kfree(arr.mods);
2852                 return err;
2853         }
2854
2855         /* or number of modules found if everything is ok. */
2856         *mods = arr.mods;
2857         return arr.mods_cnt;
2858 }
2859
2860 static int addrs_check_error_injection_list(unsigned long *addrs, u32 cnt)
2861 {
2862         u32 i;
2863
2864         for (i = 0; i < cnt; i++) {
2865                 if (!within_error_injection_list(addrs[i]))
2866                         return -EINVAL;
2867         }
2868         return 0;
2869 }
2870
2871 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2872 {
2873         struct bpf_kprobe_multi_link *link = NULL;
2874         struct bpf_link_primer link_primer;
2875         void __user *ucookies;
2876         unsigned long *addrs;
2877         u32 flags, cnt, size;
2878         void __user *uaddrs;
2879         u64 *cookies = NULL;
2880         void __user *usyms;
2881         int err;
2882
2883         /* no support for 32bit archs yet */
2884         if (sizeof(u64) != sizeof(void *))
2885                 return -EOPNOTSUPP;
2886
2887         if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2888                 return -EINVAL;
2889
2890         flags = attr->link_create.kprobe_multi.flags;
2891         if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2892                 return -EINVAL;
2893
2894         uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2895         usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2896         if (!!uaddrs == !!usyms)
2897                 return -EINVAL;
2898
2899         cnt = attr->link_create.kprobe_multi.cnt;
2900         if (!cnt)
2901                 return -EINVAL;
2902
2903         size = cnt * sizeof(*addrs);
2904         addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2905         if (!addrs)
2906                 return -ENOMEM;
2907
2908         ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2909         if (ucookies) {
2910                 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2911                 if (!cookies) {
2912                         err = -ENOMEM;
2913                         goto error;
2914                 }
2915                 if (copy_from_user(cookies, ucookies, size)) {
2916                         err = -EFAULT;
2917                         goto error;
2918                 }
2919         }
2920
2921         if (uaddrs) {
2922                 if (copy_from_user(addrs, uaddrs, size)) {
2923                         err = -EFAULT;
2924                         goto error;
2925                 }
2926         } else {
2927                 struct multi_symbols_sort data = {
2928                         .cookies = cookies,
2929                 };
2930                 struct user_syms us;
2931
2932                 err = copy_user_syms(&us, usyms, cnt);
2933                 if (err)
2934                         goto error;
2935
2936                 if (cookies)
2937                         data.funcs = us.syms;
2938
2939                 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2940                        symbols_swap_r, &data);
2941
2942                 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2943                 free_user_syms(&us);
2944                 if (err)
2945                         goto error;
2946         }
2947
2948         if (prog->kprobe_override && addrs_check_error_injection_list(addrs, cnt)) {
2949                 err = -EINVAL;
2950                 goto error;
2951         }
2952
2953         link = kzalloc(sizeof(*link), GFP_KERNEL);
2954         if (!link) {
2955                 err = -ENOMEM;
2956                 goto error;
2957         }
2958
2959         bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2960                       &bpf_kprobe_multi_link_lops, prog);
2961
2962         err = bpf_link_prime(&link->link, &link_primer);
2963         if (err)
2964                 goto error;
2965
2966         if (flags & BPF_F_KPROBE_MULTI_RETURN)
2967                 link->fp.exit_handler = kprobe_multi_link_exit_handler;
2968         else
2969                 link->fp.entry_handler = kprobe_multi_link_handler;
2970
2971         link->addrs = addrs;
2972         link->cookies = cookies;
2973         link->cnt = cnt;
2974         link->flags = flags;
2975
2976         if (cookies) {
2977                 /*
2978                  * Sorting addresses will trigger sorting cookies as well
2979                  * (check bpf_kprobe_multi_cookie_swap). This way we can
2980                  * find cookie based on the address in bpf_get_attach_cookie
2981                  * helper.
2982                  */
2983                 sort_r(addrs, cnt, sizeof(*addrs),
2984                        bpf_kprobe_multi_cookie_cmp,
2985                        bpf_kprobe_multi_cookie_swap,
2986                        link);
2987         }
2988
2989         err = get_modules_for_addrs(&link->mods, addrs, cnt);
2990         if (err < 0) {
2991                 bpf_link_cleanup(&link_primer);
2992                 return err;
2993         }
2994         link->mods_cnt = err;
2995
2996         err = register_fprobe_ips(&link->fp, addrs, cnt);
2997         if (err) {
2998                 kprobe_multi_put_modules(link->mods, link->mods_cnt);
2999                 bpf_link_cleanup(&link_primer);
3000                 return err;
3001         }
3002
3003         return bpf_link_settle(&link_primer);
3004
3005 error:
3006         kfree(link);
3007         kvfree(addrs);
3008         kvfree(cookies);
3009         return err;
3010 }
3011 #else /* !CONFIG_FPROBE */
3012 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3013 {
3014         return -EOPNOTSUPP;
3015 }
3016 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
3017 {
3018         return 0;
3019 }
3020 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3021 {
3022         return 0;
3023 }
3024 #endif
3025
3026 #ifdef CONFIG_UPROBES
3027 struct bpf_uprobe_multi_link;
3028
3029 struct bpf_uprobe {
3030         struct bpf_uprobe_multi_link *link;
3031         loff_t offset;
3032         u64 cookie;
3033         struct uprobe_consumer consumer;
3034 };
3035
3036 struct bpf_uprobe_multi_link {
3037         struct path path;
3038         struct bpf_link link;
3039         u32 cnt;
3040         struct bpf_uprobe *uprobes;
3041         struct task_struct *task;
3042 };
3043
3044 struct bpf_uprobe_multi_run_ctx {
3045         struct bpf_run_ctx run_ctx;
3046         unsigned long entry_ip;
3047         struct bpf_uprobe *uprobe;
3048 };
3049
3050 static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes,
3051                                   u32 cnt)
3052 {
3053         u32 i;
3054
3055         for (i = 0; i < cnt; i++) {
3056                 uprobe_unregister(d_real_inode(path->dentry), uprobes[i].offset,
3057                                   &uprobes[i].consumer);
3058         }
3059 }
3060
3061 static void bpf_uprobe_multi_link_release(struct bpf_link *link)
3062 {
3063         struct bpf_uprobe_multi_link *umulti_link;
3064
3065         umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3066         bpf_uprobe_unregister(&umulti_link->path, umulti_link->uprobes, umulti_link->cnt);
3067 }
3068
3069 static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)
3070 {
3071         struct bpf_uprobe_multi_link *umulti_link;
3072
3073         umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3074         if (umulti_link->task)
3075                 put_task_struct(umulti_link->task);
3076         path_put(&umulti_link->path);
3077         kvfree(umulti_link->uprobes);
3078         kfree(umulti_link);
3079 }
3080
3081 static const struct bpf_link_ops bpf_uprobe_multi_link_lops = {
3082         .release = bpf_uprobe_multi_link_release,
3083         .dealloc = bpf_uprobe_multi_link_dealloc,
3084 };
3085
3086 static int uprobe_prog_run(struct bpf_uprobe *uprobe,
3087                            unsigned long entry_ip,
3088                            struct pt_regs *regs)
3089 {
3090         struct bpf_uprobe_multi_link *link = uprobe->link;
3091         struct bpf_uprobe_multi_run_ctx run_ctx = {
3092                 .entry_ip = entry_ip,
3093                 .uprobe = uprobe,
3094         };
3095         struct bpf_prog *prog = link->link.prog;
3096         bool sleepable = prog->aux->sleepable;
3097         struct bpf_run_ctx *old_run_ctx;
3098         int err = 0;
3099
3100         if (link->task && current != link->task)
3101                 return 0;
3102
3103         if (sleepable)
3104                 rcu_read_lock_trace();
3105         else
3106                 rcu_read_lock();
3107
3108         migrate_disable();
3109
3110         old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
3111         err = bpf_prog_run(link->link.prog, regs);
3112         bpf_reset_run_ctx(old_run_ctx);
3113
3114         migrate_enable();
3115
3116         if (sleepable)
3117                 rcu_read_unlock_trace();
3118         else
3119                 rcu_read_unlock();
3120         return err;
3121 }
3122
3123 static bool
3124 uprobe_multi_link_filter(struct uprobe_consumer *con, enum uprobe_filter_ctx ctx,
3125                          struct mm_struct *mm)
3126 {
3127         struct bpf_uprobe *uprobe;
3128
3129         uprobe = container_of(con, struct bpf_uprobe, consumer);
3130         return uprobe->link->task->mm == mm;
3131 }
3132
3133 static int
3134 uprobe_multi_link_handler(struct uprobe_consumer *con, struct pt_regs *regs)
3135 {
3136         struct bpf_uprobe *uprobe;
3137
3138         uprobe = container_of(con, struct bpf_uprobe, consumer);
3139         return uprobe_prog_run(uprobe, instruction_pointer(regs), regs);
3140 }
3141
3142 static int
3143 uprobe_multi_link_ret_handler(struct uprobe_consumer *con, unsigned long func, struct pt_regs *regs)
3144 {
3145         struct bpf_uprobe *uprobe;
3146
3147         uprobe = container_of(con, struct bpf_uprobe, consumer);
3148         return uprobe_prog_run(uprobe, func, regs);
3149 }
3150
3151 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3152 {
3153         struct bpf_uprobe_multi_run_ctx *run_ctx;
3154
3155         run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx, run_ctx);
3156         return run_ctx->entry_ip;
3157 }
3158
3159 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3160 {
3161         struct bpf_uprobe_multi_run_ctx *run_ctx;
3162
3163         run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx, run_ctx);
3164         return run_ctx->uprobe->cookie;
3165 }
3166
3167 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3168 {
3169         struct bpf_uprobe_multi_link *link = NULL;
3170         unsigned long __user *uref_ctr_offsets;
3171         unsigned long *ref_ctr_offsets = NULL;
3172         struct bpf_link_primer link_primer;
3173         struct bpf_uprobe *uprobes = NULL;
3174         struct task_struct *task = NULL;
3175         unsigned long __user *uoffsets;
3176         u64 __user *ucookies;
3177         void __user *upath;
3178         u32 flags, cnt, i;
3179         struct path path;
3180         char *name;
3181         pid_t pid;
3182         int err;
3183
3184         /* no support for 32bit archs yet */
3185         if (sizeof(u64) != sizeof(void *))
3186                 return -EOPNOTSUPP;
3187
3188         if (prog->expected_attach_type != BPF_TRACE_UPROBE_MULTI)
3189                 return -EINVAL;
3190
3191         flags = attr->link_create.uprobe_multi.flags;
3192         if (flags & ~BPF_F_UPROBE_MULTI_RETURN)
3193                 return -EINVAL;
3194
3195         /*
3196          * path, offsets and cnt are mandatory,
3197          * ref_ctr_offsets and cookies are optional
3198          */
3199         upath = u64_to_user_ptr(attr->link_create.uprobe_multi.path);
3200         uoffsets = u64_to_user_ptr(attr->link_create.uprobe_multi.offsets);
3201         cnt = attr->link_create.uprobe_multi.cnt;
3202
3203         if (!upath || !uoffsets || !cnt)
3204                 return -EINVAL;
3205
3206         uref_ctr_offsets = u64_to_user_ptr(attr->link_create.uprobe_multi.ref_ctr_offsets);
3207         ucookies = u64_to_user_ptr(attr->link_create.uprobe_multi.cookies);
3208
3209         name = strndup_user(upath, PATH_MAX);
3210         if (IS_ERR(name)) {
3211                 err = PTR_ERR(name);
3212                 return err;
3213         }
3214
3215         err = kern_path(name, LOOKUP_FOLLOW, &path);
3216         kfree(name);
3217         if (err)
3218                 return err;
3219
3220         if (!d_is_reg(path.dentry)) {
3221                 err = -EBADF;
3222                 goto error_path_put;
3223         }
3224
3225         pid = attr->link_create.uprobe_multi.pid;
3226         if (pid) {
3227                 rcu_read_lock();
3228                 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3229                 rcu_read_unlock();
3230                 if (!task) {
3231                         err = -ESRCH;
3232                         goto error_path_put;
3233                 }
3234         }
3235
3236         err = -ENOMEM;
3237
3238         link = kzalloc(sizeof(*link), GFP_KERNEL);
3239         uprobes = kvcalloc(cnt, sizeof(*uprobes), GFP_KERNEL);
3240
3241         if (!uprobes || !link)
3242                 goto error_free;
3243
3244         if (uref_ctr_offsets) {
3245                 ref_ctr_offsets = kvcalloc(cnt, sizeof(*ref_ctr_offsets), GFP_KERNEL);
3246                 if (!ref_ctr_offsets)
3247                         goto error_free;
3248         }
3249
3250         for (i = 0; i < cnt; i++) {
3251                 if (ucookies && __get_user(uprobes[i].cookie, ucookies + i)) {
3252                         err = -EFAULT;
3253                         goto error_free;
3254                 }
3255                 if (uref_ctr_offsets && __get_user(ref_ctr_offsets[i], uref_ctr_offsets + i)) {
3256                         err = -EFAULT;
3257                         goto error_free;
3258                 }
3259                 if (__get_user(uprobes[i].offset, uoffsets + i)) {
3260                         err = -EFAULT;
3261                         goto error_free;
3262                 }
3263
3264                 uprobes[i].link = link;
3265
3266                 if (flags & BPF_F_UPROBE_MULTI_RETURN)
3267                         uprobes[i].consumer.ret_handler = uprobe_multi_link_ret_handler;
3268                 else
3269                         uprobes[i].consumer.handler = uprobe_multi_link_handler;
3270
3271                 if (pid)
3272                         uprobes[i].consumer.filter = uprobe_multi_link_filter;
3273         }
3274
3275         link->cnt = cnt;
3276         link->uprobes = uprobes;
3277         link->path = path;
3278         link->task = task;
3279
3280         bpf_link_init(&link->link, BPF_LINK_TYPE_UPROBE_MULTI,
3281                       &bpf_uprobe_multi_link_lops, prog);
3282
3283         for (i = 0; i < cnt; i++) {
3284                 err = uprobe_register_refctr(d_real_inode(link->path.dentry),
3285                                              uprobes[i].offset,
3286                                              ref_ctr_offsets ? ref_ctr_offsets[i] : 0,
3287                                              &uprobes[i].consumer);
3288                 if (err) {
3289                         bpf_uprobe_unregister(&path, uprobes, i);
3290                         goto error_free;
3291                 }
3292         }
3293
3294         err = bpf_link_prime(&link->link, &link_primer);
3295         if (err)
3296                 goto error_free;
3297
3298         kvfree(ref_ctr_offsets);
3299         return bpf_link_settle(&link_primer);
3300
3301 error_free:
3302         kvfree(ref_ctr_offsets);
3303         kvfree(uprobes);
3304         kfree(link);
3305         if (task)
3306                 put_task_struct(task);
3307 error_path_put:
3308         path_put(&path);
3309         return err;
3310 }
3311 #else /* !CONFIG_UPROBES */
3312 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3313 {
3314         return -EOPNOTSUPP;
3315 }
3316 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3317 {
3318         return 0;
3319 }
3320 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3321 {
3322         return 0;
3323 }
3324 #endif /* CONFIG_UPROBES */