kernel/trace/bpf_trace.c

   1 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   2  * Copyright (c) 2016 Facebook
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  */
   8 #include <linux/kernel.h>
   9 #include <linux/types.h>
  10 #include <linux/slab.h>
  11 #include <linux/bpf.h>
  12 #include <linux/bpf_perf_event.h>
  13 #include <linux/filter.h>
  14 #include <linux/uaccess.h>
  15 #include <linux/ctype.h>
  16 #include "trace.h"
  17
  18 /**
  19  * trace_call_bpf - invoke BPF program
  20  * @prog: BPF program
  21  * @ctx: opaque context pointer
  22  *
  23  * kprobe handlers execute BPF programs via this helper.
  24  * Can be used from static tracepoints in the future.
  25  *
  26  * Return: BPF programs always return an integer which is interpreted by
  27  * kprobe handler as:
  28  * 0 - return from kprobe (event is filtered out)
  29  * 1 - store kprobe event into ring buffer
  30  * Other values are reserved and currently alias to 1
  31  */
  32 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
  33 {
  34         unsigned int ret;
  35
  36         if (in_nmi()) /* not supported yet */
  37                 return 1;
  38
  39         preempt_disable();
  40
  41         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
  42                 /*
  43                  * since some bpf program is already running on this cpu,
  44                  * don't call into another bpf program (same or different)
  45                  * and don't send kprobe event into ring-buffer,
  46                  * so return zero here
  47                  */
  48                 ret = 0;
  49                 goto out;
  50         }
  51
  52         rcu_read_lock();
  53         ret = BPF_PROG_RUN(prog, ctx);
  54         rcu_read_unlock();
  55
  56  out:
  57         __this_cpu_dec(bpf_prog_active);
  58         preempt_enable();
  59
  60         return ret;
  61 }
  62 EXPORT_SYMBOL_GPL(trace_call_bpf);
  63
  64 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
  65 {
  66         int ret;
  67
  68         ret = probe_kernel_read(dst, unsafe_ptr, size);
  69         if (unlikely(ret < 0))
  70                 memset(dst, 0, size);
  71
  72         return ret;
  73 }
  74
  75 static const struct bpf_func_proto bpf_probe_read_proto = {
  76         .func           = bpf_probe_read,
  77         .gpl_only       = true,
  78         .ret_type       = RET_INTEGER,
  79         .arg1_type      = ARG_PTR_TO_RAW_STACK,
  80         .arg2_type      = ARG_CONST_STACK_SIZE,
  81         .arg3_type      = ARG_ANYTHING,
  82 };
  83
  84 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
  85            u32, size)
  86 {
  87         /*
  88          * Ensure we're in user context which is safe for the helper to
  89          * run. This helper has no business in a kthread.
  90          *
  91          * access_ok() should prevent writing to non-user memory, but in
  92          * some situations (nommu, temporary switch, etc) access_ok() does
  93          * not provide enough validation, hence the check on KERNEL_DS.
  94          */
  95
  96         if (unlikely(in_interrupt() ||
  97                      current->flags & (PF_KTHREAD | PF_EXITING)))
  98                 return -EPERM;
  99         if (unlikely(segment_eq(get_fs(), KERNEL_DS)))
 100                 return -EPERM;
 101         if (!access_ok(VERIFY_WRITE, unsafe_ptr, size))
 102                 return -EPERM;
 103
 104         return probe_kernel_write(unsafe_ptr, src, size);
 105 }
 106
 107 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 108         .func           = bpf_probe_write_user,
 109         .gpl_only       = true,
 110         .ret_type       = RET_INTEGER,
 111         .arg1_type      = ARG_ANYTHING,
 112         .arg2_type      = ARG_PTR_TO_STACK,
 113         .arg3_type      = ARG_CONST_STACK_SIZE,
 114 };
 115
 116 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 117 {
 118         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 119                             current->comm, task_pid_nr(current));
 120
 121         return &bpf_probe_write_user_proto;
 122 }
 123
 124 /*
 125  * limited trace_printk()
 126  * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
 127  */
 128 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 129            u64, arg2, u64, arg3)
 130 {
 131         bool str_seen = false;
 132         int mod[3] = {};
 133         int fmt_cnt = 0;
 134         u64 unsafe_addr;
 135         char buf[64];
 136         int i;
 137
 138         /*
 139          * bpf_check()->check_func_arg()->check_stack_boundary()
 140          * guarantees that fmt points to bpf program stack,
 141          * fmt_size bytes of it were initialized and fmt_size > 0
 142          */
 143         if (fmt[--fmt_size] != 0)
 144                 return -EINVAL;
 145
 146         /* check format string for allowed specifiers */
 147         for (i = 0; i < fmt_size; i++) {
 148                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 149                         return -EINVAL;
 150
 151                 if (fmt[i] != '%')
 152                         continue;
 153
 154                 if (fmt_cnt >= 3)
 155                         return -EINVAL;
 156
 157                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 158                 i++;
 159                 if (fmt[i] == 'l') {
 160                         mod[fmt_cnt]++;
 161                         i++;
 162                 } else if (fmt[i] == 'p' || fmt[i] == 's') {
 163                         mod[fmt_cnt]++;
 164                         i++;
 165                         if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
 166                                 return -EINVAL;
 167                         fmt_cnt++;
 168                         if (fmt[i - 1] == 's') {
 169                                 if (str_seen)
 170                                         /* allow only one '%s' per fmt string */
 171                                         return -EINVAL;
 172                                 str_seen = true;
 173
 174                                 switch (fmt_cnt) {
 175                                 case 1:
 176                                         unsafe_addr = arg1;
 177                                         arg1 = (long) buf;
 178                                         break;
 179                                 case 2:
 180                                         unsafe_addr = arg2;
 181                                         arg2 = (long) buf;
 182                                         break;
 183                                 case 3:
 184                                         unsafe_addr = arg3;
 185                                         arg3 = (long) buf;
 186                                         break;
 187                                 }
 188                                 buf[0] = 0;
 189                                 strncpy_from_unsafe(buf,
 190                                                     (void *) (long) unsafe_addr,
 191                                                     sizeof(buf));
 192                         }
 193                         continue;
 194                 }
 195
 196                 if (fmt[i] == 'l') {
 197                         mod[fmt_cnt]++;
 198                         i++;
 199                 }
 200
 201                 if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
 202                         return -EINVAL;
 203                 fmt_cnt++;
 204         }
 205
 206         return __trace_printk(1/* fake ip will not be printed */, fmt,
 207                               mod[0] == 2 ? arg1 : mod[0] == 1 ? (long) arg1 : (u32) arg1,
 208                               mod[1] == 2 ? arg2 : mod[1] == 1 ? (long) arg2 : (u32) arg2,
 209                               mod[2] == 2 ? arg3 : mod[2] == 1 ? (long) arg3 : (u32) arg3);
 210 }
 211
 212 static const struct bpf_func_proto bpf_trace_printk_proto = {
 213         .func           = bpf_trace_printk,
 214         .gpl_only       = true,
 215         .ret_type       = RET_INTEGER,
 216         .arg1_type      = ARG_PTR_TO_STACK,
 217         .arg2_type      = ARG_CONST_STACK_SIZE,
 218 };
 219
 220 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 221 {
 222         /*
 223          * this program might be calling bpf_trace_printk,
 224          * so allocate per-cpu printk buffers
 225          */
 226         trace_printk_init_buffers();
 227
 228         return &bpf_trace_printk_proto;
 229 }
 230
 231 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 232 {
 233         struct bpf_array *array = container_of(map, struct bpf_array, map);
 234         unsigned int cpu = smp_processor_id();
 235         u64 index = flags & BPF_F_INDEX_MASK;
 236         struct bpf_event_entry *ee;
 237         struct perf_event *event;
 238
 239         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 240                 return -EINVAL;
 241         if (index == BPF_F_CURRENT_CPU)
 242                 index = cpu;
 243         if (unlikely(index >= array->map.max_entries))
 244                 return -E2BIG;
 245
 246         ee = READ_ONCE(array->ptrs[index]);
 247         if (!ee)
 248                 return -ENOENT;
 249
 250         event = ee->event;
 251         if (unlikely(event->attr.type != PERF_TYPE_HARDWARE &&
 252                      event->attr.type != PERF_TYPE_RAW))
 253                 return -EINVAL;
 254
 255         /* make sure event is local and doesn't have pmu::count */
 256         if (unlikely(event->oncpu != cpu || event->pmu->count))
 257                 return -EINVAL;
 258
 259         /*
 260          * we don't know if the function is run successfully by the
 261          * return value. It can be judged in other places, such as
 262          * eBPF programs.
 263          */
 264         return perf_event_read_local(event);
 265 }
 266
 267 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 268         .func           = bpf_perf_event_read,
 269         .gpl_only       = true,
 270         .ret_type       = RET_INTEGER,
 271         .arg1_type      = ARG_CONST_MAP_PTR,
 272         .arg2_type      = ARG_ANYTHING,
 273 };
 274
 275 static __always_inline u64
 276 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 277                         u64 flags, struct perf_raw_record *raw)
 278 {
 279         struct bpf_array *array = container_of(map, struct bpf_array, map);
 280         unsigned int cpu = smp_processor_id();
 281         u64 index = flags & BPF_F_INDEX_MASK;
 282         struct perf_sample_data sample_data;
 283         struct bpf_event_entry *ee;
 284         struct perf_event *event;
 285
 286         if (index == BPF_F_CURRENT_CPU)
 287                 index = cpu;
 288         if (unlikely(index >= array->map.max_entries))
 289                 return -E2BIG;
 290
 291         ee = READ_ONCE(array->ptrs[index]);
 292         if (!ee)
 293                 return -ENOENT;
 294
 295         event = ee->event;
 296         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 297                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 298                 return -EINVAL;
 299
 300         if (unlikely(event->oncpu != cpu))
 301                 return -EOPNOTSUPP;
 302
 303         perf_sample_data_init(&sample_data, 0, 0);
 304         sample_data.raw = raw;
 305         perf_event_output(event, &sample_data, regs);
 306         return 0;
 307 }
 308
 309 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 310            u64, flags, void *, data, u64, size)
 311 {
 312         struct perf_raw_record raw = {
 313                 .frag = {
 314                         .size = size,
 315                         .data = data,
 316                 },
 317         };
 318
 319         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 320                 return -EINVAL;
 321
 322         return __bpf_perf_event_output(regs, map, flags, &raw);
 323 }
 324
 325 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 326         .func           = bpf_perf_event_output,
 327         .gpl_only       = true,
 328         .ret_type       = RET_INTEGER,
 329         .arg1_type      = ARG_PTR_TO_CTX,
 330         .arg2_type      = ARG_CONST_MAP_PTR,
 331         .arg3_type      = ARG_ANYTHING,
 332         .arg4_type      = ARG_PTR_TO_STACK,
 333         .arg5_type      = ARG_CONST_STACK_SIZE,
 334 };
 335
 336 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
 337
 338 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 339                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 340 {
 341         struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
 342         struct perf_raw_frag frag = {
 343                 .copy           = ctx_copy,
 344                 .size           = ctx_size,
 345                 .data           = ctx,
 346         };
 347         struct perf_raw_record raw = {
 348                 .frag = {
 349                         {
 350                                 .next   = ctx_size ? &frag : NULL,
 351                         },
 352                         .size   = meta_size,
 353                         .data   = meta,
 354                 },
 355         };
 356
 357         perf_fetch_caller_regs(regs);
 358
 359         return __bpf_perf_event_output(regs, map, flags, &raw);
 360 }
 361
 362 BPF_CALL_0(bpf_get_current_task)
 363 {
 364         return (long) current;
 365 }
 366
 367 static const struct bpf_func_proto bpf_get_current_task_proto = {
 368         .func           = bpf_get_current_task,
 369         .gpl_only       = true,
 370         .ret_type       = RET_INTEGER,
 371 };
 372
 373 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 374 {
 375         struct bpf_array *array = container_of(map, struct bpf_array, map);
 376         struct cgroup *cgrp;
 377
 378         if (unlikely(in_interrupt()))
 379                 return -EINVAL;
 380         if (unlikely(idx >= array->map.max_entries))
 381                 return -E2BIG;
 382
 383         cgrp = READ_ONCE(array->ptrs[idx]);
 384         if (unlikely(!cgrp))
 385                 return -EAGAIN;
 386
 387         return task_under_cgroup_hierarchy(current, cgrp);
 388 }
 389
 390 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 391         .func           = bpf_current_task_under_cgroup,
 392         .gpl_only       = false,
 393         .ret_type       = RET_INTEGER,
 394         .arg1_type      = ARG_CONST_MAP_PTR,
 395         .arg2_type      = ARG_ANYTHING,
 396 };
 397
 398 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id)
 399 {
 400         switch (func_id) {
 401         case BPF_FUNC_map_lookup_elem:
 402                 return &bpf_map_lookup_elem_proto;
 403         case BPF_FUNC_map_update_elem:
 404                 return &bpf_map_update_elem_proto;
 405         case BPF_FUNC_map_delete_elem:
 406                 return &bpf_map_delete_elem_proto;
 407         case BPF_FUNC_probe_read:
 408                 return &bpf_probe_read_proto;
 409         case BPF_FUNC_ktime_get_ns:
 410                 return &bpf_ktime_get_ns_proto;
 411         case BPF_FUNC_tail_call:
 412                 return &bpf_tail_call_proto;
 413         case BPF_FUNC_get_current_pid_tgid:
 414                 return &bpf_get_current_pid_tgid_proto;
 415         case BPF_FUNC_get_current_task:
 416                 return &bpf_get_current_task_proto;
 417         case BPF_FUNC_get_current_uid_gid:
 418                 return &bpf_get_current_uid_gid_proto;
 419         case BPF_FUNC_get_current_comm:
 420                 return &bpf_get_current_comm_proto;
 421         case BPF_FUNC_trace_printk:
 422                 return bpf_get_trace_printk_proto();
 423         case BPF_FUNC_get_smp_processor_id:
 424                 return &bpf_get_smp_processor_id_proto;
 425         case BPF_FUNC_perf_event_read:
 426                 return &bpf_perf_event_read_proto;
 427         case BPF_FUNC_probe_write_user:
 428                 return bpf_get_probe_write_proto();
 429         case BPF_FUNC_current_task_under_cgroup:
 430                 return &bpf_current_task_under_cgroup_proto;
 431         case BPF_FUNC_get_prandom_u32:
 432                 return &bpf_get_prandom_u32_proto;
 433         default:
 434                 return NULL;
 435         }
 436 }
 437
 438 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
 439 {
 440         switch (func_id) {
 441         case BPF_FUNC_perf_event_output:
 442                 return &bpf_perf_event_output_proto;
 443         case BPF_FUNC_get_stackid:
 444                 return &bpf_get_stackid_proto;
 445         default:
 446                 return tracing_func_proto(func_id);
 447         }
 448 }
 449
 450 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
 451 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 452                                         enum bpf_reg_type *reg_type)
 453 {
 454         if (off < 0 || off >= sizeof(struct pt_regs))
 455                 return false;
 456         if (type != BPF_READ)
 457                 return false;
 458         if (off % size != 0)
 459                 return false;
 460         return true;
 461 }
 462
 463 static const struct bpf_verifier_ops kprobe_prog_ops = {
 464         .get_func_proto  = kprobe_prog_func_proto,
 465         .is_valid_access = kprobe_prog_is_valid_access,
 466 };
 467
 468 static struct bpf_prog_type_list kprobe_tl = {
 469         .ops    = &kprobe_prog_ops,
 470         .type   = BPF_PROG_TYPE_KPROBE,
 471 };
 472
 473 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
 474            u64, flags, void *, data, u64, size)
 475 {
 476         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
 477
 478         /*
 479          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
 480          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
 481          * from there and call the same bpf_perf_event_output() helper inline.
 482          */
 483         return ____bpf_perf_event_output(regs, map, flags, data, size);
 484 }
 485
 486 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
 487         .func           = bpf_perf_event_output_tp,
 488         .gpl_only       = true,
 489         .ret_type       = RET_INTEGER,
 490         .arg1_type      = ARG_PTR_TO_CTX,
 491         .arg2_type      = ARG_CONST_MAP_PTR,
 492         .arg3_type      = ARG_ANYTHING,
 493         .arg4_type      = ARG_PTR_TO_STACK,
 494         .arg5_type      = ARG_CONST_STACK_SIZE,
 495 };
 496
 497 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
 498            u64, flags)
 499 {
 500         struct pt_regs *regs = *(struct pt_regs **)tp_buff;
 501
 502         /*
 503          * Same comment as in bpf_perf_event_output_tp(), only that this time
 504          * the other helper's function body cannot be inlined due to being
 505          * external, thus we need to call raw helper function.
 506          */
 507         return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
 508                                flags, 0, 0);
 509 }
 510
 511 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
 512         .func           = bpf_get_stackid_tp,
 513         .gpl_only       = true,
 514         .ret_type       = RET_INTEGER,
 515         .arg1_type      = ARG_PTR_TO_CTX,
 516         .arg2_type      = ARG_CONST_MAP_PTR,
 517         .arg3_type      = ARG_ANYTHING,
 518 };
 519
 520 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id)
 521 {
 522         switch (func_id) {
 523         case BPF_FUNC_perf_event_output:
 524                 return &bpf_perf_event_output_proto_tp;
 525         case BPF_FUNC_get_stackid:
 526                 return &bpf_get_stackid_proto_tp;
 527         default:
 528                 return tracing_func_proto(func_id);
 529         }
 530 }
 531
 532 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 533                                     enum bpf_reg_type *reg_type)
 534 {
 535         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
 536                 return false;
 537         if (type != BPF_READ)
 538                 return false;
 539         if (off % size != 0)
 540                 return false;
 541         return true;
 542 }
 543
 544 static const struct bpf_verifier_ops tracepoint_prog_ops = {
 545         .get_func_proto  = tp_prog_func_proto,
 546         .is_valid_access = tp_prog_is_valid_access,
 547 };
 548
 549 static struct bpf_prog_type_list tracepoint_tl = {
 550         .ops    = &tracepoint_prog_ops,
 551         .type   = BPF_PROG_TYPE_TRACEPOINT,
 552 };
 553
 554 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 555                                     enum bpf_reg_type *reg_type)
 556 {
 557         if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
 558                 return false;
 559         if (type != BPF_READ)
 560                 return false;
 561         if (off % size != 0)
 562                 return false;
 563         if (off == offsetof(struct bpf_perf_event_data, sample_period)) {
 564                 if (size != sizeof(u64))
 565                         return false;
 566         } else {
 567                 if (size != sizeof(long))
 568                         return false;
 569         }
 570         return true;
 571 }
 572
 573 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, int dst_reg,
 574                                       int src_reg, int ctx_off,
 575                                       struct bpf_insn *insn_buf,
 576                                       struct bpf_prog *prog)
 577 {
 578         struct bpf_insn *insn = insn_buf;
 579
 580         switch (ctx_off) {
 581         case offsetof(struct bpf_perf_event_data, sample_period):
 582                 BUILD_BUG_ON(FIELD_SIZEOF(struct perf_sample_data, period) != sizeof(u64));
 583
 584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
 585                                                        data), dst_reg, src_reg,
 586                                       offsetof(struct bpf_perf_event_data_kern, data));
 587                 *insn++ = BPF_LDX_MEM(BPF_DW, dst_reg, dst_reg,
 588                                       offsetof(struct perf_sample_data, period));
 589                 break;
 590         default:
 591                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
 592                                                        regs), dst_reg, src_reg,
 593                                       offsetof(struct bpf_perf_event_data_kern, regs));
 594                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), dst_reg, dst_reg, ctx_off);
 595                 break;
 596         }
 597
 598         return insn - insn_buf;
 599 }
 600
 601 static const struct bpf_verifier_ops perf_event_prog_ops = {
 602         .get_func_proto         = tp_prog_func_proto,
 603         .is_valid_access        = pe_prog_is_valid_access,
 604         .convert_ctx_access     = pe_prog_convert_ctx_access,
 605 };
 606
 607 static struct bpf_prog_type_list perf_event_tl = {
 608         .ops    = &perf_event_prog_ops,
 609         .type   = BPF_PROG_TYPE_PERF_EVENT,
 610 };
 611
 612 static int __init register_kprobe_prog_ops(void)
 613 {
 614         bpf_register_prog_type(&kprobe_tl);
 615         bpf_register_prog_type(&tracepoint_tl);
 616         bpf_register_prog_type(&perf_event_tl);
 617         return 0;
 618 }
 619 late_initcall(register_kprobe_prog_ops);