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 /*
  85  * limited trace_printk()
  86  * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
  87  */
  88 static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5)
  89 {
  90         char *fmt = (char *) (long) r1;
  91         bool str_seen = false;
  92         int mod[3] = {};
  93         int fmt_cnt = 0;
  94         u64 unsafe_addr;
  95         char buf[64];
  96         int i;
  97
  98         /*
  99          * bpf_check()->check_func_arg()->check_stack_boundary()
 100          * guarantees that fmt points to bpf program stack,
 101          * fmt_size bytes of it were initialized and fmt_size > 0
 102          */
 103         if (fmt[--fmt_size] != 0)
 104                 return -EINVAL;
 105
 106         /* check format string for allowed specifiers */
 107         for (i = 0; i < fmt_size; i++) {
 108                 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
 109                         return -EINVAL;
 110
 111                 if (fmt[i] != '%')
 112                         continue;
 113
 114                 if (fmt_cnt >= 3)
 115                         return -EINVAL;
 116
 117                 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
 118                 i++;
 119                 if (fmt[i] == 'l') {
 120                         mod[fmt_cnt]++;
 121                         i++;
 122                 } else if (fmt[i] == 'p' || fmt[i] == 's') {
 123                         mod[fmt_cnt]++;
 124                         i++;
 125                         if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
 126                                 return -EINVAL;
 127                         fmt_cnt++;
 128                         if (fmt[i - 1] == 's') {
 129                                 if (str_seen)
 130                                         /* allow only one '%s' per fmt string */
 131                                         return -EINVAL;
 132                                 str_seen = true;
 133
 134                                 switch (fmt_cnt) {
 135                                 case 1:
 136                                         unsafe_addr = r3;
 137                                         r3 = (long) buf;
 138                                         break;
 139                                 case 2:
 140                                         unsafe_addr = r4;
 141                                         r4 = (long) buf;
 142                                         break;
 143                                 case 3:
 144                                         unsafe_addr = r5;
 145                                         r5 = (long) buf;
 146                                         break;
 147                                 }
 148                                 buf[0] = 0;
 149                                 strncpy_from_unsafe(buf,
 150                                                     (void *) (long) unsafe_addr,
 151                                                     sizeof(buf));
 152                         }
 153                         continue;
 154                 }
 155
 156                 if (fmt[i] == 'l') {
 157                         mod[fmt_cnt]++;
 158                         i++;
 159                 }
 160
 161                 if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
 162                         return -EINVAL;
 163                 fmt_cnt++;
 164         }
 165
 166         return __trace_printk(1/* fake ip will not be printed */, fmt,
 167                               mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3,
 168                               mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4,
 169                               mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5);
 170 }
 171
 172 static const struct bpf_func_proto bpf_trace_printk_proto = {
 173         .func           = bpf_trace_printk,
 174         .gpl_only       = true,
 175         .ret_type       = RET_INTEGER,
 176         .arg1_type      = ARG_PTR_TO_STACK,
 177         .arg2_type      = ARG_CONST_STACK_SIZE,
 178 };
 179
 180 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 181 {
 182         /*
 183          * this program might be calling bpf_trace_printk,
 184          * so allocate per-cpu printk buffers
 185          */
 186         trace_printk_init_buffers();
 187
 188         return &bpf_trace_printk_proto;
 189 }
 190
 191 static u64 bpf_perf_event_read(u64 r1, u64 flags, u64 r3, u64 r4, u64 r5)
 192 {
 193         struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
 194         struct bpf_array *array = container_of(map, struct bpf_array, map);
 195         unsigned int cpu = smp_processor_id();
 196         u64 index = flags & BPF_F_INDEX_MASK;
 197         struct bpf_event_entry *ee;
 198         struct perf_event *event;
 199
 200         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 201                 return -EINVAL;
 202         if (index == BPF_F_CURRENT_CPU)
 203                 index = cpu;
 204         if (unlikely(index >= array->map.max_entries))
 205                 return -E2BIG;
 206
 207         ee = READ_ONCE(array->ptrs[index]);
 208         if (!ee)
 209                 return -ENOENT;
 210
 211         event = ee->event;
 212         if (unlikely(event->attr.type != PERF_TYPE_HARDWARE &&
 213                      event->attr.type != PERF_TYPE_RAW))
 214                 return -EINVAL;
 215
 216         /* make sure event is local and doesn't have pmu::count */
 217         if (unlikely(event->oncpu != cpu || event->pmu->count))
 218                 return -EINVAL;
 219
 220         /*
 221          * we don't know if the function is run successfully by the
 222          * return value. It can be judged in other places, such as
 223          * eBPF programs.
 224          */
 225         return perf_event_read_local(event);
 226 }
 227
 228 static const struct bpf_func_proto bpf_perf_event_read_proto = {
 229         .func           = bpf_perf_event_read,
 230         .gpl_only       = true,
 231         .ret_type       = RET_INTEGER,
 232         .arg1_type      = ARG_CONST_MAP_PTR,
 233         .arg2_type      = ARG_ANYTHING,
 234 };
 235
 236 static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size)
 237 {
 238         struct pt_regs *regs = (struct pt_regs *) (long) r1;
 239         struct bpf_map *map = (struct bpf_map *) (long) r2;
 240         struct bpf_array *array = container_of(map, struct bpf_array, map);
 241         unsigned int cpu = smp_processor_id();
 242         u64 index = flags & BPF_F_INDEX_MASK;
 243         void *data = (void *) (long) r4;
 244         struct perf_sample_data sample_data;
 245         struct bpf_event_entry *ee;
 246         struct perf_event *event;
 247         struct perf_raw_record raw = {
 248                 .size = size,
 249                 .data = data,
 250         };
 251
 252         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 253                 return -EINVAL;
 254         if (index == BPF_F_CURRENT_CPU)
 255                 index = cpu;
 256         if (unlikely(index >= array->map.max_entries))
 257                 return -E2BIG;
 258
 259         ee = READ_ONCE(array->ptrs[index]);
 260         if (!ee)
 261                 return -ENOENT;
 262
 263         event = ee->event;
 264         if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 265                      event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 266                 return -EINVAL;
 267
 268         if (unlikely(event->oncpu != cpu))
 269                 return -EOPNOTSUPP;
 270
 271         perf_sample_data_init(&sample_data, 0, 0);
 272         sample_data.raw = &raw;
 273         perf_event_output(event, &sample_data, regs);
 274         return 0;
 275 }
 276
 277 static const struct bpf_func_proto bpf_perf_event_output_proto = {
 278         .func           = bpf_perf_event_output,
 279         .gpl_only       = true,
 280         .ret_type       = RET_INTEGER,
 281         .arg1_type      = ARG_PTR_TO_CTX,
 282         .arg2_type      = ARG_CONST_MAP_PTR,
 283         .arg3_type      = ARG_ANYTHING,
 284         .arg4_type      = ARG_PTR_TO_STACK,
 285         .arg5_type      = ARG_CONST_STACK_SIZE,
 286 };
 287
 288 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs);
 289
 290 static u64 bpf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size)
 291 {
 292         struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs);
 293
 294         perf_fetch_caller_regs(regs);
 295
 296         return bpf_perf_event_output((long)regs, r2, flags, r4, size);
 297 }
 298
 299 static const struct bpf_func_proto bpf_event_output_proto = {
 300         .func           = bpf_event_output,
 301         .gpl_only       = true,
 302         .ret_type       = RET_INTEGER,
 303         .arg1_type      = ARG_PTR_TO_CTX,
 304         .arg2_type      = ARG_CONST_MAP_PTR,
 305         .arg3_type      = ARG_ANYTHING,
 306         .arg4_type      = ARG_PTR_TO_STACK,
 307         .arg5_type      = ARG_CONST_STACK_SIZE,
 308 };
 309
 310 const struct bpf_func_proto *bpf_get_event_output_proto(void)
 311 {
 312         return &bpf_event_output_proto;
 313 }
 314
 315 static u64 bpf_get_current_task(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 316 {
 317         return (long) current;
 318 }
 319
 320 static const struct bpf_func_proto bpf_get_current_task_proto = {
 321         .func           = bpf_get_current_task,
 322         .gpl_only       = true,
 323         .ret_type       = RET_INTEGER,
 324 };
 325
 326 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id)
 327 {
 328         switch (func_id) {
 329         case BPF_FUNC_map_lookup_elem:
 330                 return &bpf_map_lookup_elem_proto;
 331         case BPF_FUNC_map_update_elem:
 332                 return &bpf_map_update_elem_proto;
 333         case BPF_FUNC_map_delete_elem:
 334                 return &bpf_map_delete_elem_proto;
 335         case BPF_FUNC_probe_read:
 336                 return &bpf_probe_read_proto;
 337         case BPF_FUNC_ktime_get_ns:
 338                 return &bpf_ktime_get_ns_proto;
 339         case BPF_FUNC_tail_call:
 340                 return &bpf_tail_call_proto;
 341         case BPF_FUNC_get_current_pid_tgid:
 342                 return &bpf_get_current_pid_tgid_proto;
 343         case BPF_FUNC_get_current_task:
 344                 return &bpf_get_current_task_proto;
 345         case BPF_FUNC_get_current_uid_gid:
 346                 return &bpf_get_current_uid_gid_proto;
 347         case BPF_FUNC_get_current_comm:
 348                 return &bpf_get_current_comm_proto;
 349         case BPF_FUNC_trace_printk:
 350                 return bpf_get_trace_printk_proto();
 351         case BPF_FUNC_get_smp_processor_id:
 352                 return &bpf_get_smp_processor_id_proto;
 353         case BPF_FUNC_perf_event_read:
 354                 return &bpf_perf_event_read_proto;
 355         default:
 356                 return NULL;
 357         }
 358 }
 359
 360 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
 361 {
 362         switch (func_id) {
 363         case BPF_FUNC_perf_event_output:
 364                 return &bpf_perf_event_output_proto;
 365         case BPF_FUNC_get_stackid:
 366                 return &bpf_get_stackid_proto;
 367         default:
 368                 return tracing_func_proto(func_id);
 369         }
 370 }
 371
 372 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
 373 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 374                                         enum bpf_reg_type *reg_type)
 375 {
 376         if (off < 0 || off >= sizeof(struct pt_regs))
 377                 return false;
 378         if (type != BPF_READ)
 379                 return false;
 380         if (off % size != 0)
 381                 return false;
 382         return true;
 383 }
 384
 385 static const struct bpf_verifier_ops kprobe_prog_ops = {
 386         .get_func_proto  = kprobe_prog_func_proto,
 387         .is_valid_access = kprobe_prog_is_valid_access,
 388 };
 389
 390 static struct bpf_prog_type_list kprobe_tl = {
 391         .ops    = &kprobe_prog_ops,
 392         .type   = BPF_PROG_TYPE_KPROBE,
 393 };
 394
 395 static u64 bpf_perf_event_output_tp(u64 r1, u64 r2, u64 index, u64 r4, u64 size)
 396 {
 397         /*
 398          * r1 points to perf tracepoint buffer where first 8 bytes are hidden
 399          * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
 400          * from there and call the same bpf_perf_event_output() helper
 401          */
 402         u64 ctx = *(long *)(uintptr_t)r1;
 403
 404         return bpf_perf_event_output(ctx, r2, index, r4, size);
 405 }
 406
 407 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
 408         .func           = bpf_perf_event_output_tp,
 409         .gpl_only       = true,
 410         .ret_type       = RET_INTEGER,
 411         .arg1_type      = ARG_PTR_TO_CTX,
 412         .arg2_type      = ARG_CONST_MAP_PTR,
 413         .arg3_type      = ARG_ANYTHING,
 414         .arg4_type      = ARG_PTR_TO_STACK,
 415         .arg5_type      = ARG_CONST_STACK_SIZE,
 416 };
 417
 418 static u64 bpf_get_stackid_tp(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 419 {
 420         u64 ctx = *(long *)(uintptr_t)r1;
 421
 422         return bpf_get_stackid(ctx, r2, r3, r4, r5);
 423 }
 424
 425 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
 426         .func           = bpf_get_stackid_tp,
 427         .gpl_only       = true,
 428         .ret_type       = RET_INTEGER,
 429         .arg1_type      = ARG_PTR_TO_CTX,
 430         .arg2_type      = ARG_CONST_MAP_PTR,
 431         .arg3_type      = ARG_ANYTHING,
 432 };
 433
 434 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id)
 435 {
 436         switch (func_id) {
 437         case BPF_FUNC_perf_event_output:
 438                 return &bpf_perf_event_output_proto_tp;
 439         case BPF_FUNC_get_stackid:
 440                 return &bpf_get_stackid_proto_tp;
 441         default:
 442                 return tracing_func_proto(func_id);
 443         }
 444 }
 445
 446 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
 447                                     enum bpf_reg_type *reg_type)
 448 {
 449         if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
 450                 return false;
 451         if (type != BPF_READ)
 452                 return false;
 453         if (off % size != 0)
 454                 return false;
 455         return true;
 456 }
 457
 458 static const struct bpf_verifier_ops tracepoint_prog_ops = {
 459         .get_func_proto  = tp_prog_func_proto,
 460         .is_valid_access = tp_prog_is_valid_access,
 461 };
 462
 463 static struct bpf_prog_type_list tracepoint_tl = {
 464         .ops    = &tracepoint_prog_ops,
 465         .type   = BPF_PROG_TYPE_TRACEPOINT,
 466 };
 467
 468 static int __init register_kprobe_prog_ops(void)
 469 {
 470         bpf_register_prog_type(&kprobe_tl);
 471         bpf_register_prog_type(&tracepoint_tl);
 472         return 0;
 473 }
 474 late_initcall(register_kprobe_prog_ops);