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_UNINIT_MEM, 80 .arg2_type = ARG_CONST_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(uaccess_kernel())) 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_MEM, 113 .arg3_type = ARG_CONST_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_MEM, 217 .arg2_type = ARG_CONST_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 u64 value = 0; 238 int err; 239 240 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 241 return -EINVAL; 242 if (index == BPF_F_CURRENT_CPU) 243 index = cpu; 244 if (unlikely(index >= array->map.max_entries)) 245 return -E2BIG; 246 247 ee = READ_ONCE(array->ptrs[index]); 248 if (!ee) 249 return -ENOENT; 250 251 err = perf_event_read_local(ee->event, &value); 252 /* 253 * this api is ugly since we miss [-22..-2] range of valid 254 * counter values, but that's uapi 255 */ 256 if (err) 257 return err; 258 return value; 259 } 260 261 static const struct bpf_func_proto bpf_perf_event_read_proto = { 262 .func = bpf_perf_event_read, 263 .gpl_only = true, 264 .ret_type = RET_INTEGER, 265 .arg1_type = ARG_CONST_MAP_PTR, 266 .arg2_type = ARG_ANYTHING, 267 }; 268 269 static DEFINE_PER_CPU(struct perf_sample_data, bpf_sd); 270 271 static __always_inline u64 272 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, 273 u64 flags, struct perf_raw_record *raw) 274 { 275 struct bpf_array *array = container_of(map, struct bpf_array, map); 276 struct perf_sample_data *sd = this_cpu_ptr(&bpf_sd); 277 unsigned int cpu = smp_processor_id(); 278 u64 index = flags & BPF_F_INDEX_MASK; 279 struct bpf_event_entry *ee; 280 struct perf_event *event; 281 282 if (index == BPF_F_CURRENT_CPU) 283 index = cpu; 284 if (unlikely(index >= array->map.max_entries)) 285 return -E2BIG; 286 287 ee = READ_ONCE(array->ptrs[index]); 288 if (!ee) 289 return -ENOENT; 290 291 event = ee->event; 292 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || 293 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) 294 return -EINVAL; 295 296 if (unlikely(event->oncpu != cpu)) 297 return -EOPNOTSUPP; 298 299 perf_sample_data_init(sd, 0, 0); 300 sd->raw = raw; 301 perf_event_output(event, sd, regs); 302 return 0; 303 } 304 305 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, 306 u64, flags, void *, data, u64, size) 307 { 308 struct perf_raw_record raw = { 309 .frag = { 310 .size = size, 311 .data = data, 312 }, 313 }; 314 315 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 316 return -EINVAL; 317 318 return __bpf_perf_event_output(regs, map, flags, &raw); 319 } 320 321 static const struct bpf_func_proto bpf_perf_event_output_proto = { 322 .func = bpf_perf_event_output, 323 .gpl_only = true, 324 .ret_type = RET_INTEGER, 325 .arg1_type = ARG_PTR_TO_CTX, 326 .arg2_type = ARG_CONST_MAP_PTR, 327 .arg3_type = ARG_ANYTHING, 328 .arg4_type = ARG_PTR_TO_MEM, 329 .arg5_type = ARG_CONST_SIZE, 330 }; 331 332 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); 333 334 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 335 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) 336 { 337 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); 338 struct perf_raw_frag frag = { 339 .copy = ctx_copy, 340 .size = ctx_size, 341 .data = ctx, 342 }; 343 struct perf_raw_record raw = { 344 .frag = { 345 { 346 .next = ctx_size ? &frag : NULL, 347 }, 348 .size = meta_size, 349 .data = meta, 350 }, 351 }; 352 353 perf_fetch_caller_regs(regs); 354 355 return __bpf_perf_event_output(regs, map, flags, &raw); 356 } 357 358 BPF_CALL_0(bpf_get_current_task) 359 { 360 return (long) current; 361 } 362 363 static const struct bpf_func_proto bpf_get_current_task_proto = { 364 .func = bpf_get_current_task, 365 .gpl_only = true, 366 .ret_type = RET_INTEGER, 367 }; 368 369 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) 370 { 371 struct bpf_array *array = container_of(map, struct bpf_array, map); 372 struct cgroup *cgrp; 373 374 if (unlikely(in_interrupt())) 375 return -EINVAL; 376 if (unlikely(idx >= array->map.max_entries)) 377 return -E2BIG; 378 379 cgrp = READ_ONCE(array->ptrs[idx]); 380 if (unlikely(!cgrp)) 381 return -EAGAIN; 382 383 return task_under_cgroup_hierarchy(current, cgrp); 384 } 385 386 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { 387 .func = bpf_current_task_under_cgroup, 388 .gpl_only = false, 389 .ret_type = RET_INTEGER, 390 .arg1_type = ARG_CONST_MAP_PTR, 391 .arg2_type = ARG_ANYTHING, 392 }; 393 394 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size, 395 const void *, unsafe_ptr) 396 { 397 int ret; 398 399 /* 400 * The strncpy_from_unsafe() call will likely not fill the entire 401 * buffer, but that's okay in this circumstance as we're probing 402 * arbitrary memory anyway similar to bpf_probe_read() and might 403 * as well probe the stack. Thus, memory is explicitly cleared 404 * only in error case, so that improper users ignoring return 405 * code altogether don't copy garbage; otherwise length of string 406 * is returned that can be used for bpf_perf_event_output() et al. 407 */ 408 ret = strncpy_from_unsafe(dst, unsafe_ptr, size); 409 if (unlikely(ret < 0)) 410 memset(dst, 0, size); 411 412 return ret; 413 } 414 415 static const struct bpf_func_proto bpf_probe_read_str_proto = { 416 .func = bpf_probe_read_str, 417 .gpl_only = true, 418 .ret_type = RET_INTEGER, 419 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 420 .arg2_type = ARG_CONST_SIZE, 421 .arg3_type = ARG_ANYTHING, 422 }; 423 424 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id) 425 { 426 switch (func_id) { 427 case BPF_FUNC_map_lookup_elem: 428 return &bpf_map_lookup_elem_proto; 429 case BPF_FUNC_map_update_elem: 430 return &bpf_map_update_elem_proto; 431 case BPF_FUNC_map_delete_elem: 432 return &bpf_map_delete_elem_proto; 433 case BPF_FUNC_probe_read: 434 return &bpf_probe_read_proto; 435 case BPF_FUNC_ktime_get_ns: 436 return &bpf_ktime_get_ns_proto; 437 case BPF_FUNC_tail_call: 438 return &bpf_tail_call_proto; 439 case BPF_FUNC_get_current_pid_tgid: 440 return &bpf_get_current_pid_tgid_proto; 441 case BPF_FUNC_get_current_task: 442 return &bpf_get_current_task_proto; 443 case BPF_FUNC_get_current_uid_gid: 444 return &bpf_get_current_uid_gid_proto; 445 case BPF_FUNC_get_current_comm: 446 return &bpf_get_current_comm_proto; 447 case BPF_FUNC_trace_printk: 448 return bpf_get_trace_printk_proto(); 449 case BPF_FUNC_get_smp_processor_id: 450 return &bpf_get_smp_processor_id_proto; 451 case BPF_FUNC_get_numa_node_id: 452 return &bpf_get_numa_node_id_proto; 453 case BPF_FUNC_perf_event_read: 454 return &bpf_perf_event_read_proto; 455 case BPF_FUNC_probe_write_user: 456 return bpf_get_probe_write_proto(); 457 case BPF_FUNC_current_task_under_cgroup: 458 return &bpf_current_task_under_cgroup_proto; 459 case BPF_FUNC_get_prandom_u32: 460 return &bpf_get_prandom_u32_proto; 461 case BPF_FUNC_probe_read_str: 462 return &bpf_probe_read_str_proto; 463 default: 464 return NULL; 465 } 466 } 467 468 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id) 469 { 470 switch (func_id) { 471 case BPF_FUNC_perf_event_output: 472 return &bpf_perf_event_output_proto; 473 case BPF_FUNC_get_stackid: 474 return &bpf_get_stackid_proto; 475 default: 476 return tracing_func_proto(func_id); 477 } 478 } 479 480 /* bpf+kprobe programs can access fields of 'struct pt_regs' */ 481 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 482 enum bpf_reg_type *reg_type, int *ctx_field_size) 483 { 484 if (off < 0 || off >= sizeof(struct pt_regs)) 485 return false; 486 if (type != BPF_READ) 487 return false; 488 if (off % size != 0) 489 return false; 490 /* 491 * Assertion for 32 bit to make sure last 8 byte access 492 * (BPF_DW) to the last 4 byte member is disallowed. 493 */ 494 if (off + size > sizeof(struct pt_regs)) 495 return false; 496 497 return true; 498 } 499 500 const struct bpf_verifier_ops kprobe_prog_ops = { 501 .get_func_proto = kprobe_prog_func_proto, 502 .is_valid_access = kprobe_prog_is_valid_access, 503 }; 504 505 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, 506 u64, flags, void *, data, u64, size) 507 { 508 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 509 510 /* 511 * r1 points to perf tracepoint buffer where first 8 bytes are hidden 512 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it 513 * from there and call the same bpf_perf_event_output() helper inline. 514 */ 515 return ____bpf_perf_event_output(regs, map, flags, data, size); 516 } 517 518 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { 519 .func = bpf_perf_event_output_tp, 520 .gpl_only = true, 521 .ret_type = RET_INTEGER, 522 .arg1_type = ARG_PTR_TO_CTX, 523 .arg2_type = ARG_CONST_MAP_PTR, 524 .arg3_type = ARG_ANYTHING, 525 .arg4_type = ARG_PTR_TO_MEM, 526 .arg5_type = ARG_CONST_SIZE, 527 }; 528 529 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, 530 u64, flags) 531 { 532 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 533 534 /* 535 * Same comment as in bpf_perf_event_output_tp(), only that this time 536 * the other helper's function body cannot be inlined due to being 537 * external, thus we need to call raw helper function. 538 */ 539 return bpf_get_stackid((unsigned long) regs, (unsigned long) map, 540 flags, 0, 0); 541 } 542 543 static const struct bpf_func_proto bpf_get_stackid_proto_tp = { 544 .func = bpf_get_stackid_tp, 545 .gpl_only = true, 546 .ret_type = RET_INTEGER, 547 .arg1_type = ARG_PTR_TO_CTX, 548 .arg2_type = ARG_CONST_MAP_PTR, 549 .arg3_type = ARG_ANYTHING, 550 }; 551 552 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id) 553 { 554 switch (func_id) { 555 case BPF_FUNC_perf_event_output: 556 return &bpf_perf_event_output_proto_tp; 557 case BPF_FUNC_get_stackid: 558 return &bpf_get_stackid_proto_tp; 559 default: 560 return tracing_func_proto(func_id); 561 } 562 } 563 564 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, 565 enum bpf_reg_type *reg_type, int *ctx_field_size) 566 { 567 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) 568 return false; 569 if (type != BPF_READ) 570 return false; 571 if (off % size != 0) 572 return false; 573 574 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); 575 return true; 576 } 577 578 const struct bpf_verifier_ops tracepoint_prog_ops = { 579 .get_func_proto = tp_prog_func_proto, 580 .is_valid_access = tp_prog_is_valid_access, 581 }; 582 583 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 584 enum bpf_reg_type *reg_type, int *ctx_field_size) 585 { 586 int sample_period_off; 587 588 if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) 589 return false; 590 if (type != BPF_READ) 591 return false; 592 if (off % size != 0) 593 return false; 594 595 /* permit 1, 2, 4 byte narrower and 8 normal read access to sample_period */ 596 sample_period_off = offsetof(struct bpf_perf_event_data, sample_period); 597 if (off >= sample_period_off && off < sample_period_off + sizeof(__u64)) { 598 *ctx_field_size = 8; 599 #ifdef __LITTLE_ENDIAN 600 return (off & 0x7) == 0 && size <= 8 && (size & (size - 1)) == 0; 601 #else 602 return ((off & 0x7) + size) == 8 && size <= 8 && (size & (size - 1)) == 0; 603 #endif 604 } else { 605 if (size != sizeof(long)) 606 return false; 607 } 608 return true; 609 } 610 611 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, 612 const struct bpf_insn *si, 613 struct bpf_insn *insn_buf, 614 struct bpf_prog *prog) 615 { 616 struct bpf_insn *insn = insn_buf; 617 618 switch (si->off) { 619 case offsetof(struct bpf_perf_event_data, sample_period): 620 BUILD_BUG_ON(FIELD_SIZEOF(struct perf_sample_data, period) != sizeof(u64)); 621 622 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 623 data), si->dst_reg, si->src_reg, 624 offsetof(struct bpf_perf_event_data_kern, data)); 625 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 626 offsetof(struct perf_sample_data, period)); 627 break; 628 default: 629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 630 regs), si->dst_reg, si->src_reg, 631 offsetof(struct bpf_perf_event_data_kern, regs)); 632 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, 633 si->off); 634 break; 635 } 636 637 return insn - insn_buf; 638 } 639 640 const struct bpf_verifier_ops perf_event_prog_ops = { 641 .get_func_proto = tp_prog_func_proto, 642 .is_valid_access = pe_prog_is_valid_access, 643 .convert_ctx_access = pe_prog_convert_ctx_access, 644 }; 645