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 * Only limited trace_printk() conversion specifiers allowed: 126 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s 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] != 'i' && fmt[i] != 'd' && 202 fmt[i] != 'u' && fmt[i] != 'x') 203 return -EINVAL; 204 fmt_cnt++; 205 } 206 207 return __trace_printk(1/* fake ip will not be printed */, fmt, 208 mod[0] == 2 ? arg1 : mod[0] == 1 ? (long) arg1 : (u32) arg1, 209 mod[1] == 2 ? arg2 : mod[1] == 1 ? (long) arg2 : (u32) arg2, 210 mod[2] == 2 ? arg3 : mod[2] == 1 ? (long) arg3 : (u32) arg3); 211 } 212 213 static const struct bpf_func_proto bpf_trace_printk_proto = { 214 .func = bpf_trace_printk, 215 .gpl_only = true, 216 .ret_type = RET_INTEGER, 217 .arg1_type = ARG_PTR_TO_MEM, 218 .arg2_type = ARG_CONST_SIZE, 219 }; 220 221 const struct bpf_func_proto *bpf_get_trace_printk_proto(void) 222 { 223 /* 224 * this program might be calling bpf_trace_printk, 225 * so allocate per-cpu printk buffers 226 */ 227 trace_printk_init_buffers(); 228 229 return &bpf_trace_printk_proto; 230 } 231 232 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) 233 { 234 struct bpf_array *array = container_of(map, struct bpf_array, map); 235 unsigned int cpu = smp_processor_id(); 236 u64 index = flags & BPF_F_INDEX_MASK; 237 struct bpf_event_entry *ee; 238 u64 value = 0; 239 int err; 240 241 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 242 return -EINVAL; 243 if (index == BPF_F_CURRENT_CPU) 244 index = cpu; 245 if (unlikely(index >= array->map.max_entries)) 246 return -E2BIG; 247 248 ee = READ_ONCE(array->ptrs[index]); 249 if (!ee) 250 return -ENOENT; 251 252 err = perf_event_read_local(ee->event, &value); 253 /* 254 * this api is ugly since we miss [-22..-2] range of valid 255 * counter values, but that's uapi 256 */ 257 if (err) 258 return err; 259 return value; 260 } 261 262 static const struct bpf_func_proto bpf_perf_event_read_proto = { 263 .func = bpf_perf_event_read, 264 .gpl_only = true, 265 .ret_type = RET_INTEGER, 266 .arg1_type = ARG_CONST_MAP_PTR, 267 .arg2_type = ARG_ANYTHING, 268 }; 269 270 static DEFINE_PER_CPU(struct perf_sample_data, bpf_sd); 271 272 static __always_inline u64 273 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, 274 u64 flags, struct perf_raw_record *raw) 275 { 276 struct bpf_array *array = container_of(map, struct bpf_array, map); 277 struct perf_sample_data *sd = this_cpu_ptr(&bpf_sd); 278 unsigned int cpu = smp_processor_id(); 279 u64 index = flags & BPF_F_INDEX_MASK; 280 struct bpf_event_entry *ee; 281 struct perf_event *event; 282 283 if (index == BPF_F_CURRENT_CPU) 284 index = cpu; 285 if (unlikely(index >= array->map.max_entries)) 286 return -E2BIG; 287 288 ee = READ_ONCE(array->ptrs[index]); 289 if (!ee) 290 return -ENOENT; 291 292 event = ee->event; 293 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || 294 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) 295 return -EINVAL; 296 297 if (unlikely(event->oncpu != cpu)) 298 return -EOPNOTSUPP; 299 300 perf_sample_data_init(sd, 0, 0); 301 sd->raw = raw; 302 perf_event_output(event, sd, regs); 303 return 0; 304 } 305 306 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, 307 u64, flags, void *, data, u64, size) 308 { 309 struct perf_raw_record raw = { 310 .frag = { 311 .size = size, 312 .data = data, 313 }, 314 }; 315 316 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 317 return -EINVAL; 318 319 return __bpf_perf_event_output(regs, map, flags, &raw); 320 } 321 322 static const struct bpf_func_proto bpf_perf_event_output_proto = { 323 .func = bpf_perf_event_output, 324 .gpl_only = true, 325 .ret_type = RET_INTEGER, 326 .arg1_type = ARG_PTR_TO_CTX, 327 .arg2_type = ARG_CONST_MAP_PTR, 328 .arg3_type = ARG_ANYTHING, 329 .arg4_type = ARG_PTR_TO_MEM, 330 .arg5_type = ARG_CONST_SIZE, 331 }; 332 333 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); 334 335 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 336 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) 337 { 338 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); 339 struct perf_raw_frag frag = { 340 .copy = ctx_copy, 341 .size = ctx_size, 342 .data = ctx, 343 }; 344 struct perf_raw_record raw = { 345 .frag = { 346 { 347 .next = ctx_size ? &frag : NULL, 348 }, 349 .size = meta_size, 350 .data = meta, 351 }, 352 }; 353 354 perf_fetch_caller_regs(regs); 355 356 return __bpf_perf_event_output(regs, map, flags, &raw); 357 } 358 359 BPF_CALL_0(bpf_get_current_task) 360 { 361 return (long) current; 362 } 363 364 static const struct bpf_func_proto bpf_get_current_task_proto = { 365 .func = bpf_get_current_task, 366 .gpl_only = true, 367 .ret_type = RET_INTEGER, 368 }; 369 370 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) 371 { 372 struct bpf_array *array = container_of(map, struct bpf_array, map); 373 struct cgroup *cgrp; 374 375 if (unlikely(in_interrupt())) 376 return -EINVAL; 377 if (unlikely(idx >= array->map.max_entries)) 378 return -E2BIG; 379 380 cgrp = READ_ONCE(array->ptrs[idx]); 381 if (unlikely(!cgrp)) 382 return -EAGAIN; 383 384 return task_under_cgroup_hierarchy(current, cgrp); 385 } 386 387 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { 388 .func = bpf_current_task_under_cgroup, 389 .gpl_only = false, 390 .ret_type = RET_INTEGER, 391 .arg1_type = ARG_CONST_MAP_PTR, 392 .arg2_type = ARG_ANYTHING, 393 }; 394 395 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size, 396 const void *, unsafe_ptr) 397 { 398 int ret; 399 400 /* 401 * The strncpy_from_unsafe() call will likely not fill the entire 402 * buffer, but that's okay in this circumstance as we're probing 403 * arbitrary memory anyway similar to bpf_probe_read() and might 404 * as well probe the stack. Thus, memory is explicitly cleared 405 * only in error case, so that improper users ignoring return 406 * code altogether don't copy garbage; otherwise length of string 407 * is returned that can be used for bpf_perf_event_output() et al. 408 */ 409 ret = strncpy_from_unsafe(dst, unsafe_ptr, size); 410 if (unlikely(ret < 0)) 411 memset(dst, 0, size); 412 413 return ret; 414 } 415 416 static const struct bpf_func_proto bpf_probe_read_str_proto = { 417 .func = bpf_probe_read_str, 418 .gpl_only = true, 419 .ret_type = RET_INTEGER, 420 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 421 .arg2_type = ARG_CONST_SIZE, 422 .arg3_type = ARG_ANYTHING, 423 }; 424 425 static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id) 426 { 427 switch (func_id) { 428 case BPF_FUNC_map_lookup_elem: 429 return &bpf_map_lookup_elem_proto; 430 case BPF_FUNC_map_update_elem: 431 return &bpf_map_update_elem_proto; 432 case BPF_FUNC_map_delete_elem: 433 return &bpf_map_delete_elem_proto; 434 case BPF_FUNC_probe_read: 435 return &bpf_probe_read_proto; 436 case BPF_FUNC_ktime_get_ns: 437 return &bpf_ktime_get_ns_proto; 438 case BPF_FUNC_tail_call: 439 return &bpf_tail_call_proto; 440 case BPF_FUNC_get_current_pid_tgid: 441 return &bpf_get_current_pid_tgid_proto; 442 case BPF_FUNC_get_current_task: 443 return &bpf_get_current_task_proto; 444 case BPF_FUNC_get_current_uid_gid: 445 return &bpf_get_current_uid_gid_proto; 446 case BPF_FUNC_get_current_comm: 447 return &bpf_get_current_comm_proto; 448 case BPF_FUNC_trace_printk: 449 return bpf_get_trace_printk_proto(); 450 case BPF_FUNC_get_smp_processor_id: 451 return &bpf_get_smp_processor_id_proto; 452 case BPF_FUNC_get_numa_node_id: 453 return &bpf_get_numa_node_id_proto; 454 case BPF_FUNC_perf_event_read: 455 return &bpf_perf_event_read_proto; 456 case BPF_FUNC_probe_write_user: 457 return bpf_get_probe_write_proto(); 458 case BPF_FUNC_current_task_under_cgroup: 459 return &bpf_current_task_under_cgroup_proto; 460 case BPF_FUNC_get_prandom_u32: 461 return &bpf_get_prandom_u32_proto; 462 case BPF_FUNC_probe_read_str: 463 return &bpf_probe_read_str_proto; 464 default: 465 return NULL; 466 } 467 } 468 469 static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id) 470 { 471 switch (func_id) { 472 case BPF_FUNC_perf_event_output: 473 return &bpf_perf_event_output_proto; 474 case BPF_FUNC_get_stackid: 475 return &bpf_get_stackid_proto; 476 default: 477 return tracing_func_proto(func_id); 478 } 479 } 480 481 /* bpf+kprobe programs can access fields of 'struct pt_regs' */ 482 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 483 struct bpf_insn_access_aux *info) 484 { 485 if (off < 0 || off >= sizeof(struct pt_regs)) 486 return false; 487 if (type != BPF_READ) 488 return false; 489 if (off % size != 0) 490 return false; 491 /* 492 * Assertion for 32 bit to make sure last 8 byte access 493 * (BPF_DW) to the last 4 byte member is disallowed. 494 */ 495 if (off + size > sizeof(struct pt_regs)) 496 return false; 497 498 return true; 499 } 500 501 const struct bpf_verifier_ops kprobe_prog_ops = { 502 .get_func_proto = kprobe_prog_func_proto, 503 .is_valid_access = kprobe_prog_is_valid_access, 504 }; 505 506 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, 507 u64, flags, void *, data, u64, size) 508 { 509 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 510 511 /* 512 * r1 points to perf tracepoint buffer where first 8 bytes are hidden 513 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it 514 * from there and call the same bpf_perf_event_output() helper inline. 515 */ 516 return ____bpf_perf_event_output(regs, map, flags, data, size); 517 } 518 519 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { 520 .func = bpf_perf_event_output_tp, 521 .gpl_only = true, 522 .ret_type = RET_INTEGER, 523 .arg1_type = ARG_PTR_TO_CTX, 524 .arg2_type = ARG_CONST_MAP_PTR, 525 .arg3_type = ARG_ANYTHING, 526 .arg4_type = ARG_PTR_TO_MEM, 527 .arg5_type = ARG_CONST_SIZE, 528 }; 529 530 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, 531 u64, flags) 532 { 533 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 534 535 /* 536 * Same comment as in bpf_perf_event_output_tp(), only that this time 537 * the other helper's function body cannot be inlined due to being 538 * external, thus we need to call raw helper function. 539 */ 540 return bpf_get_stackid((unsigned long) regs, (unsigned long) map, 541 flags, 0, 0); 542 } 543 544 static const struct bpf_func_proto bpf_get_stackid_proto_tp = { 545 .func = bpf_get_stackid_tp, 546 .gpl_only = true, 547 .ret_type = RET_INTEGER, 548 .arg1_type = ARG_PTR_TO_CTX, 549 .arg2_type = ARG_CONST_MAP_PTR, 550 .arg3_type = ARG_ANYTHING, 551 }; 552 553 static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id) 554 { 555 switch (func_id) { 556 case BPF_FUNC_perf_event_output: 557 return &bpf_perf_event_output_proto_tp; 558 case BPF_FUNC_get_stackid: 559 return &bpf_get_stackid_proto_tp; 560 default: 561 return tracing_func_proto(func_id); 562 } 563 } 564 565 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, 566 struct bpf_insn_access_aux *info) 567 { 568 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) 569 return false; 570 if (type != BPF_READ) 571 return false; 572 if (off % size != 0) 573 return false; 574 575 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); 576 return true; 577 } 578 579 const struct bpf_verifier_ops tracepoint_prog_ops = { 580 .get_func_proto = tp_prog_func_proto, 581 .is_valid_access = tp_prog_is_valid_access, 582 }; 583 584 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 585 struct bpf_insn_access_aux *info) 586 { 587 const int size_sp = FIELD_SIZEOF(struct bpf_perf_event_data, 588 sample_period); 589 590 if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) 591 return false; 592 if (type != BPF_READ) 593 return false; 594 if (off % size != 0) 595 return false; 596 597 switch (off) { 598 case bpf_ctx_range(struct bpf_perf_event_data, sample_period): 599 bpf_ctx_record_field_size(info, size_sp); 600 if (!bpf_ctx_narrow_access_ok(off, size, size_sp)) 601 return false; 602 break; 603 default: 604 if (size != sizeof(long)) 605 return false; 606 } 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, u32 *target_size) 615 { 616 struct bpf_insn *insn = insn_buf; 617 618 switch (si->off) { 619 case offsetof(struct bpf_perf_event_data, sample_period): 620 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 621 data), si->dst_reg, si->src_reg, 622 offsetof(struct bpf_perf_event_data_kern, data)); 623 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 624 bpf_target_off(struct perf_sample_data, period, 8, 625 target_size)); 626 break; 627 default: 628 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 629 regs), si->dst_reg, si->src_reg, 630 offsetof(struct bpf_perf_event_data_kern, regs)); 631 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, 632 si->off); 633 break; 634 } 635 636 return insn - insn_buf; 637 } 638 639 const struct bpf_verifier_ops perf_event_prog_ops = { 640 .get_func_proto = tp_prog_func_proto, 641 .is_valid_access = pe_prog_is_valid_access, 642 .convert_ctx_access = pe_prog_convert_ctx_access, 643 }; 644