1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com 3 * Copyright (c) 2016 Facebook 4 */ 5 #include <linux/kernel.h> 6 #include <linux/types.h> 7 #include <linux/slab.h> 8 #include <linux/bpf.h> 9 #include <linux/bpf_perf_event.h> 10 #include <linux/filter.h> 11 #include <linux/uaccess.h> 12 #include <linux/ctype.h> 13 #include <linux/kprobes.h> 14 #include <linux/syscalls.h> 15 #include <linux/error-injection.h> 16 17 #include <asm/tlb.h> 18 19 #include "trace_probe.h" 20 #include "trace.h" 21 22 #ifdef CONFIG_MODULES 23 struct bpf_trace_module { 24 struct module *module; 25 struct list_head list; 26 }; 27 28 static LIST_HEAD(bpf_trace_modules); 29 static DEFINE_MUTEX(bpf_module_mutex); 30 31 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) 32 { 33 struct bpf_raw_event_map *btp, *ret = NULL; 34 struct bpf_trace_module *btm; 35 unsigned int i; 36 37 mutex_lock(&bpf_module_mutex); 38 list_for_each_entry(btm, &bpf_trace_modules, list) { 39 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) { 40 btp = &btm->module->bpf_raw_events[i]; 41 if (!strcmp(btp->tp->name, name)) { 42 if (try_module_get(btm->module)) 43 ret = btp; 44 goto out; 45 } 46 } 47 } 48 out: 49 mutex_unlock(&bpf_module_mutex); 50 return ret; 51 } 52 #else 53 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) 54 { 55 return NULL; 56 } 57 #endif /* CONFIG_MODULES */ 58 59 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 60 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 61 62 /** 63 * trace_call_bpf - invoke BPF program 64 * @call: tracepoint event 65 * @ctx: opaque context pointer 66 * 67 * kprobe handlers execute BPF programs via this helper. 68 * Can be used from static tracepoints in the future. 69 * 70 * Return: BPF programs always return an integer which is interpreted by 71 * kprobe handler as: 72 * 0 - return from kprobe (event is filtered out) 73 * 1 - store kprobe event into ring buffer 74 * Other values are reserved and currently alias to 1 75 */ 76 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx) 77 { 78 unsigned int ret; 79 80 if (in_nmi()) /* not supported yet */ 81 return 1; 82 83 preempt_disable(); 84 85 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { 86 /* 87 * since some bpf program is already running on this cpu, 88 * don't call into another bpf program (same or different) 89 * and don't send kprobe event into ring-buffer, 90 * so return zero here 91 */ 92 ret = 0; 93 goto out; 94 } 95 96 /* 97 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock 98 * to all call sites, we did a bpf_prog_array_valid() there to check 99 * whether call->prog_array is empty or not, which is 100 * a heurisitc to speed up execution. 101 * 102 * If bpf_prog_array_valid() fetched prog_array was 103 * non-NULL, we go into trace_call_bpf() and do the actual 104 * proper rcu_dereference() under RCU lock. 105 * If it turns out that prog_array is NULL then, we bail out. 106 * For the opposite, if the bpf_prog_array_valid() fetched pointer 107 * was NULL, you'll skip the prog_array with the risk of missing 108 * out of events when it was updated in between this and the 109 * rcu_dereference() which is accepted risk. 110 */ 111 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN); 112 113 out: 114 __this_cpu_dec(bpf_prog_active); 115 preempt_enable(); 116 117 return ret; 118 } 119 EXPORT_SYMBOL_GPL(trace_call_bpf); 120 121 #ifdef CONFIG_BPF_KPROBE_OVERRIDE 122 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc) 123 { 124 regs_set_return_value(regs, rc); 125 override_function_with_return(regs); 126 return 0; 127 } 128 129 static const struct bpf_func_proto bpf_override_return_proto = { 130 .func = bpf_override_return, 131 .gpl_only = true, 132 .ret_type = RET_INTEGER, 133 .arg1_type = ARG_PTR_TO_CTX, 134 .arg2_type = ARG_ANYTHING, 135 }; 136 #endif 137 138 BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr) 139 { 140 int ret; 141 142 ret = probe_kernel_read(dst, unsafe_ptr, size); 143 if (unlikely(ret < 0)) 144 memset(dst, 0, size); 145 146 return ret; 147 } 148 149 static const struct bpf_func_proto bpf_probe_read_proto = { 150 .func = bpf_probe_read, 151 .gpl_only = true, 152 .ret_type = RET_INTEGER, 153 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 154 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 155 .arg3_type = ARG_ANYTHING, 156 }; 157 158 BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src, 159 u32, size) 160 { 161 /* 162 * Ensure we're in user context which is safe for the helper to 163 * run. This helper has no business in a kthread. 164 * 165 * access_ok() should prevent writing to non-user memory, but in 166 * some situations (nommu, temporary switch, etc) access_ok() does 167 * not provide enough validation, hence the check on KERNEL_DS. 168 * 169 * nmi_uaccess_okay() ensures the probe is not run in an interim 170 * state, when the task or mm are switched. This is specifically 171 * required to prevent the use of temporary mm. 172 */ 173 174 if (unlikely(in_interrupt() || 175 current->flags & (PF_KTHREAD | PF_EXITING))) 176 return -EPERM; 177 if (unlikely(uaccess_kernel())) 178 return -EPERM; 179 if (unlikely(!nmi_uaccess_okay())) 180 return -EPERM; 181 if (!access_ok(unsafe_ptr, size)) 182 return -EPERM; 183 184 return probe_kernel_write(unsafe_ptr, src, size); 185 } 186 187 static const struct bpf_func_proto bpf_probe_write_user_proto = { 188 .func = bpf_probe_write_user, 189 .gpl_only = true, 190 .ret_type = RET_INTEGER, 191 .arg1_type = ARG_ANYTHING, 192 .arg2_type = ARG_PTR_TO_MEM, 193 .arg3_type = ARG_CONST_SIZE, 194 }; 195 196 static const struct bpf_func_proto *bpf_get_probe_write_proto(void) 197 { 198 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", 199 current->comm, task_pid_nr(current)); 200 201 return &bpf_probe_write_user_proto; 202 } 203 204 /* 205 * Only limited trace_printk() conversion specifiers allowed: 206 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s 207 */ 208 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, 209 u64, arg2, u64, arg3) 210 { 211 bool str_seen = false; 212 int mod[3] = {}; 213 int fmt_cnt = 0; 214 u64 unsafe_addr; 215 char buf[64]; 216 int i; 217 218 /* 219 * bpf_check()->check_func_arg()->check_stack_boundary() 220 * guarantees that fmt points to bpf program stack, 221 * fmt_size bytes of it were initialized and fmt_size > 0 222 */ 223 if (fmt[--fmt_size] != 0) 224 return -EINVAL; 225 226 /* check format string for allowed specifiers */ 227 for (i = 0; i < fmt_size; i++) { 228 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) 229 return -EINVAL; 230 231 if (fmt[i] != '%') 232 continue; 233 234 if (fmt_cnt >= 3) 235 return -EINVAL; 236 237 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ 238 i++; 239 if (fmt[i] == 'l') { 240 mod[fmt_cnt]++; 241 i++; 242 } else if (fmt[i] == 'p' || fmt[i] == 's') { 243 mod[fmt_cnt]++; 244 /* disallow any further format extensions */ 245 if (fmt[i + 1] != 0 && 246 !isspace(fmt[i + 1]) && 247 !ispunct(fmt[i + 1])) 248 return -EINVAL; 249 fmt_cnt++; 250 if (fmt[i] == 's') { 251 if (str_seen) 252 /* allow only one '%s' per fmt string */ 253 return -EINVAL; 254 str_seen = true; 255 256 switch (fmt_cnt) { 257 case 1: 258 unsafe_addr = arg1; 259 arg1 = (long) buf; 260 break; 261 case 2: 262 unsafe_addr = arg2; 263 arg2 = (long) buf; 264 break; 265 case 3: 266 unsafe_addr = arg3; 267 arg3 = (long) buf; 268 break; 269 } 270 buf[0] = 0; 271 strncpy_from_unsafe(buf, 272 (void *) (long) unsafe_addr, 273 sizeof(buf)); 274 } 275 continue; 276 } 277 278 if (fmt[i] == 'l') { 279 mod[fmt_cnt]++; 280 i++; 281 } 282 283 if (fmt[i] != 'i' && fmt[i] != 'd' && 284 fmt[i] != 'u' && fmt[i] != 'x') 285 return -EINVAL; 286 fmt_cnt++; 287 } 288 289 /* Horrid workaround for getting va_list handling working with different 290 * argument type combinations generically for 32 and 64 bit archs. 291 */ 292 #define __BPF_TP_EMIT() __BPF_ARG3_TP() 293 #define __BPF_TP(...) \ 294 __trace_printk(0 /* Fake ip */, \ 295 fmt, ##__VA_ARGS__) 296 297 #define __BPF_ARG1_TP(...) \ 298 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \ 299 ? __BPF_TP(arg1, ##__VA_ARGS__) \ 300 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \ 301 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \ 302 : __BPF_TP((u32)arg1, ##__VA_ARGS__))) 303 304 #define __BPF_ARG2_TP(...) \ 305 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \ 306 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \ 307 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \ 308 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \ 309 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__))) 310 311 #define __BPF_ARG3_TP(...) \ 312 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \ 313 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \ 314 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \ 315 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \ 316 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__))) 317 318 return __BPF_TP_EMIT(); 319 } 320 321 static const struct bpf_func_proto bpf_trace_printk_proto = { 322 .func = bpf_trace_printk, 323 .gpl_only = true, 324 .ret_type = RET_INTEGER, 325 .arg1_type = ARG_PTR_TO_MEM, 326 .arg2_type = ARG_CONST_SIZE, 327 }; 328 329 const struct bpf_func_proto *bpf_get_trace_printk_proto(void) 330 { 331 /* 332 * this program might be calling bpf_trace_printk, 333 * so allocate per-cpu printk buffers 334 */ 335 trace_printk_init_buffers(); 336 337 return &bpf_trace_printk_proto; 338 } 339 340 static __always_inline int 341 get_map_perf_counter(struct bpf_map *map, u64 flags, 342 u64 *value, u64 *enabled, u64 *running) 343 { 344 struct bpf_array *array = container_of(map, struct bpf_array, map); 345 unsigned int cpu = smp_processor_id(); 346 u64 index = flags & BPF_F_INDEX_MASK; 347 struct bpf_event_entry *ee; 348 349 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 350 return -EINVAL; 351 if (index == BPF_F_CURRENT_CPU) 352 index = cpu; 353 if (unlikely(index >= array->map.max_entries)) 354 return -E2BIG; 355 356 ee = READ_ONCE(array->ptrs[index]); 357 if (!ee) 358 return -ENOENT; 359 360 return perf_event_read_local(ee->event, value, enabled, running); 361 } 362 363 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) 364 { 365 u64 value = 0; 366 int err; 367 368 err = get_map_perf_counter(map, flags, &value, NULL, NULL); 369 /* 370 * this api is ugly since we miss [-22..-2] range of valid 371 * counter values, but that's uapi 372 */ 373 if (err) 374 return err; 375 return value; 376 } 377 378 static const struct bpf_func_proto bpf_perf_event_read_proto = { 379 .func = bpf_perf_event_read, 380 .gpl_only = true, 381 .ret_type = RET_INTEGER, 382 .arg1_type = ARG_CONST_MAP_PTR, 383 .arg2_type = ARG_ANYTHING, 384 }; 385 386 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags, 387 struct bpf_perf_event_value *, buf, u32, size) 388 { 389 int err = -EINVAL; 390 391 if (unlikely(size != sizeof(struct bpf_perf_event_value))) 392 goto clear; 393 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled, 394 &buf->running); 395 if (unlikely(err)) 396 goto clear; 397 return 0; 398 clear: 399 memset(buf, 0, size); 400 return err; 401 } 402 403 static const struct bpf_func_proto bpf_perf_event_read_value_proto = { 404 .func = bpf_perf_event_read_value, 405 .gpl_only = true, 406 .ret_type = RET_INTEGER, 407 .arg1_type = ARG_CONST_MAP_PTR, 408 .arg2_type = ARG_ANYTHING, 409 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 410 .arg4_type = ARG_CONST_SIZE, 411 }; 412 413 static DEFINE_PER_CPU(struct perf_sample_data, bpf_trace_sd); 414 415 static __always_inline u64 416 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, 417 u64 flags, struct perf_sample_data *sd) 418 { 419 struct bpf_array *array = container_of(map, struct bpf_array, map); 420 unsigned int cpu = smp_processor_id(); 421 u64 index = flags & BPF_F_INDEX_MASK; 422 struct bpf_event_entry *ee; 423 struct perf_event *event; 424 425 if (index == BPF_F_CURRENT_CPU) 426 index = cpu; 427 if (unlikely(index >= array->map.max_entries)) 428 return -E2BIG; 429 430 ee = READ_ONCE(array->ptrs[index]); 431 if (!ee) 432 return -ENOENT; 433 434 event = ee->event; 435 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || 436 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) 437 return -EINVAL; 438 439 if (unlikely(event->oncpu != cpu)) 440 return -EOPNOTSUPP; 441 442 return perf_event_output(event, sd, regs); 443 } 444 445 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, 446 u64, flags, void *, data, u64, size) 447 { 448 struct perf_sample_data *sd = this_cpu_ptr(&bpf_trace_sd); 449 struct perf_raw_record raw = { 450 .frag = { 451 .size = size, 452 .data = data, 453 }, 454 }; 455 456 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 457 return -EINVAL; 458 459 perf_sample_data_init(sd, 0, 0); 460 sd->raw = &raw; 461 462 return __bpf_perf_event_output(regs, map, flags, sd); 463 } 464 465 static const struct bpf_func_proto bpf_perf_event_output_proto = { 466 .func = bpf_perf_event_output, 467 .gpl_only = true, 468 .ret_type = RET_INTEGER, 469 .arg1_type = ARG_PTR_TO_CTX, 470 .arg2_type = ARG_CONST_MAP_PTR, 471 .arg3_type = ARG_ANYTHING, 472 .arg4_type = ARG_PTR_TO_MEM, 473 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 474 }; 475 476 static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); 477 static DEFINE_PER_CPU(struct perf_sample_data, bpf_misc_sd); 478 479 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 480 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) 481 { 482 struct perf_sample_data *sd = this_cpu_ptr(&bpf_misc_sd); 483 struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); 484 struct perf_raw_frag frag = { 485 .copy = ctx_copy, 486 .size = ctx_size, 487 .data = ctx, 488 }; 489 struct perf_raw_record raw = { 490 .frag = { 491 { 492 .next = ctx_size ? &frag : NULL, 493 }, 494 .size = meta_size, 495 .data = meta, 496 }, 497 }; 498 499 perf_fetch_caller_regs(regs); 500 perf_sample_data_init(sd, 0, 0); 501 sd->raw = &raw; 502 503 return __bpf_perf_event_output(regs, map, flags, sd); 504 } 505 506 BPF_CALL_0(bpf_get_current_task) 507 { 508 return (long) current; 509 } 510 511 static const struct bpf_func_proto bpf_get_current_task_proto = { 512 .func = bpf_get_current_task, 513 .gpl_only = true, 514 .ret_type = RET_INTEGER, 515 }; 516 517 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) 518 { 519 struct bpf_array *array = container_of(map, struct bpf_array, map); 520 struct cgroup *cgrp; 521 522 if (unlikely(idx >= array->map.max_entries)) 523 return -E2BIG; 524 525 cgrp = READ_ONCE(array->ptrs[idx]); 526 if (unlikely(!cgrp)) 527 return -EAGAIN; 528 529 return task_under_cgroup_hierarchy(current, cgrp); 530 } 531 532 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { 533 .func = bpf_current_task_under_cgroup, 534 .gpl_only = false, 535 .ret_type = RET_INTEGER, 536 .arg1_type = ARG_CONST_MAP_PTR, 537 .arg2_type = ARG_ANYTHING, 538 }; 539 540 BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size, 541 const void *, unsafe_ptr) 542 { 543 int ret; 544 545 /* 546 * The strncpy_from_unsafe() call will likely not fill the entire 547 * buffer, but that's okay in this circumstance as we're probing 548 * arbitrary memory anyway similar to bpf_probe_read() and might 549 * as well probe the stack. Thus, memory is explicitly cleared 550 * only in error case, so that improper users ignoring return 551 * code altogether don't copy garbage; otherwise length of string 552 * is returned that can be used for bpf_perf_event_output() et al. 553 */ 554 ret = strncpy_from_unsafe(dst, unsafe_ptr, size); 555 if (unlikely(ret < 0)) 556 memset(dst, 0, size); 557 558 return ret; 559 } 560 561 static const struct bpf_func_proto bpf_probe_read_str_proto = { 562 .func = bpf_probe_read_str, 563 .gpl_only = true, 564 .ret_type = RET_INTEGER, 565 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 566 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 567 .arg3_type = ARG_ANYTHING, 568 }; 569 570 struct send_signal_irq_work { 571 struct irq_work irq_work; 572 struct task_struct *task; 573 u32 sig; 574 }; 575 576 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work); 577 578 static void do_bpf_send_signal(struct irq_work *entry) 579 { 580 struct send_signal_irq_work *work; 581 582 work = container_of(entry, struct send_signal_irq_work, irq_work); 583 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID); 584 } 585 586 BPF_CALL_1(bpf_send_signal, u32, sig) 587 { 588 struct send_signal_irq_work *work = NULL; 589 590 /* Similar to bpf_probe_write_user, task needs to be 591 * in a sound condition and kernel memory access be 592 * permitted in order to send signal to the current 593 * task. 594 */ 595 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING))) 596 return -EPERM; 597 if (unlikely(uaccess_kernel())) 598 return -EPERM; 599 if (unlikely(!nmi_uaccess_okay())) 600 return -EPERM; 601 602 if (in_nmi()) { 603 work = this_cpu_ptr(&send_signal_work); 604 if (work->irq_work.flags & IRQ_WORK_BUSY) 605 return -EBUSY; 606 607 /* Add the current task, which is the target of sending signal, 608 * to the irq_work. The current task may change when queued 609 * irq works get executed. 610 */ 611 work->task = current; 612 work->sig = sig; 613 irq_work_queue(&work->irq_work); 614 return 0; 615 } 616 617 return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID); 618 } 619 620 static const struct bpf_func_proto bpf_send_signal_proto = { 621 .func = bpf_send_signal, 622 .gpl_only = false, 623 .ret_type = RET_INTEGER, 624 .arg1_type = ARG_ANYTHING, 625 }; 626 627 static const struct bpf_func_proto * 628 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 629 { 630 switch (func_id) { 631 case BPF_FUNC_map_lookup_elem: 632 return &bpf_map_lookup_elem_proto; 633 case BPF_FUNC_map_update_elem: 634 return &bpf_map_update_elem_proto; 635 case BPF_FUNC_map_delete_elem: 636 return &bpf_map_delete_elem_proto; 637 case BPF_FUNC_map_push_elem: 638 return &bpf_map_push_elem_proto; 639 case BPF_FUNC_map_pop_elem: 640 return &bpf_map_pop_elem_proto; 641 case BPF_FUNC_map_peek_elem: 642 return &bpf_map_peek_elem_proto; 643 case BPF_FUNC_probe_read: 644 return &bpf_probe_read_proto; 645 case BPF_FUNC_ktime_get_ns: 646 return &bpf_ktime_get_ns_proto; 647 case BPF_FUNC_tail_call: 648 return &bpf_tail_call_proto; 649 case BPF_FUNC_get_current_pid_tgid: 650 return &bpf_get_current_pid_tgid_proto; 651 case BPF_FUNC_get_current_task: 652 return &bpf_get_current_task_proto; 653 case BPF_FUNC_get_current_uid_gid: 654 return &bpf_get_current_uid_gid_proto; 655 case BPF_FUNC_get_current_comm: 656 return &bpf_get_current_comm_proto; 657 case BPF_FUNC_trace_printk: 658 return bpf_get_trace_printk_proto(); 659 case BPF_FUNC_get_smp_processor_id: 660 return &bpf_get_smp_processor_id_proto; 661 case BPF_FUNC_get_numa_node_id: 662 return &bpf_get_numa_node_id_proto; 663 case BPF_FUNC_perf_event_read: 664 return &bpf_perf_event_read_proto; 665 case BPF_FUNC_probe_write_user: 666 return bpf_get_probe_write_proto(); 667 case BPF_FUNC_current_task_under_cgroup: 668 return &bpf_current_task_under_cgroup_proto; 669 case BPF_FUNC_get_prandom_u32: 670 return &bpf_get_prandom_u32_proto; 671 case BPF_FUNC_probe_read_str: 672 return &bpf_probe_read_str_proto; 673 #ifdef CONFIG_CGROUPS 674 case BPF_FUNC_get_current_cgroup_id: 675 return &bpf_get_current_cgroup_id_proto; 676 #endif 677 case BPF_FUNC_send_signal: 678 return &bpf_send_signal_proto; 679 default: 680 return NULL; 681 } 682 } 683 684 static const struct bpf_func_proto * 685 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 686 { 687 switch (func_id) { 688 case BPF_FUNC_perf_event_output: 689 return &bpf_perf_event_output_proto; 690 case BPF_FUNC_get_stackid: 691 return &bpf_get_stackid_proto; 692 case BPF_FUNC_get_stack: 693 return &bpf_get_stack_proto; 694 case BPF_FUNC_perf_event_read_value: 695 return &bpf_perf_event_read_value_proto; 696 #ifdef CONFIG_BPF_KPROBE_OVERRIDE 697 case BPF_FUNC_override_return: 698 return &bpf_override_return_proto; 699 #endif 700 default: 701 return tracing_func_proto(func_id, prog); 702 } 703 } 704 705 /* bpf+kprobe programs can access fields of 'struct pt_regs' */ 706 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 707 const struct bpf_prog *prog, 708 struct bpf_insn_access_aux *info) 709 { 710 if (off < 0 || off >= sizeof(struct pt_regs)) 711 return false; 712 if (type != BPF_READ) 713 return false; 714 if (off % size != 0) 715 return false; 716 /* 717 * Assertion for 32 bit to make sure last 8 byte access 718 * (BPF_DW) to the last 4 byte member is disallowed. 719 */ 720 if (off + size > sizeof(struct pt_regs)) 721 return false; 722 723 return true; 724 } 725 726 const struct bpf_verifier_ops kprobe_verifier_ops = { 727 .get_func_proto = kprobe_prog_func_proto, 728 .is_valid_access = kprobe_prog_is_valid_access, 729 }; 730 731 const struct bpf_prog_ops kprobe_prog_ops = { 732 }; 733 734 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, 735 u64, flags, void *, data, u64, size) 736 { 737 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 738 739 /* 740 * r1 points to perf tracepoint buffer where first 8 bytes are hidden 741 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it 742 * from there and call the same bpf_perf_event_output() helper inline. 743 */ 744 return ____bpf_perf_event_output(regs, map, flags, data, size); 745 } 746 747 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { 748 .func = bpf_perf_event_output_tp, 749 .gpl_only = true, 750 .ret_type = RET_INTEGER, 751 .arg1_type = ARG_PTR_TO_CTX, 752 .arg2_type = ARG_CONST_MAP_PTR, 753 .arg3_type = ARG_ANYTHING, 754 .arg4_type = ARG_PTR_TO_MEM, 755 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 756 }; 757 758 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, 759 u64, flags) 760 { 761 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 762 763 /* 764 * Same comment as in bpf_perf_event_output_tp(), only that this time 765 * the other helper's function body cannot be inlined due to being 766 * external, thus we need to call raw helper function. 767 */ 768 return bpf_get_stackid((unsigned long) regs, (unsigned long) map, 769 flags, 0, 0); 770 } 771 772 static const struct bpf_func_proto bpf_get_stackid_proto_tp = { 773 .func = bpf_get_stackid_tp, 774 .gpl_only = true, 775 .ret_type = RET_INTEGER, 776 .arg1_type = ARG_PTR_TO_CTX, 777 .arg2_type = ARG_CONST_MAP_PTR, 778 .arg3_type = ARG_ANYTHING, 779 }; 780 781 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size, 782 u64, flags) 783 { 784 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 785 786 return bpf_get_stack((unsigned long) regs, (unsigned long) buf, 787 (unsigned long) size, flags, 0); 788 } 789 790 static const struct bpf_func_proto bpf_get_stack_proto_tp = { 791 .func = bpf_get_stack_tp, 792 .gpl_only = true, 793 .ret_type = RET_INTEGER, 794 .arg1_type = ARG_PTR_TO_CTX, 795 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 796 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 797 .arg4_type = ARG_ANYTHING, 798 }; 799 800 static const struct bpf_func_proto * 801 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 802 { 803 switch (func_id) { 804 case BPF_FUNC_perf_event_output: 805 return &bpf_perf_event_output_proto_tp; 806 case BPF_FUNC_get_stackid: 807 return &bpf_get_stackid_proto_tp; 808 case BPF_FUNC_get_stack: 809 return &bpf_get_stack_proto_tp; 810 default: 811 return tracing_func_proto(func_id, prog); 812 } 813 } 814 815 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, 816 const struct bpf_prog *prog, 817 struct bpf_insn_access_aux *info) 818 { 819 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) 820 return false; 821 if (type != BPF_READ) 822 return false; 823 if (off % size != 0) 824 return false; 825 826 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); 827 return true; 828 } 829 830 const struct bpf_verifier_ops tracepoint_verifier_ops = { 831 .get_func_proto = tp_prog_func_proto, 832 .is_valid_access = tp_prog_is_valid_access, 833 }; 834 835 const struct bpf_prog_ops tracepoint_prog_ops = { 836 }; 837 838 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx, 839 struct bpf_perf_event_value *, buf, u32, size) 840 { 841 int err = -EINVAL; 842 843 if (unlikely(size != sizeof(struct bpf_perf_event_value))) 844 goto clear; 845 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled, 846 &buf->running); 847 if (unlikely(err)) 848 goto clear; 849 return 0; 850 clear: 851 memset(buf, 0, size); 852 return err; 853 } 854 855 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = { 856 .func = bpf_perf_prog_read_value, 857 .gpl_only = true, 858 .ret_type = RET_INTEGER, 859 .arg1_type = ARG_PTR_TO_CTX, 860 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 861 .arg3_type = ARG_CONST_SIZE, 862 }; 863 864 static const struct bpf_func_proto * 865 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 866 { 867 switch (func_id) { 868 case BPF_FUNC_perf_event_output: 869 return &bpf_perf_event_output_proto_tp; 870 case BPF_FUNC_get_stackid: 871 return &bpf_get_stackid_proto_tp; 872 case BPF_FUNC_get_stack: 873 return &bpf_get_stack_proto_tp; 874 case BPF_FUNC_perf_prog_read_value: 875 return &bpf_perf_prog_read_value_proto; 876 default: 877 return tracing_func_proto(func_id, prog); 878 } 879 } 880 881 /* 882 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp 883 * to avoid potential recursive reuse issue when/if tracepoints are added 884 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack 885 */ 886 static DEFINE_PER_CPU(struct pt_regs, bpf_raw_tp_regs); 887 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args, 888 struct bpf_map *, map, u64, flags, void *, data, u64, size) 889 { 890 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs); 891 892 perf_fetch_caller_regs(regs); 893 return ____bpf_perf_event_output(regs, map, flags, data, size); 894 } 895 896 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = { 897 .func = bpf_perf_event_output_raw_tp, 898 .gpl_only = true, 899 .ret_type = RET_INTEGER, 900 .arg1_type = ARG_PTR_TO_CTX, 901 .arg2_type = ARG_CONST_MAP_PTR, 902 .arg3_type = ARG_ANYTHING, 903 .arg4_type = ARG_PTR_TO_MEM, 904 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 905 }; 906 907 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args, 908 struct bpf_map *, map, u64, flags) 909 { 910 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs); 911 912 perf_fetch_caller_regs(regs); 913 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */ 914 return bpf_get_stackid((unsigned long) regs, (unsigned long) map, 915 flags, 0, 0); 916 } 917 918 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = { 919 .func = bpf_get_stackid_raw_tp, 920 .gpl_only = true, 921 .ret_type = RET_INTEGER, 922 .arg1_type = ARG_PTR_TO_CTX, 923 .arg2_type = ARG_CONST_MAP_PTR, 924 .arg3_type = ARG_ANYTHING, 925 }; 926 927 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args, 928 void *, buf, u32, size, u64, flags) 929 { 930 struct pt_regs *regs = this_cpu_ptr(&bpf_raw_tp_regs); 931 932 perf_fetch_caller_regs(regs); 933 return bpf_get_stack((unsigned long) regs, (unsigned long) buf, 934 (unsigned long) size, flags, 0); 935 } 936 937 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = { 938 .func = bpf_get_stack_raw_tp, 939 .gpl_only = true, 940 .ret_type = RET_INTEGER, 941 .arg1_type = ARG_PTR_TO_CTX, 942 .arg2_type = ARG_PTR_TO_MEM, 943 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 944 .arg4_type = ARG_ANYTHING, 945 }; 946 947 static const struct bpf_func_proto * 948 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 949 { 950 switch (func_id) { 951 case BPF_FUNC_perf_event_output: 952 return &bpf_perf_event_output_proto_raw_tp; 953 case BPF_FUNC_get_stackid: 954 return &bpf_get_stackid_proto_raw_tp; 955 case BPF_FUNC_get_stack: 956 return &bpf_get_stack_proto_raw_tp; 957 default: 958 return tracing_func_proto(func_id, prog); 959 } 960 } 961 962 static bool raw_tp_prog_is_valid_access(int off, int size, 963 enum bpf_access_type type, 964 const struct bpf_prog *prog, 965 struct bpf_insn_access_aux *info) 966 { 967 /* largest tracepoint in the kernel has 12 args */ 968 if (off < 0 || off >= sizeof(__u64) * 12) 969 return false; 970 if (type != BPF_READ) 971 return false; 972 if (off % size != 0) 973 return false; 974 return true; 975 } 976 977 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = { 978 .get_func_proto = raw_tp_prog_func_proto, 979 .is_valid_access = raw_tp_prog_is_valid_access, 980 }; 981 982 const struct bpf_prog_ops raw_tracepoint_prog_ops = { 983 }; 984 985 static bool raw_tp_writable_prog_is_valid_access(int off, int size, 986 enum bpf_access_type type, 987 const struct bpf_prog *prog, 988 struct bpf_insn_access_aux *info) 989 { 990 if (off == 0) { 991 if (size != sizeof(u64) || type != BPF_READ) 992 return false; 993 info->reg_type = PTR_TO_TP_BUFFER; 994 } 995 return raw_tp_prog_is_valid_access(off, size, type, prog, info); 996 } 997 998 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = { 999 .get_func_proto = raw_tp_prog_func_proto, 1000 .is_valid_access = raw_tp_writable_prog_is_valid_access, 1001 }; 1002 1003 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = { 1004 }; 1005 1006 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 1007 const struct bpf_prog *prog, 1008 struct bpf_insn_access_aux *info) 1009 { 1010 const int size_u64 = sizeof(u64); 1011 1012 if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) 1013 return false; 1014 if (type != BPF_READ) 1015 return false; 1016 if (off % size != 0) { 1017 if (sizeof(unsigned long) != 4) 1018 return false; 1019 if (size != 8) 1020 return false; 1021 if (off % size != 4) 1022 return false; 1023 } 1024 1025 switch (off) { 1026 case bpf_ctx_range(struct bpf_perf_event_data, sample_period): 1027 bpf_ctx_record_field_size(info, size_u64); 1028 if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) 1029 return false; 1030 break; 1031 case bpf_ctx_range(struct bpf_perf_event_data, addr): 1032 bpf_ctx_record_field_size(info, size_u64); 1033 if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) 1034 return false; 1035 break; 1036 default: 1037 if (size != sizeof(long)) 1038 return false; 1039 } 1040 1041 return true; 1042 } 1043 1044 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, 1045 const struct bpf_insn *si, 1046 struct bpf_insn *insn_buf, 1047 struct bpf_prog *prog, u32 *target_size) 1048 { 1049 struct bpf_insn *insn = insn_buf; 1050 1051 switch (si->off) { 1052 case offsetof(struct bpf_perf_event_data, sample_period): 1053 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 1054 data), si->dst_reg, si->src_reg, 1055 offsetof(struct bpf_perf_event_data_kern, data)); 1056 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 1057 bpf_target_off(struct perf_sample_data, period, 8, 1058 target_size)); 1059 break; 1060 case offsetof(struct bpf_perf_event_data, addr): 1061 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 1062 data), si->dst_reg, si->src_reg, 1063 offsetof(struct bpf_perf_event_data_kern, data)); 1064 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 1065 bpf_target_off(struct perf_sample_data, addr, 8, 1066 target_size)); 1067 break; 1068 default: 1069 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 1070 regs), si->dst_reg, si->src_reg, 1071 offsetof(struct bpf_perf_event_data_kern, regs)); 1072 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, 1073 si->off); 1074 break; 1075 } 1076 1077 return insn - insn_buf; 1078 } 1079 1080 const struct bpf_verifier_ops perf_event_verifier_ops = { 1081 .get_func_proto = pe_prog_func_proto, 1082 .is_valid_access = pe_prog_is_valid_access, 1083 .convert_ctx_access = pe_prog_convert_ctx_access, 1084 }; 1085 1086 const struct bpf_prog_ops perf_event_prog_ops = { 1087 }; 1088 1089 static DEFINE_MUTEX(bpf_event_mutex); 1090 1091 #define BPF_TRACE_MAX_PROGS 64 1092 1093 int perf_event_attach_bpf_prog(struct perf_event *event, 1094 struct bpf_prog *prog) 1095 { 1096 struct bpf_prog_array __rcu *old_array; 1097 struct bpf_prog_array *new_array; 1098 int ret = -EEXIST; 1099 1100 /* 1101 * Kprobe override only works if they are on the function entry, 1102 * and only if they are on the opt-in list. 1103 */ 1104 if (prog->kprobe_override && 1105 (!trace_kprobe_on_func_entry(event->tp_event) || 1106 !trace_kprobe_error_injectable(event->tp_event))) 1107 return -EINVAL; 1108 1109 mutex_lock(&bpf_event_mutex); 1110 1111 if (event->prog) 1112 goto unlock; 1113 1114 old_array = event->tp_event->prog_array; 1115 if (old_array && 1116 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) { 1117 ret = -E2BIG; 1118 goto unlock; 1119 } 1120 1121 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array); 1122 if (ret < 0) 1123 goto unlock; 1124 1125 /* set the new array to event->tp_event and set event->prog */ 1126 event->prog = prog; 1127 rcu_assign_pointer(event->tp_event->prog_array, new_array); 1128 bpf_prog_array_free(old_array); 1129 1130 unlock: 1131 mutex_unlock(&bpf_event_mutex); 1132 return ret; 1133 } 1134 1135 void perf_event_detach_bpf_prog(struct perf_event *event) 1136 { 1137 struct bpf_prog_array __rcu *old_array; 1138 struct bpf_prog_array *new_array; 1139 int ret; 1140 1141 mutex_lock(&bpf_event_mutex); 1142 1143 if (!event->prog) 1144 goto unlock; 1145 1146 old_array = event->tp_event->prog_array; 1147 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array); 1148 if (ret == -ENOENT) 1149 goto unlock; 1150 if (ret < 0) { 1151 bpf_prog_array_delete_safe(old_array, event->prog); 1152 } else { 1153 rcu_assign_pointer(event->tp_event->prog_array, new_array); 1154 bpf_prog_array_free(old_array); 1155 } 1156 1157 bpf_prog_put(event->prog); 1158 event->prog = NULL; 1159 1160 unlock: 1161 mutex_unlock(&bpf_event_mutex); 1162 } 1163 1164 int perf_event_query_prog_array(struct perf_event *event, void __user *info) 1165 { 1166 struct perf_event_query_bpf __user *uquery = info; 1167 struct perf_event_query_bpf query = {}; 1168 u32 *ids, prog_cnt, ids_len; 1169 int ret; 1170 1171 if (!capable(CAP_SYS_ADMIN)) 1172 return -EPERM; 1173 if (event->attr.type != PERF_TYPE_TRACEPOINT) 1174 return -EINVAL; 1175 if (copy_from_user(&query, uquery, sizeof(query))) 1176 return -EFAULT; 1177 1178 ids_len = query.ids_len; 1179 if (ids_len > BPF_TRACE_MAX_PROGS) 1180 return -E2BIG; 1181 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN); 1182 if (!ids) 1183 return -ENOMEM; 1184 /* 1185 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which 1186 * is required when user only wants to check for uquery->prog_cnt. 1187 * There is no need to check for it since the case is handled 1188 * gracefully in bpf_prog_array_copy_info. 1189 */ 1190 1191 mutex_lock(&bpf_event_mutex); 1192 ret = bpf_prog_array_copy_info(event->tp_event->prog_array, 1193 ids, 1194 ids_len, 1195 &prog_cnt); 1196 mutex_unlock(&bpf_event_mutex); 1197 1198 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) || 1199 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32))) 1200 ret = -EFAULT; 1201 1202 kfree(ids); 1203 return ret; 1204 } 1205 1206 extern struct bpf_raw_event_map __start__bpf_raw_tp[]; 1207 extern struct bpf_raw_event_map __stop__bpf_raw_tp[]; 1208 1209 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name) 1210 { 1211 struct bpf_raw_event_map *btp = __start__bpf_raw_tp; 1212 1213 for (; btp < __stop__bpf_raw_tp; btp++) { 1214 if (!strcmp(btp->tp->name, name)) 1215 return btp; 1216 } 1217 1218 return bpf_get_raw_tracepoint_module(name); 1219 } 1220 1221 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp) 1222 { 1223 struct module *mod = __module_address((unsigned long)btp); 1224 1225 if (mod) 1226 module_put(mod); 1227 } 1228 1229 static __always_inline 1230 void __bpf_trace_run(struct bpf_prog *prog, u64 *args) 1231 { 1232 rcu_read_lock(); 1233 preempt_disable(); 1234 (void) BPF_PROG_RUN(prog, args); 1235 preempt_enable(); 1236 rcu_read_unlock(); 1237 } 1238 1239 #define UNPACK(...) __VA_ARGS__ 1240 #define REPEAT_1(FN, DL, X, ...) FN(X) 1241 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__) 1242 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__) 1243 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__) 1244 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__) 1245 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__) 1246 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__) 1247 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__) 1248 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__) 1249 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__) 1250 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__) 1251 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__) 1252 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__) 1253 1254 #define SARG(X) u64 arg##X 1255 #define COPY(X) args[X] = arg##X 1256 1257 #define __DL_COM (,) 1258 #define __DL_SEM (;) 1259 1260 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 1261 1262 #define BPF_TRACE_DEFN_x(x) \ 1263 void bpf_trace_run##x(struct bpf_prog *prog, \ 1264 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \ 1265 { \ 1266 u64 args[x]; \ 1267 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \ 1268 __bpf_trace_run(prog, args); \ 1269 } \ 1270 EXPORT_SYMBOL_GPL(bpf_trace_run##x) 1271 BPF_TRACE_DEFN_x(1); 1272 BPF_TRACE_DEFN_x(2); 1273 BPF_TRACE_DEFN_x(3); 1274 BPF_TRACE_DEFN_x(4); 1275 BPF_TRACE_DEFN_x(5); 1276 BPF_TRACE_DEFN_x(6); 1277 BPF_TRACE_DEFN_x(7); 1278 BPF_TRACE_DEFN_x(8); 1279 BPF_TRACE_DEFN_x(9); 1280 BPF_TRACE_DEFN_x(10); 1281 BPF_TRACE_DEFN_x(11); 1282 BPF_TRACE_DEFN_x(12); 1283 1284 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) 1285 { 1286 struct tracepoint *tp = btp->tp; 1287 1288 /* 1289 * check that program doesn't access arguments beyond what's 1290 * available in this tracepoint 1291 */ 1292 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64)) 1293 return -EINVAL; 1294 1295 if (prog->aux->max_tp_access > btp->writable_size) 1296 return -EINVAL; 1297 1298 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog); 1299 } 1300 1301 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) 1302 { 1303 return __bpf_probe_register(btp, prog); 1304 } 1305 1306 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog) 1307 { 1308 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog); 1309 } 1310 1311 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id, 1312 u32 *fd_type, const char **buf, 1313 u64 *probe_offset, u64 *probe_addr) 1314 { 1315 bool is_tracepoint, is_syscall_tp; 1316 struct bpf_prog *prog; 1317 int flags, err = 0; 1318 1319 prog = event->prog; 1320 if (!prog) 1321 return -ENOENT; 1322 1323 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */ 1324 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) 1325 return -EOPNOTSUPP; 1326 1327 *prog_id = prog->aux->id; 1328 flags = event->tp_event->flags; 1329 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT; 1330 is_syscall_tp = is_syscall_trace_event(event->tp_event); 1331 1332 if (is_tracepoint || is_syscall_tp) { 1333 *buf = is_tracepoint ? event->tp_event->tp->name 1334 : event->tp_event->name; 1335 *fd_type = BPF_FD_TYPE_TRACEPOINT; 1336 *probe_offset = 0x0; 1337 *probe_addr = 0x0; 1338 } else { 1339 /* kprobe/uprobe */ 1340 err = -EOPNOTSUPP; 1341 #ifdef CONFIG_KPROBE_EVENTS 1342 if (flags & TRACE_EVENT_FL_KPROBE) 1343 err = bpf_get_kprobe_info(event, fd_type, buf, 1344 probe_offset, probe_addr, 1345 event->attr.type == PERF_TYPE_TRACEPOINT); 1346 #endif 1347 #ifdef CONFIG_UPROBE_EVENTS 1348 if (flags & TRACE_EVENT_FL_UPROBE) 1349 err = bpf_get_uprobe_info(event, fd_type, buf, 1350 probe_offset, 1351 event->attr.type == PERF_TYPE_TRACEPOINT); 1352 #endif 1353 } 1354 1355 return err; 1356 } 1357 1358 #ifdef CONFIG_MODULES 1359 static int bpf_event_notify(struct notifier_block *nb, unsigned long op, 1360 void *module) 1361 { 1362 struct bpf_trace_module *btm, *tmp; 1363 struct module *mod = module; 1364 1365 if (mod->num_bpf_raw_events == 0 || 1366 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING)) 1367 return 0; 1368 1369 mutex_lock(&bpf_module_mutex); 1370 1371 switch (op) { 1372 case MODULE_STATE_COMING: 1373 btm = kzalloc(sizeof(*btm), GFP_KERNEL); 1374 if (btm) { 1375 btm->module = module; 1376 list_add(&btm->list, &bpf_trace_modules); 1377 } 1378 break; 1379 case MODULE_STATE_GOING: 1380 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) { 1381 if (btm->module == module) { 1382 list_del(&btm->list); 1383 kfree(btm); 1384 break; 1385 } 1386 } 1387 break; 1388 } 1389 1390 mutex_unlock(&bpf_module_mutex); 1391 1392 return 0; 1393 } 1394 1395 static struct notifier_block bpf_module_nb = { 1396 .notifier_call = bpf_event_notify, 1397 }; 1398 1399 static int __init bpf_event_init(void) 1400 { 1401 register_module_notifier(&bpf_module_nb); 1402 return 0; 1403 } 1404 1405 static int __init send_signal_irq_work_init(void) 1406 { 1407 int cpu; 1408 struct send_signal_irq_work *work; 1409 1410 for_each_possible_cpu(cpu) { 1411 work = per_cpu_ptr(&send_signal_work, cpu); 1412 init_irq_work(&work->irq_work, do_bpf_send_signal); 1413 } 1414 return 0; 1415 } 1416 1417 fs_initcall(bpf_event_init); 1418 subsys_initcall(send_signal_irq_work_init); 1419 #endif /* CONFIG_MODULES */ 1420