kernel/trace/bpf_trace.c

   1 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  */
   7 #include <linux/kernel.h>
   8 #include <linux/types.h>
   9 #include <linux/slab.h>
  10 #include <linux/bpf.h>
  11 #include <linux/filter.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/ctype.h>
  14 #include "trace.h"
  15
  16 /**
  17  * trace_call_bpf - invoke BPF program
  18  * @prog: BPF program
  19  * @ctx: opaque context pointer
  20  *
  21  * kprobe handlers execute BPF programs via this helper.
  22  * Can be used from static tracepoints in the future.
  23  *
  24  * Return: BPF programs always return an integer which is interpreted by
  25  * kprobe handler as:
  26  * 0 - return from kprobe (event is filtered out)
  27  * 1 - store kprobe event into ring buffer
  28  * Other values are reserved and currently alias to 1
  29  */
  30 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
  31 {
  32         unsigned int ret;
  33
  34         if (in_nmi()) /* not supported yet */
  35                 return 1;
  36
  37         preempt_disable();
  38
  39         if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
  40                 /*
  41                  * since some bpf program is already running on this cpu,
  42                  * don't call into another bpf program (same or different)
  43                  * and don't send kprobe event into ring-buffer,
  44                  * so return zero here
  45                  */
  46                 ret = 0;
  47                 goto out;
  48         }
  49
  50         rcu_read_lock();
  51         ret = BPF_PROG_RUN(prog, ctx);
  52         rcu_read_unlock();
  53
  54  out:
  55         __this_cpu_dec(bpf_prog_active);
  56         preempt_enable();
  57
  58         return ret;
  59 }
  60 EXPORT_SYMBOL_GPL(trace_call_bpf);
  61
  62 static u64 bpf_probe_read(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
  63 {
  64         void *dst = (void *) (long) r1;
  65         int ret, size = (int) r2;
  66         void *unsafe_ptr = (void *) (long) r3;
  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 static u64 bpf_probe_write_user(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
  85 {
  86         void *unsafe_ptr = (void *) (long) r1;
  87         void *src = (void *) (long) r2;
  88         int size = (int) r3;
  89
  90         /*
  91          * Ensure we're in user context which is safe for the helper to
  92          * run. This helper has no business in a kthread.
  93          *
  94          * access_ok() should prevent writing to non-user memory, but in
  95          * some situations (nommu, temporary switch, etc) access_ok() does
  96          * not provide enough validation, hence the check on KERNEL_DS.
  97          */
  98
  99         if (unlikely(in_interrupt() ||
 100                      current->flags & (PF_KTHREAD | PF_EXITING)))
 101                 return -EPERM;
 102         if (unlikely(segment_eq(get_fs(), KERNEL_DS)))
 103                 return -EPERM;
 104         if (!access_ok(VERIFY_WRITE, unsafe_ptr, size))
 105                 return -EPERM;
 106
 107         return probe_kernel_write(unsafe_ptr, src, size);
 108 }
 109
 110 static const struct bpf_func_proto bpf_probe_write_user_proto = {
 111         .func           = bpf_probe_write_user,
 112         .gpl_only       = true,
 113         .ret_type       = RET_INTEGER,
 114         .arg1_type      = ARG_ANYTHING,
 115         .arg2_type      = ARG_PTR_TO_STACK,
 116         .arg3_type      = ARG_CONST_STACK_SIZE,
 117 };
 118
 119 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 120 {
 121         pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
 122                             current->comm, task_pid_nr(current));
 123
 124         return &bpf_probe_write_user_proto;
 125 }
 126
 127 /*
 128  * limited trace_printk()
 129  * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
 130  */
 131 static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5)
 132 {
 133         char *fmt = (char *) (long) r1;
 134         bool str_seen = false;
 135         int mod[3] = {};
 136         int fmt_cnt = 0;
 137         u64 unsafe_addr;
 138         char buf[64];
 139         int i;
 140
 141         /*
 142          * bpf_check()->check_func_arg()->check_stack_boundary()
 143          * guarantees that fmt points to bpf program stack,
 144          * fmt_size bytes of it were initialized and fmt_size > 0
 145          */
 146         if (fmt[--fmt_size] != 0)
 147                 return -EINVAL;
 148
 149         /* check format string for allowed specifiers */
 150         for (i = 0; i < fmt_size; i++) {
 151                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 152                         return -EINVAL;
 153
 154                 if (fmt[i] != '%')
 155                         continue;
 156
 157                 if (fmt_cnt >= 3)
 158                         return -EINVAL;
 159
 160                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 161                 i++;
 162                 if (fmt[i] == 'l') {
 163                         mod[fmt_cnt]++;
 164                         i++;
 165                 } else if (fmt[i] == 'p' || fmt[i] == 's') {
 166                         mod[fmt_cnt]++;
 167                         i++;
 168                         if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
 169                                 return -EINVAL;
 170                         fmt_cnt++;
 171                         if (fmt[i - 1] == 's') {
 172                                 if (str_seen)
 173                                         /* allow only one '%s' per fmt string */
 174                                         return -EINVAL;
 175                                 str_seen = true;
 176
 177                                 switch (fmt_cnt) {
 178                                 case 1:
 179                                         unsafe_addr = r3;
 180                                         r3 = (long) buf;
 181                                         break;
 182                                 case 2:
 183                                         unsafe_addr = r4;
 184                                         r4 = (long) buf;
 185                                         break;
 186                                 case 3:
 187                                         unsafe_addr = r5;
 188                                         r5 = (long) buf;
 189                                         break;
 190                                 }
 191                                 buf[0] = 0;
 192                                 strncpy_from_unsafe(buf,
 193                                                     (void *) (long) unsafe_addr,
 194                                                     sizeof(buf));
 195                         }
 196                         continue;
 197                 }
 198
 199                 if (fmt[i] == 'l') {
 200                         mod[fmt_cnt]++;
 201                         i++;
 202                 }
 203
 204                 if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
 205                         return -EINVAL;
 206                 fmt_cnt++;
 207         }
 208
 209         return __trace_printk(1/* fake ip will not be printed */, fmt,
 210                               mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3,
 211                               mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4,
 212                               mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5);
 213 }
 214
 215 static const struct bpf_func_proto bpf_trace_printk_proto = {
 216         .func           = bpf_trace_printk,
 217         .gpl_only       = true,
 218         .ret_type       = RET_INTEGER,
 219         .arg1_type      = ARG_PTR_TO_STACK,
 220         .arg2_type      = ARG_CONST_STACK_SIZE,
 221 };
 222
 223 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 224 {
 225         /*
 226          * this program might be calling bpf_trace_printk,
 227          * so allocate per-cpu printk buffers
 228          */
 229         trace_printk_init_buffers();
 230
 231         return &bpf_trace_printk_proto;
 232 }
 233
 234 static u64 bpf_perf_event_read(u64 r1, u64 flags, u64 r3, u64 r4, u64 r5)
 235 {
 236         struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
 237         struct bpf_array *array = container_of(map, struct bpf_array, map);
 238         unsigned int cpu = smp_processor_id();
 239         u64 index = flags & BPF_F_INDEX_MASK;
 240         struct bpf_event_entry *ee;
 241         struct perf_event *event;
 242
 243         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 244                 return -EINVAL;
 245         if (index == BPF_F_CURRENT_CPU)
 246                 index = cpu;
 247         if (unlikely(index >= array->map.max_entries))
 248                 return -E2BIG;
 249
 250         ee = READ_ONCE(array->ptrs[index]);
 251         if (!ee)
 252                 return -ENOENT;
 253
 254         event = ee->event;
 255         if (unlikely(event->attr.type != PERF_TYPE_HARDWARE &&
 256                      event->attr.type != PERF_TYPE_RAW))
 257                 return -EINVAL;
 258
 259         /* make sure event is local and doesn't have pmu::count */
 260         if (unlikely(event->oncpu != cpu || event->pmu->count))
 261                 return -EINVAL;
 262
 263         /*
 264          * we don't know if the function is run successfully by the
 265          * return value. It can be judged in other places, such as
 266          * eBPF programs.
 267          */
 268         return perf_event_read_local(event);
 269 }
 270
 271 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 272         .func           = bpf_perf_event_read,
 273         .gpl_only       = true,
 274         .ret_type       = RET_INTEGER,
 275         .arg1_type      = ARG_CONST_MAP_PTR,
 276         .arg2_type      = ARG_ANYTHING,
 277 };
 278
 279 static __always_inline u64
 280 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 281                         u64 flags, struct perf_raw_record *raw)
 282 {
 283         struct bpf_array *array = container_of(map, struct bpf_array, map);
 284         unsigned int cpu = smp_processor_id();
 285         u64 index = flags & BPF_F_INDEX_MASK;
 286         struct perf_sample_data sample_data;
 287         struct bpf_event_entry *ee;
 288         struct perf_event *event;
 289
 290         if (index == BPF_F_CURRENT_CPU)
 291                 index = cpu;
 292         if (unlikely(index >= array->map.max_entries))
 293                 return -E2BIG;
 294
 295         ee = READ_ONCE(array->ptrs[index]);
 296         if (!ee)
 297                 return -ENOENT;
 298
 299         event = ee->event;
 300         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 301                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 302                 return -EINVAL;
 303
 304         if (unlikely(event->oncpu != cpu))
 305                 return -EOPNOTSUPP;
 306
 307         perf_sample_data_init(&sample_data, 0, 0);
 308         sample_data.raw = raw;
 309         perf_event_output(event, &sample_data, regs);
 310         return 0;
 311 }
 312
 313 static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size)
 314 {
 315         struct pt_regs *regs = (struct pt_regs *)(long) r1;
 316         struct bpf_map *map  = (struct bpf_map *)(long) r2;
 317         void *data = (void *)(long) r4;
 318         struct perf_raw_record raw = {
 319                 .frag = {
 320                         .size = size,
 321                         .data = data,
 322                 },
 323         };
 324
 325         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 326                 return -EINVAL;
 327
 328         return __bpf_perf_event_output(regs, map, flags, &raw);
 329 }
 330
 331 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 332         .func           = bpf_perf_event_output,
 333         .gpl_only       = true,
 334         .ret_type       = RET_INTEGER,
 335         .arg1_type      = ARG_PTR_TO_CTX,
 336         .arg2_type      = ARG_CONST_MAP_PTR,
 337         .arg3_type      = ARG_ANYTHING,
 338         .arg4_type      = ARG_PTR_TO_STACK,
 339         .arg5_type      = ARG_CONST_STACK_SIZE,
 340 };
 341
 342 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
 343
 344 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
 345                      void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
 346 {
 347         struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
 348         struct perf_raw_frag frag = {
 349                 .copy           = ctx_copy,
 350                 .size           = ctx_size,
 351                 .data           = ctx,
 352         };
 353         struct perf_raw_record raw = {
 354                 .frag = {
 355                         {
 356                                 .next   = ctx_size ? &frag : NULL,
 357                         },
 358                         .size   = meta_size,
 359                         .data   = meta,
 360                 },
 361         };
 362
 363         perf_fetch_caller_regs(regs);
 364
 365         return __bpf_perf_event_output(regs, map, flags, &raw);
 366 }
 367
 368 static u64 bpf_get_current_task(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 369 {
 370         return (long) current;
 371 }
 372
 373 static const struct bpf_func_proto bpf_get_current_task_proto = {
 374         .func           = bpf_get_current_task,
 375         .gpl_only       = true,
 376         .ret_type       = RET_INTEGER,
 377 };
 378
 379 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id)
 380 {
 381         switch (func_id) {
 382         case BPF_FUNC_map_lookup_elem:
 383                 return &bpf_map_lookup_elem_proto;
 384         case BPF_FUNC_map_update_elem:
 385                 return &bpf_map_update_elem_proto;
 386         case BPF_FUNC_map_delete_elem:
 387                 return &bpf_map_delete_elem_proto;
 388         case BPF_FUNC_probe_read:
 389                 return &bpf_probe_read_proto;
 390         case BPF_FUNC_ktime_get_ns:
 391                 return &bpf_ktime_get_ns_proto;
 392         case BPF_FUNC_tail_call:
 393                 return &bpf_tail_call_proto;
 394         case BPF_FUNC_get_current_pid_tgid:
 395                 return &bpf_get_current_pid_tgid_proto;
 396         case BPF_FUNC_get_current_task:
 397                 return &bpf_get_current_task_proto;
 398         case BPF_FUNC_get_current_uid_gid:
 399                 return &bpf_get_current_uid_gid_proto;
 400         case BPF_FUNC_get_current_comm:
 401                 return &bpf_get_current_comm_proto;
 402         case BPF_FUNC_trace_printk:
 403                 return bpf_get_trace_printk_proto();
 404         case BPF_FUNC_get_smp_processor_id:
 405                 return &bpf_get_smp_processor_id_proto;
 406         case BPF_FUNC_perf_event_read:
 407                 return &bpf_perf_event_read_proto;
 408         case BPF_FUNC_probe_write_user:
 409                 return bpf_get_probe_write_proto();
 410         default:
 411                 return NULL;
 412         }
 413 }
 414
 415 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
 416 {
 417         switch (func_id) {
 418         case BPF_FUNC_perf_event_output:
 419                 return &bpf_perf_event_output_proto;
 420         case BPF_FUNC_get_stackid:
 421                 return &bpf_get_stackid_proto;
 422         default:
 423                 return tracing_func_proto(func_id);
 424         }
 425 }
 426
 427 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
 428 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 429                                         enum bpf_reg_type *reg_type)
 430 {
 431         if (off < 0 || off >= sizeof(struct pt_regs))
 432                 return false;
 433         if (type != BPF_READ)
 434                 return false;
 435         if (off % size != 0)
 436                 return false;
 437         return true;
 438 }
 439
 440 static const struct bpf_verifier_ops kprobe_prog_ops = {
 441         .get_func_proto  = kprobe_prog_func_proto,
 442         .is_valid_access = kprobe_prog_is_valid_access,
 443 };
 444
 445 static struct bpf_prog_type_list kprobe_tl = {
 446         .ops    = &kprobe_prog_ops,
 447         .type   = BPF_PROG_TYPE_KPROBE,
 448 };
 449
 450 static u64 bpf_perf_event_output_tp(u64 r1, u64 r2, u64 index, u64 r4, u64 size)
 451 {
 452         /*
 453          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
 454          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
 455          * from there and call the same bpf_perf_event_output() helper
 456          */
 457         u64 ctx = *(long *)(uintptr_t)r1;
 458
 459         return bpf_perf_event_output(ctx, r2, index, r4, size);
 460 }
 461
 462 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
 463         .func           = bpf_perf_event_output_tp,
 464         .gpl_only       = true,
 465         .ret_type       = RET_INTEGER,
 466         .arg1_type      = ARG_PTR_TO_CTX,
 467         .arg2_type      = ARG_CONST_MAP_PTR,
 468         .arg3_type      = ARG_ANYTHING,
 469         .arg4_type      = ARG_PTR_TO_STACK,
 470         .arg5_type      = ARG_CONST_STACK_SIZE,
 471 };
 472
 473 static u64 bpf_get_stackid_tp(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 474 {
 475         u64 ctx = *(long *)(uintptr_t)r1;
 476
 477         return bpf_get_stackid(ctx, r2, r3, r4, r5);
 478 }
 479
 480 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
 481         .func           = bpf_get_stackid_tp,
 482         .gpl_only       = true,
 483         .ret_type       = RET_INTEGER,
 484         .arg1_type      = ARG_PTR_TO_CTX,
 485         .arg2_type      = ARG_CONST_MAP_PTR,
 486         .arg3_type      = ARG_ANYTHING,
 487 };
 488
 489 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id)
 490 {
 491         switch (func_id) {
 492         case BPF_FUNC_perf_event_output:
 493                 return &bpf_perf_event_output_proto_tp;
 494         case BPF_FUNC_get_stackid:
 495                 return &bpf_get_stackid_proto_tp;
 496         default:
 497                 return tracing_func_proto(func_id);
 498         }
 499 }
 500
 501 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 502                                     enum bpf_reg_type *reg_type)
 503 {
 504         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
 505                 return false;
 506         if (type != BPF_READ)
 507                 return false;
 508         if (off % size != 0)
 509                 return false;
 510         return true;
 511 }
 512
 513 static const struct bpf_verifier_ops tracepoint_prog_ops = {
 514         .get_func_proto  = tp_prog_func_proto,
 515         .is_valid_access = tp_prog_is_valid_access,
 516 };
 517
 518 static struct bpf_prog_type_list tracepoint_tl = {
 519         .ops    = &tracepoint_prog_ops,
 520         .type   = BPF_PROG_TYPE_TRACEPOINT,
 521 };
 522
 523 static int __init register_kprobe_prog_ops(void)
 524 {
 525         bpf_register_prog_type(&kprobe_tl);
 526         bpf_register_prog_type(&tracepoint_tl);
 527         return 0;
 528 }
 529 late_initcall(register_kprobe_prog_ops);