1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2021, Microsoft Corporation. 4 * 5 * Authors: 6 * Beau Belgrave <beaub@linux.microsoft.com> 7 */ 8 9 #include <linux/bitmap.h> 10 #include <linux/cdev.h> 11 #include <linux/hashtable.h> 12 #include <linux/list.h> 13 #include <linux/io.h> 14 #include <linux/uio.h> 15 #include <linux/ioctl.h> 16 #include <linux/jhash.h> 17 #include <linux/trace_events.h> 18 #include <linux/tracefs.h> 19 #include <linux/types.h> 20 #include <linux/uaccess.h> 21 #include <uapi/linux/user_events.h> 22 #include "trace.h" 23 #include "trace_dynevent.h" 24 25 #define USER_EVENTS_PREFIX_LEN (sizeof(USER_EVENTS_PREFIX)-1) 26 27 #define FIELD_DEPTH_TYPE 0 28 #define FIELD_DEPTH_NAME 1 29 #define FIELD_DEPTH_SIZE 2 30 31 /* 32 * Limits how many trace_event calls user processes can create: 33 * Must be multiple of PAGE_SIZE. 34 */ 35 #define MAX_PAGES 1 36 #define MAX_EVENTS (MAX_PAGES * PAGE_SIZE) 37 38 /* Limit how long of an event name plus args within the subsystem. */ 39 #define MAX_EVENT_DESC 512 40 #define EVENT_NAME(user_event) ((user_event)->tracepoint.name) 41 #define MAX_FIELD_ARRAY_SIZE 1024 42 #define MAX_FIELD_ARG_NAME 256 43 44 #define MAX_BPF_COPY_SIZE PAGE_SIZE 45 #define MAX_STACK_BPF_DATA 512 46 47 static char *register_page_data; 48 49 static DEFINE_MUTEX(reg_mutex); 50 static DEFINE_HASHTABLE(register_table, 4); 51 static DECLARE_BITMAP(page_bitmap, MAX_EVENTS); 52 53 /* 54 * Stores per-event properties, as users register events 55 * within a file a user_event might be created if it does not 56 * already exist. These are globally used and their lifetime 57 * is tied to the refcnt member. These cannot go away until the 58 * refcnt reaches zero. 59 */ 60 struct user_event { 61 struct tracepoint tracepoint; 62 struct trace_event_call call; 63 struct trace_event_class class; 64 struct dyn_event devent; 65 struct hlist_node node; 66 struct list_head fields; 67 struct list_head validators; 68 atomic_t refcnt; 69 int index; 70 int flags; 71 int min_size; 72 }; 73 74 /* 75 * Stores per-file events references, as users register events 76 * within a file this structure is modified and freed via RCU. 77 * The lifetime of this struct is tied to the lifetime of the file. 78 * These are not shared and only accessible by the file that created it. 79 */ 80 struct user_event_refs { 81 struct rcu_head rcu; 82 int count; 83 struct user_event *events[]; 84 }; 85 86 #define VALIDATOR_ENSURE_NULL (1 << 0) 87 #define VALIDATOR_REL (1 << 1) 88 89 struct user_event_validator { 90 struct list_head link; 91 int offset; 92 int flags; 93 }; 94 95 typedef void (*user_event_func_t) (struct user_event *user, struct iov_iter *i, 96 void *tpdata, bool *faulted); 97 98 static int user_event_parse(char *name, char *args, char *flags, 99 struct user_event **newuser); 100 101 static u32 user_event_key(char *name) 102 { 103 return jhash(name, strlen(name), 0); 104 } 105 106 static __always_inline __must_check 107 size_t copy_nofault(void *addr, size_t bytes, struct iov_iter *i) 108 { 109 size_t ret; 110 111 pagefault_disable(); 112 113 ret = copy_from_iter_nocache(addr, bytes, i); 114 115 pagefault_enable(); 116 117 return ret; 118 } 119 120 static struct list_head *user_event_get_fields(struct trace_event_call *call) 121 { 122 struct user_event *user = (struct user_event *)call->data; 123 124 return &user->fields; 125 } 126 127 /* 128 * Parses a register command for user_events 129 * Format: event_name[:FLAG1[,FLAG2...]] [field1[;field2...]] 130 * 131 * Example event named 'test' with a 20 char 'msg' field with an unsigned int 132 * 'id' field after: 133 * test char[20] msg;unsigned int id 134 * 135 * NOTE: Offsets are from the user data perspective, they are not from the 136 * trace_entry/buffer perspective. We automatically add the common properties 137 * sizes to the offset for the user. 138 */ 139 static int user_event_parse_cmd(char *raw_command, struct user_event **newuser) 140 { 141 char *name = raw_command; 142 char *args = strpbrk(name, " "); 143 char *flags; 144 145 if (args) 146 *args++ = '\0'; 147 148 flags = strpbrk(name, ":"); 149 150 if (flags) 151 *flags++ = '\0'; 152 153 return user_event_parse(name, args, flags, newuser); 154 } 155 156 static int user_field_array_size(const char *type) 157 { 158 const char *start = strchr(type, '['); 159 char val[8]; 160 char *bracket; 161 int size = 0; 162 163 if (start == NULL) 164 return -EINVAL; 165 166 if (strscpy(val, start + 1, sizeof(val)) <= 0) 167 return -EINVAL; 168 169 bracket = strchr(val, ']'); 170 171 if (!bracket) 172 return -EINVAL; 173 174 *bracket = '\0'; 175 176 if (kstrtouint(val, 0, &size)) 177 return -EINVAL; 178 179 if (size > MAX_FIELD_ARRAY_SIZE) 180 return -EINVAL; 181 182 return size; 183 } 184 185 static int user_field_size(const char *type) 186 { 187 /* long is not allowed from a user, since it's ambigious in size */ 188 if (strcmp(type, "s64") == 0) 189 return sizeof(s64); 190 if (strcmp(type, "u64") == 0) 191 return sizeof(u64); 192 if (strcmp(type, "s32") == 0) 193 return sizeof(s32); 194 if (strcmp(type, "u32") == 0) 195 return sizeof(u32); 196 if (strcmp(type, "int") == 0) 197 return sizeof(int); 198 if (strcmp(type, "unsigned int") == 0) 199 return sizeof(unsigned int); 200 if (strcmp(type, "s16") == 0) 201 return sizeof(s16); 202 if (strcmp(type, "u16") == 0) 203 return sizeof(u16); 204 if (strcmp(type, "short") == 0) 205 return sizeof(short); 206 if (strcmp(type, "unsigned short") == 0) 207 return sizeof(unsigned short); 208 if (strcmp(type, "s8") == 0) 209 return sizeof(s8); 210 if (strcmp(type, "u8") == 0) 211 return sizeof(u8); 212 if (strcmp(type, "char") == 0) 213 return sizeof(char); 214 if (strcmp(type, "unsigned char") == 0) 215 return sizeof(unsigned char); 216 if (str_has_prefix(type, "char[")) 217 return user_field_array_size(type); 218 if (str_has_prefix(type, "unsigned char[")) 219 return user_field_array_size(type); 220 if (str_has_prefix(type, "__data_loc ")) 221 return sizeof(u32); 222 if (str_has_prefix(type, "__rel_loc ")) 223 return sizeof(u32); 224 225 /* Uknown basic type, error */ 226 return -EINVAL; 227 } 228 229 static void user_event_destroy_validators(struct user_event *user) 230 { 231 struct user_event_validator *validator, *next; 232 struct list_head *head = &user->validators; 233 234 list_for_each_entry_safe(validator, next, head, link) { 235 list_del(&validator->link); 236 kfree(validator); 237 } 238 } 239 240 static void user_event_destroy_fields(struct user_event *user) 241 { 242 struct ftrace_event_field *field, *next; 243 struct list_head *head = &user->fields; 244 245 list_for_each_entry_safe(field, next, head, link) { 246 list_del(&field->link); 247 kfree(field); 248 } 249 } 250 251 static int user_event_add_field(struct user_event *user, const char *type, 252 const char *name, int offset, int size, 253 int is_signed, int filter_type) 254 { 255 struct user_event_validator *validator; 256 struct ftrace_event_field *field; 257 int validator_flags = 0; 258 259 field = kmalloc(sizeof(*field), GFP_KERNEL); 260 261 if (!field) 262 return -ENOMEM; 263 264 if (str_has_prefix(type, "__data_loc ")) 265 goto add_validator; 266 267 if (str_has_prefix(type, "__rel_loc ")) { 268 validator_flags |= VALIDATOR_REL; 269 goto add_validator; 270 } 271 272 goto add_field; 273 274 add_validator: 275 if (strstr(type, "char") != 0) 276 validator_flags |= VALIDATOR_ENSURE_NULL; 277 278 validator = kmalloc(sizeof(*validator), GFP_KERNEL); 279 280 if (!validator) { 281 kfree(field); 282 return -ENOMEM; 283 } 284 285 validator->flags = validator_flags; 286 validator->offset = offset; 287 288 /* Want sequential access when validating */ 289 list_add_tail(&validator->link, &user->validators); 290 291 add_field: 292 field->type = type; 293 field->name = name; 294 field->offset = offset; 295 field->size = size; 296 field->is_signed = is_signed; 297 field->filter_type = filter_type; 298 299 list_add(&field->link, &user->fields); 300 301 /* 302 * Min size from user writes that are required, this does not include 303 * the size of trace_entry (common fields). 304 */ 305 user->min_size = (offset + size) - sizeof(struct trace_entry); 306 307 return 0; 308 } 309 310 /* 311 * Parses the values of a field within the description 312 * Format: type name [size] 313 */ 314 static int user_event_parse_field(char *field, struct user_event *user, 315 u32 *offset) 316 { 317 char *part, *type, *name; 318 u32 depth = 0, saved_offset = *offset; 319 int len, size = -EINVAL; 320 bool is_struct = false; 321 322 field = skip_spaces(field); 323 324 if (*field == '\0') 325 return 0; 326 327 /* Handle types that have a space within */ 328 len = str_has_prefix(field, "unsigned "); 329 if (len) 330 goto skip_next; 331 332 len = str_has_prefix(field, "struct "); 333 if (len) { 334 is_struct = true; 335 goto skip_next; 336 } 337 338 len = str_has_prefix(field, "__data_loc unsigned "); 339 if (len) 340 goto skip_next; 341 342 len = str_has_prefix(field, "__data_loc "); 343 if (len) 344 goto skip_next; 345 346 len = str_has_prefix(field, "__rel_loc unsigned "); 347 if (len) 348 goto skip_next; 349 350 len = str_has_prefix(field, "__rel_loc "); 351 if (len) 352 goto skip_next; 353 354 goto parse; 355 skip_next: 356 type = field; 357 field = strpbrk(field + len, " "); 358 359 if (field == NULL) 360 return -EINVAL; 361 362 *field++ = '\0'; 363 depth++; 364 parse: 365 while ((part = strsep(&field, " ")) != NULL) { 366 switch (depth++) { 367 case FIELD_DEPTH_TYPE: 368 type = part; 369 break; 370 case FIELD_DEPTH_NAME: 371 name = part; 372 break; 373 case FIELD_DEPTH_SIZE: 374 if (!is_struct) 375 return -EINVAL; 376 377 if (kstrtou32(part, 10, &size)) 378 return -EINVAL; 379 break; 380 default: 381 return -EINVAL; 382 } 383 } 384 385 if (depth < FIELD_DEPTH_SIZE) 386 return -EINVAL; 387 388 if (depth == FIELD_DEPTH_SIZE) 389 size = user_field_size(type); 390 391 if (size == 0) 392 return -EINVAL; 393 394 if (size < 0) 395 return size; 396 397 *offset = saved_offset + size; 398 399 return user_event_add_field(user, type, name, saved_offset, size, 400 type[0] != 'u', FILTER_OTHER); 401 } 402 403 static void user_event_parse_flags(struct user_event *user, char *flags) 404 { 405 char *flag; 406 407 if (flags == NULL) 408 return; 409 410 while ((flag = strsep(&flags, ",")) != NULL) { 411 if (strcmp(flag, "BPF_ITER") == 0) 412 user->flags |= FLAG_BPF_ITER; 413 } 414 } 415 416 static int user_event_parse_fields(struct user_event *user, char *args) 417 { 418 char *field; 419 u32 offset = sizeof(struct trace_entry); 420 int ret = -EINVAL; 421 422 if (args == NULL) 423 return 0; 424 425 while ((field = strsep(&args, ";")) != NULL) { 426 ret = user_event_parse_field(field, user, &offset); 427 428 if (ret) 429 break; 430 } 431 432 return ret; 433 } 434 435 static struct trace_event_fields user_event_fields_array[1]; 436 437 static const char *user_field_format(const char *type) 438 { 439 if (strcmp(type, "s64") == 0) 440 return "%lld"; 441 if (strcmp(type, "u64") == 0) 442 return "%llu"; 443 if (strcmp(type, "s32") == 0) 444 return "%d"; 445 if (strcmp(type, "u32") == 0) 446 return "%u"; 447 if (strcmp(type, "int") == 0) 448 return "%d"; 449 if (strcmp(type, "unsigned int") == 0) 450 return "%u"; 451 if (strcmp(type, "s16") == 0) 452 return "%d"; 453 if (strcmp(type, "u16") == 0) 454 return "%u"; 455 if (strcmp(type, "short") == 0) 456 return "%d"; 457 if (strcmp(type, "unsigned short") == 0) 458 return "%u"; 459 if (strcmp(type, "s8") == 0) 460 return "%d"; 461 if (strcmp(type, "u8") == 0) 462 return "%u"; 463 if (strcmp(type, "char") == 0) 464 return "%d"; 465 if (strcmp(type, "unsigned char") == 0) 466 return "%u"; 467 if (strstr(type, "char[") != 0) 468 return "%s"; 469 470 /* Unknown, likely struct, allowed treat as 64-bit */ 471 return "%llu"; 472 } 473 474 static bool user_field_is_dyn_string(const char *type, const char **str_func) 475 { 476 if (str_has_prefix(type, "__data_loc ")) { 477 *str_func = "__get_str"; 478 goto check; 479 } 480 481 if (str_has_prefix(type, "__rel_loc ")) { 482 *str_func = "__get_rel_str"; 483 goto check; 484 } 485 486 return false; 487 check: 488 return strstr(type, "char") != 0; 489 } 490 491 #define LEN_OR_ZERO (len ? len - pos : 0) 492 static int user_event_set_print_fmt(struct user_event *user, char *buf, int len) 493 { 494 struct ftrace_event_field *field, *next; 495 struct list_head *head = &user->fields; 496 int pos = 0, depth = 0; 497 const char *str_func; 498 499 pos += snprintf(buf + pos, LEN_OR_ZERO, "\""); 500 501 list_for_each_entry_safe_reverse(field, next, head, link) { 502 if (depth != 0) 503 pos += snprintf(buf + pos, LEN_OR_ZERO, " "); 504 505 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s=%s", 506 field->name, user_field_format(field->type)); 507 508 depth++; 509 } 510 511 pos += snprintf(buf + pos, LEN_OR_ZERO, "\""); 512 513 list_for_each_entry_safe_reverse(field, next, head, link) { 514 if (user_field_is_dyn_string(field->type, &str_func)) 515 pos += snprintf(buf + pos, LEN_OR_ZERO, 516 ", %s(%s)", str_func, field->name); 517 else 518 pos += snprintf(buf + pos, LEN_OR_ZERO, 519 ", REC->%s", field->name); 520 } 521 522 return pos + 1; 523 } 524 #undef LEN_OR_ZERO 525 526 static int user_event_create_print_fmt(struct user_event *user) 527 { 528 char *print_fmt; 529 int len; 530 531 len = user_event_set_print_fmt(user, NULL, 0); 532 533 print_fmt = kmalloc(len, GFP_KERNEL); 534 535 if (!print_fmt) 536 return -ENOMEM; 537 538 user_event_set_print_fmt(user, print_fmt, len); 539 540 user->call.print_fmt = print_fmt; 541 542 return 0; 543 } 544 545 static enum print_line_t user_event_print_trace(struct trace_iterator *iter, 546 int flags, 547 struct trace_event *event) 548 { 549 /* Unsafe to try to decode user provided print_fmt, use hex */ 550 trace_print_hex_dump_seq(&iter->seq, "", DUMP_PREFIX_OFFSET, 16, 551 1, iter->ent, iter->ent_size, true); 552 553 return trace_handle_return(&iter->seq); 554 } 555 556 static struct trace_event_functions user_event_funcs = { 557 .trace = user_event_print_trace, 558 }; 559 560 static int destroy_user_event(struct user_event *user) 561 { 562 int ret = 0; 563 564 /* Must destroy fields before call removal */ 565 user_event_destroy_fields(user); 566 567 ret = trace_remove_event_call(&user->call); 568 569 if (ret) 570 return ret; 571 572 dyn_event_remove(&user->devent); 573 574 register_page_data[user->index] = 0; 575 clear_bit(user->index, page_bitmap); 576 hash_del(&user->node); 577 578 user_event_destroy_validators(user); 579 kfree(user->call.print_fmt); 580 kfree(EVENT_NAME(user)); 581 kfree(user); 582 583 return ret; 584 } 585 586 static struct user_event *find_user_event(char *name, u32 *outkey) 587 { 588 struct user_event *user; 589 u32 key = user_event_key(name); 590 591 *outkey = key; 592 593 hash_for_each_possible(register_table, user, node, key) 594 if (!strcmp(EVENT_NAME(user), name)) 595 return user; 596 597 return NULL; 598 } 599 600 static int user_event_validate(struct user_event *user, void *data, int len) 601 { 602 struct list_head *head = &user->validators; 603 struct user_event_validator *validator; 604 void *pos, *end = data + len; 605 u32 loc, offset, size; 606 607 list_for_each_entry(validator, head, link) { 608 pos = data + validator->offset; 609 610 /* Already done min_size check, no bounds check here */ 611 loc = *(u32 *)pos; 612 offset = loc & 0xffff; 613 size = loc >> 16; 614 615 if (likely(validator->flags & VALIDATOR_REL)) 616 pos += offset + sizeof(loc); 617 else 618 pos = data + offset; 619 620 pos += size; 621 622 if (unlikely(pos > end)) 623 return -EFAULT; 624 625 if (likely(validator->flags & VALIDATOR_ENSURE_NULL)) 626 if (unlikely(*(char *)(pos - 1) != '\0')) 627 return -EFAULT; 628 } 629 630 return 0; 631 } 632 633 /* 634 * Writes the user supplied payload out to a trace file. 635 */ 636 static void user_event_ftrace(struct user_event *user, struct iov_iter *i, 637 void *tpdata, bool *faulted) 638 { 639 struct trace_event_file *file; 640 struct trace_entry *entry; 641 struct trace_event_buffer event_buffer; 642 size_t size = sizeof(*entry) + i->count; 643 644 file = (struct trace_event_file *)tpdata; 645 646 if (!file || 647 !(file->flags & EVENT_FILE_FL_ENABLED) || 648 trace_trigger_soft_disabled(file)) 649 return; 650 651 /* Allocates and fills trace_entry, + 1 of this is data payload */ 652 entry = trace_event_buffer_reserve(&event_buffer, file, size); 653 654 if (unlikely(!entry)) 655 return; 656 657 if (unlikely(!copy_nofault(entry + 1, i->count, i))) 658 goto discard; 659 660 if (!list_empty(&user->validators) && 661 unlikely(user_event_validate(user, entry, size))) 662 goto discard; 663 664 trace_event_buffer_commit(&event_buffer); 665 666 return; 667 discard: 668 *faulted = true; 669 __trace_event_discard_commit(event_buffer.buffer, 670 event_buffer.event); 671 } 672 673 #ifdef CONFIG_PERF_EVENTS 674 static void user_event_bpf(struct user_event *user, struct iov_iter *i) 675 { 676 struct user_bpf_context context; 677 struct user_bpf_iter bpf_i; 678 char fast_data[MAX_STACK_BPF_DATA]; 679 void *temp = NULL; 680 681 if ((user->flags & FLAG_BPF_ITER) && iter_is_iovec(i)) { 682 /* Raw iterator */ 683 context.data_type = USER_BPF_DATA_ITER; 684 context.data_len = i->count; 685 context.iter = &bpf_i; 686 687 bpf_i.iov_offset = i->iov_offset; 688 bpf_i.iov = i->iov; 689 bpf_i.nr_segs = i->nr_segs; 690 } else if (i->nr_segs == 1 && iter_is_iovec(i)) { 691 /* Single buffer from user */ 692 context.data_type = USER_BPF_DATA_USER; 693 context.data_len = i->count; 694 context.udata = i->iov->iov_base + i->iov_offset; 695 } else { 696 /* Multi buffer from user */ 697 struct iov_iter copy = *i; 698 size_t copy_size = min_t(size_t, i->count, MAX_BPF_COPY_SIZE); 699 700 context.data_type = USER_BPF_DATA_KERNEL; 701 context.kdata = fast_data; 702 703 if (unlikely(copy_size > sizeof(fast_data))) { 704 temp = kmalloc(copy_size, GFP_NOWAIT); 705 706 if (temp) 707 context.kdata = temp; 708 else 709 copy_size = sizeof(fast_data); 710 } 711 712 context.data_len = copy_nofault(context.kdata, 713 copy_size, ©); 714 } 715 716 trace_call_bpf(&user->call, &context); 717 718 kfree(temp); 719 } 720 721 /* 722 * Writes the user supplied payload out to perf ring buffer or eBPF program. 723 */ 724 static void user_event_perf(struct user_event *user, struct iov_iter *i, 725 void *tpdata, bool *faulted) 726 { 727 struct hlist_head *perf_head; 728 729 if (bpf_prog_array_valid(&user->call)) 730 user_event_bpf(user, i); 731 732 perf_head = this_cpu_ptr(user->call.perf_events); 733 734 if (perf_head && !hlist_empty(perf_head)) { 735 struct trace_entry *perf_entry; 736 struct pt_regs *regs; 737 size_t size = sizeof(*perf_entry) + i->count; 738 int context; 739 740 perf_entry = perf_trace_buf_alloc(ALIGN(size, 8), 741 ®s, &context); 742 743 if (unlikely(!perf_entry)) 744 return; 745 746 perf_fetch_caller_regs(regs); 747 748 if (unlikely(!copy_nofault(perf_entry + 1, i->count, i))) 749 goto discard; 750 751 if (!list_empty(&user->validators) && 752 unlikely(user_event_validate(user, perf_entry, size))) 753 goto discard; 754 755 perf_trace_buf_submit(perf_entry, size, context, 756 user->call.event.type, 1, regs, 757 perf_head, NULL); 758 759 return; 760 discard: 761 *faulted = true; 762 perf_swevent_put_recursion_context(context); 763 } 764 } 765 #endif 766 767 /* 768 * Update the register page that is shared between user processes. 769 */ 770 static void update_reg_page_for(struct user_event *user) 771 { 772 struct tracepoint *tp = &user->tracepoint; 773 char status = 0; 774 775 if (atomic_read(&tp->key.enabled) > 0) { 776 struct tracepoint_func *probe_func_ptr; 777 user_event_func_t probe_func; 778 779 rcu_read_lock_sched(); 780 781 probe_func_ptr = rcu_dereference_sched(tp->funcs); 782 783 if (probe_func_ptr) { 784 do { 785 probe_func = probe_func_ptr->func; 786 787 if (probe_func == user_event_ftrace) 788 status |= EVENT_STATUS_FTRACE; 789 #ifdef CONFIG_PERF_EVENTS 790 else if (probe_func == user_event_perf) 791 status |= EVENT_STATUS_PERF; 792 #endif 793 else 794 status |= EVENT_STATUS_OTHER; 795 } while ((++probe_func_ptr)->func); 796 } 797 798 rcu_read_unlock_sched(); 799 } 800 801 register_page_data[user->index] = status; 802 } 803 804 /* 805 * Register callback for our events from tracing sub-systems. 806 */ 807 static int user_event_reg(struct trace_event_call *call, 808 enum trace_reg type, 809 void *data) 810 { 811 struct user_event *user = (struct user_event *)call->data; 812 int ret = 0; 813 814 if (!user) 815 return -ENOENT; 816 817 switch (type) { 818 case TRACE_REG_REGISTER: 819 ret = tracepoint_probe_register(call->tp, 820 call->class->probe, 821 data); 822 if (!ret) 823 goto inc; 824 break; 825 826 case TRACE_REG_UNREGISTER: 827 tracepoint_probe_unregister(call->tp, 828 call->class->probe, 829 data); 830 goto dec; 831 832 #ifdef CONFIG_PERF_EVENTS 833 case TRACE_REG_PERF_REGISTER: 834 ret = tracepoint_probe_register(call->tp, 835 call->class->perf_probe, 836 data); 837 if (!ret) 838 goto inc; 839 break; 840 841 case TRACE_REG_PERF_UNREGISTER: 842 tracepoint_probe_unregister(call->tp, 843 call->class->perf_probe, 844 data); 845 goto dec; 846 847 case TRACE_REG_PERF_OPEN: 848 case TRACE_REG_PERF_CLOSE: 849 case TRACE_REG_PERF_ADD: 850 case TRACE_REG_PERF_DEL: 851 break; 852 #endif 853 } 854 855 return ret; 856 inc: 857 atomic_inc(&user->refcnt); 858 update_reg_page_for(user); 859 return 0; 860 dec: 861 update_reg_page_for(user); 862 atomic_dec(&user->refcnt); 863 return 0; 864 } 865 866 static int user_event_create(const char *raw_command) 867 { 868 struct user_event *user; 869 char *name; 870 int ret; 871 872 if (!str_has_prefix(raw_command, USER_EVENTS_PREFIX)) 873 return -ECANCELED; 874 875 raw_command += USER_EVENTS_PREFIX_LEN; 876 raw_command = skip_spaces(raw_command); 877 878 name = kstrdup(raw_command, GFP_KERNEL); 879 880 if (!name) 881 return -ENOMEM; 882 883 mutex_lock(®_mutex); 884 ret = user_event_parse_cmd(name, &user); 885 mutex_unlock(®_mutex); 886 887 if (ret) 888 kfree(name); 889 890 return ret; 891 } 892 893 static int user_event_show(struct seq_file *m, struct dyn_event *ev) 894 { 895 struct user_event *user = container_of(ev, struct user_event, devent); 896 struct ftrace_event_field *field, *next; 897 struct list_head *head; 898 int depth = 0; 899 900 seq_printf(m, "%s%s", USER_EVENTS_PREFIX, EVENT_NAME(user)); 901 902 head = trace_get_fields(&user->call); 903 904 list_for_each_entry_safe_reverse(field, next, head, link) { 905 if (depth == 0) 906 seq_puts(m, " "); 907 else 908 seq_puts(m, "; "); 909 910 seq_printf(m, "%s %s", field->type, field->name); 911 912 if (str_has_prefix(field->type, "struct ")) 913 seq_printf(m, " %d", field->size); 914 915 depth++; 916 } 917 918 seq_puts(m, "\n"); 919 920 return 0; 921 } 922 923 static bool user_event_is_busy(struct dyn_event *ev) 924 { 925 struct user_event *user = container_of(ev, struct user_event, devent); 926 927 return atomic_read(&user->refcnt) != 0; 928 } 929 930 static int user_event_free(struct dyn_event *ev) 931 { 932 struct user_event *user = container_of(ev, struct user_event, devent); 933 934 if (atomic_read(&user->refcnt) != 0) 935 return -EBUSY; 936 937 return destroy_user_event(user); 938 } 939 940 static bool user_field_match(struct ftrace_event_field *field, int argc, 941 const char **argv, int *iout) 942 { 943 char *field_name, *arg_name; 944 int len, pos, i = *iout; 945 bool colon = false, match = false; 946 947 if (i >= argc) 948 return false; 949 950 len = MAX_FIELD_ARG_NAME; 951 field_name = kmalloc(len, GFP_KERNEL); 952 arg_name = kmalloc(len, GFP_KERNEL); 953 954 if (!arg_name || !field_name) 955 goto out; 956 957 pos = 0; 958 959 for (; i < argc; ++i) { 960 if (i != *iout) 961 pos += snprintf(arg_name + pos, len - pos, " "); 962 963 pos += snprintf(arg_name + pos, len - pos, argv[i]); 964 965 if (strchr(argv[i], ';')) { 966 ++i; 967 colon = true; 968 break; 969 } 970 } 971 972 pos = 0; 973 974 pos += snprintf(field_name + pos, len - pos, field->type); 975 pos += snprintf(field_name + pos, len - pos, " "); 976 pos += snprintf(field_name + pos, len - pos, field->name); 977 978 if (colon) 979 pos += snprintf(field_name + pos, len - pos, ";"); 980 981 *iout = i; 982 983 match = strcmp(arg_name, field_name) == 0; 984 out: 985 kfree(arg_name); 986 kfree(field_name); 987 988 return match; 989 } 990 991 static bool user_fields_match(struct user_event *user, int argc, 992 const char **argv) 993 { 994 struct ftrace_event_field *field, *next; 995 struct list_head *head = &user->fields; 996 int i = 0; 997 998 list_for_each_entry_safe_reverse(field, next, head, link) 999 if (!user_field_match(field, argc, argv, &i)) 1000 return false; 1001 1002 if (i != argc) 1003 return false; 1004 1005 return true; 1006 } 1007 1008 static bool user_event_match(const char *system, const char *event, 1009 int argc, const char **argv, struct dyn_event *ev) 1010 { 1011 struct user_event *user = container_of(ev, struct user_event, devent); 1012 bool match; 1013 1014 match = strcmp(EVENT_NAME(user), event) == 0 && 1015 (!system || strcmp(system, USER_EVENTS_SYSTEM) == 0); 1016 1017 if (match && argc > 0) 1018 match = user_fields_match(user, argc, argv); 1019 1020 return match; 1021 } 1022 1023 static struct dyn_event_operations user_event_dops = { 1024 .create = user_event_create, 1025 .show = user_event_show, 1026 .is_busy = user_event_is_busy, 1027 .free = user_event_free, 1028 .match = user_event_match, 1029 }; 1030 1031 static int user_event_trace_register(struct user_event *user) 1032 { 1033 int ret; 1034 1035 ret = register_trace_event(&user->call.event); 1036 1037 if (!ret) 1038 return -ENODEV; 1039 1040 ret = trace_add_event_call(&user->call); 1041 1042 if (ret) 1043 unregister_trace_event(&user->call.event); 1044 1045 return ret; 1046 } 1047 1048 /* 1049 * Parses the event name, arguments and flags then registers if successful. 1050 * The name buffer lifetime is owned by this method for success cases only. 1051 */ 1052 static int user_event_parse(char *name, char *args, char *flags, 1053 struct user_event **newuser) 1054 { 1055 int ret; 1056 int index; 1057 u32 key; 1058 struct user_event *user = find_user_event(name, &key); 1059 1060 if (user) { 1061 *newuser = user; 1062 /* 1063 * Name is allocated by caller, free it since it already exists. 1064 * Caller only worries about failure cases for freeing. 1065 */ 1066 kfree(name); 1067 return 0; 1068 } 1069 1070 index = find_first_zero_bit(page_bitmap, MAX_EVENTS); 1071 1072 if (index == MAX_EVENTS) 1073 return -EMFILE; 1074 1075 user = kzalloc(sizeof(*user), GFP_KERNEL); 1076 1077 if (!user) 1078 return -ENOMEM; 1079 1080 INIT_LIST_HEAD(&user->class.fields); 1081 INIT_LIST_HEAD(&user->fields); 1082 INIT_LIST_HEAD(&user->validators); 1083 1084 user->tracepoint.name = name; 1085 1086 user_event_parse_flags(user, flags); 1087 1088 ret = user_event_parse_fields(user, args); 1089 1090 if (ret) 1091 goto put_user; 1092 1093 ret = user_event_create_print_fmt(user); 1094 1095 if (ret) 1096 goto put_user; 1097 1098 user->call.data = user; 1099 user->call.class = &user->class; 1100 user->call.name = name; 1101 user->call.flags = TRACE_EVENT_FL_TRACEPOINT; 1102 user->call.tp = &user->tracepoint; 1103 user->call.event.funcs = &user_event_funcs; 1104 1105 user->class.system = USER_EVENTS_SYSTEM; 1106 user->class.fields_array = user_event_fields_array; 1107 user->class.get_fields = user_event_get_fields; 1108 user->class.reg = user_event_reg; 1109 user->class.probe = user_event_ftrace; 1110 #ifdef CONFIG_PERF_EVENTS 1111 user->class.perf_probe = user_event_perf; 1112 #endif 1113 1114 mutex_lock(&event_mutex); 1115 ret = user_event_trace_register(user); 1116 mutex_unlock(&event_mutex); 1117 1118 if (ret) 1119 goto put_user; 1120 1121 user->index = index; 1122 dyn_event_init(&user->devent, &user_event_dops); 1123 dyn_event_add(&user->devent, &user->call); 1124 set_bit(user->index, page_bitmap); 1125 hash_add(register_table, &user->node, key); 1126 1127 *newuser = user; 1128 return 0; 1129 put_user: 1130 user_event_destroy_fields(user); 1131 user_event_destroy_validators(user); 1132 kfree(user); 1133 return ret; 1134 } 1135 1136 /* 1137 * Deletes a previously created event if it is no longer being used. 1138 */ 1139 static int delete_user_event(char *name) 1140 { 1141 u32 key; 1142 int ret; 1143 struct user_event *user = find_user_event(name, &key); 1144 1145 if (!user) 1146 return -ENOENT; 1147 1148 if (atomic_read(&user->refcnt) != 0) 1149 return -EBUSY; 1150 1151 mutex_lock(&event_mutex); 1152 ret = destroy_user_event(user); 1153 mutex_unlock(&event_mutex); 1154 1155 return ret; 1156 } 1157 1158 /* 1159 * Validates the user payload and writes via iterator. 1160 */ 1161 static ssize_t user_events_write_core(struct file *file, struct iov_iter *i) 1162 { 1163 struct user_event_refs *refs; 1164 struct user_event *user = NULL; 1165 struct tracepoint *tp; 1166 ssize_t ret = i->count; 1167 int idx; 1168 1169 if (unlikely(copy_from_iter(&idx, sizeof(idx), i) != sizeof(idx))) 1170 return -EFAULT; 1171 1172 rcu_read_lock_sched(); 1173 1174 refs = rcu_dereference_sched(file->private_data); 1175 1176 /* 1177 * The refs->events array is protected by RCU, and new items may be 1178 * added. But the user retrieved from indexing into the events array 1179 * shall be immutable while the file is opened. 1180 */ 1181 if (likely(refs && idx < refs->count)) 1182 user = refs->events[idx]; 1183 1184 rcu_read_unlock_sched(); 1185 1186 if (unlikely(user == NULL)) 1187 return -ENOENT; 1188 1189 if (unlikely(i->count < user->min_size)) 1190 return -EINVAL; 1191 1192 tp = &user->tracepoint; 1193 1194 /* 1195 * It's possible key.enabled disables after this check, however 1196 * we don't mind if a few events are included in this condition. 1197 */ 1198 if (likely(atomic_read(&tp->key.enabled) > 0)) { 1199 struct tracepoint_func *probe_func_ptr; 1200 user_event_func_t probe_func; 1201 struct iov_iter copy; 1202 void *tpdata; 1203 bool faulted; 1204 1205 if (unlikely(fault_in_iov_iter_readable(i, i->count))) 1206 return -EFAULT; 1207 1208 faulted = false; 1209 1210 rcu_read_lock_sched(); 1211 1212 probe_func_ptr = rcu_dereference_sched(tp->funcs); 1213 1214 if (probe_func_ptr) { 1215 do { 1216 copy = *i; 1217 probe_func = probe_func_ptr->func; 1218 tpdata = probe_func_ptr->data; 1219 probe_func(user, ©, tpdata, &faulted); 1220 } while ((++probe_func_ptr)->func); 1221 } 1222 1223 rcu_read_unlock_sched(); 1224 1225 if (unlikely(faulted)) 1226 return -EFAULT; 1227 } 1228 1229 return ret; 1230 } 1231 1232 static ssize_t user_events_write(struct file *file, const char __user *ubuf, 1233 size_t count, loff_t *ppos) 1234 { 1235 struct iovec iov; 1236 struct iov_iter i; 1237 1238 if (unlikely(*ppos != 0)) 1239 return -EFAULT; 1240 1241 if (unlikely(import_single_range(READ, (char *)ubuf, count, &iov, &i))) 1242 return -EFAULT; 1243 1244 return user_events_write_core(file, &i); 1245 } 1246 1247 static ssize_t user_events_write_iter(struct kiocb *kp, struct iov_iter *i) 1248 { 1249 return user_events_write_core(kp->ki_filp, i); 1250 } 1251 1252 static int user_events_ref_add(struct file *file, struct user_event *user) 1253 { 1254 struct user_event_refs *refs, *new_refs; 1255 int i, size, count = 0; 1256 1257 refs = rcu_dereference_protected(file->private_data, 1258 lockdep_is_held(®_mutex)); 1259 1260 if (refs) { 1261 count = refs->count; 1262 1263 for (i = 0; i < count; ++i) 1264 if (refs->events[i] == user) 1265 return i; 1266 } 1267 1268 size = struct_size(refs, events, count + 1); 1269 1270 new_refs = kzalloc(size, GFP_KERNEL); 1271 1272 if (!new_refs) 1273 return -ENOMEM; 1274 1275 new_refs->count = count + 1; 1276 1277 for (i = 0; i < count; ++i) 1278 new_refs->events[i] = refs->events[i]; 1279 1280 new_refs->events[i] = user; 1281 1282 atomic_inc(&user->refcnt); 1283 1284 rcu_assign_pointer(file->private_data, new_refs); 1285 1286 if (refs) 1287 kfree_rcu(refs, rcu); 1288 1289 return i; 1290 } 1291 1292 static long user_reg_get(struct user_reg __user *ureg, struct user_reg *kreg) 1293 { 1294 u32 size; 1295 long ret; 1296 1297 ret = get_user(size, &ureg->size); 1298 1299 if (ret) 1300 return ret; 1301 1302 if (size > PAGE_SIZE) 1303 return -E2BIG; 1304 1305 return copy_struct_from_user(kreg, sizeof(*kreg), ureg, size); 1306 } 1307 1308 /* 1309 * Registers a user_event on behalf of a user process. 1310 */ 1311 static long user_events_ioctl_reg(struct file *file, unsigned long uarg) 1312 { 1313 struct user_reg __user *ureg = (struct user_reg __user *)uarg; 1314 struct user_reg reg; 1315 struct user_event *user; 1316 char *name; 1317 long ret; 1318 1319 ret = user_reg_get(ureg, ®); 1320 1321 if (ret) 1322 return ret; 1323 1324 name = strndup_user((const char __user *)(uintptr_t)reg.name_args, 1325 MAX_EVENT_DESC); 1326 1327 if (IS_ERR(name)) { 1328 ret = PTR_ERR(name); 1329 return ret; 1330 } 1331 1332 ret = user_event_parse_cmd(name, &user); 1333 1334 if (ret) { 1335 kfree(name); 1336 return ret; 1337 } 1338 1339 ret = user_events_ref_add(file, user); 1340 1341 /* Positive number is index and valid */ 1342 if (ret < 0) 1343 return ret; 1344 1345 put_user((u32)ret, &ureg->write_index); 1346 put_user(user->index, &ureg->status_index); 1347 1348 return 0; 1349 } 1350 1351 /* 1352 * Deletes a user_event on behalf of a user process. 1353 */ 1354 static long user_events_ioctl_del(struct file *file, unsigned long uarg) 1355 { 1356 void __user *ubuf = (void __user *)uarg; 1357 char *name; 1358 long ret; 1359 1360 name = strndup_user(ubuf, MAX_EVENT_DESC); 1361 1362 if (IS_ERR(name)) 1363 return PTR_ERR(name); 1364 1365 ret = delete_user_event(name); 1366 1367 kfree(name); 1368 1369 return ret; 1370 } 1371 1372 /* 1373 * Handles the ioctl from user mode to register or alter operations. 1374 */ 1375 static long user_events_ioctl(struct file *file, unsigned int cmd, 1376 unsigned long uarg) 1377 { 1378 long ret = -ENOTTY; 1379 1380 switch (cmd) { 1381 case DIAG_IOCSREG: 1382 mutex_lock(®_mutex); 1383 ret = user_events_ioctl_reg(file, uarg); 1384 mutex_unlock(®_mutex); 1385 break; 1386 1387 case DIAG_IOCSDEL: 1388 mutex_lock(®_mutex); 1389 ret = user_events_ioctl_del(file, uarg); 1390 mutex_unlock(®_mutex); 1391 break; 1392 } 1393 1394 return ret; 1395 } 1396 1397 /* 1398 * Handles the final close of the file from user mode. 1399 */ 1400 static int user_events_release(struct inode *node, struct file *file) 1401 { 1402 struct user_event_refs *refs; 1403 struct user_event *user; 1404 int i; 1405 1406 /* 1407 * Ensure refs cannot change under any situation by taking the 1408 * register mutex during the final freeing of the references. 1409 */ 1410 mutex_lock(®_mutex); 1411 1412 refs = file->private_data; 1413 1414 if (!refs) 1415 goto out; 1416 1417 /* 1418 * The lifetime of refs has reached an end, it's tied to this file. 1419 * The underlying user_events are ref counted, and cannot be freed. 1420 * After this decrement, the user_events may be freed elsewhere. 1421 */ 1422 for (i = 0; i < refs->count; ++i) { 1423 user = refs->events[i]; 1424 1425 if (user) 1426 atomic_dec(&user->refcnt); 1427 } 1428 out: 1429 file->private_data = NULL; 1430 1431 mutex_unlock(®_mutex); 1432 1433 kfree(refs); 1434 1435 return 0; 1436 } 1437 1438 static const struct file_operations user_data_fops = { 1439 .write = user_events_write, 1440 .write_iter = user_events_write_iter, 1441 .unlocked_ioctl = user_events_ioctl, 1442 .release = user_events_release, 1443 }; 1444 1445 /* 1446 * Maps the shared page into the user process for checking if event is enabled. 1447 */ 1448 static int user_status_mmap(struct file *file, struct vm_area_struct *vma) 1449 { 1450 unsigned long size = vma->vm_end - vma->vm_start; 1451 1452 if (size != MAX_EVENTS) 1453 return -EINVAL; 1454 1455 return remap_pfn_range(vma, vma->vm_start, 1456 virt_to_phys(register_page_data) >> PAGE_SHIFT, 1457 size, vm_get_page_prot(VM_READ)); 1458 } 1459 1460 static void *user_seq_start(struct seq_file *m, loff_t *pos) 1461 { 1462 if (*pos) 1463 return NULL; 1464 1465 return (void *)1; 1466 } 1467 1468 static void *user_seq_next(struct seq_file *m, void *p, loff_t *pos) 1469 { 1470 ++*pos; 1471 return NULL; 1472 } 1473 1474 static void user_seq_stop(struct seq_file *m, void *p) 1475 { 1476 } 1477 1478 static int user_seq_show(struct seq_file *m, void *p) 1479 { 1480 struct user_event *user; 1481 char status; 1482 int i, active = 0, busy = 0, flags; 1483 1484 mutex_lock(®_mutex); 1485 1486 hash_for_each(register_table, i, user, node) { 1487 status = register_page_data[user->index]; 1488 flags = user->flags; 1489 1490 seq_printf(m, "%d:%s", user->index, EVENT_NAME(user)); 1491 1492 if (flags != 0 || status != 0) 1493 seq_puts(m, " #"); 1494 1495 if (status != 0) { 1496 seq_puts(m, " Used by"); 1497 if (status & EVENT_STATUS_FTRACE) 1498 seq_puts(m, " ftrace"); 1499 if (status & EVENT_STATUS_PERF) 1500 seq_puts(m, " perf"); 1501 if (status & EVENT_STATUS_OTHER) 1502 seq_puts(m, " other"); 1503 busy++; 1504 } 1505 1506 if (flags & FLAG_BPF_ITER) 1507 seq_puts(m, " FLAG:BPF_ITER"); 1508 1509 seq_puts(m, "\n"); 1510 active++; 1511 } 1512 1513 mutex_unlock(®_mutex); 1514 1515 seq_puts(m, "\n"); 1516 seq_printf(m, "Active: %d\n", active); 1517 seq_printf(m, "Busy: %d\n", busy); 1518 seq_printf(m, "Max: %ld\n", MAX_EVENTS); 1519 1520 return 0; 1521 } 1522 1523 static const struct seq_operations user_seq_ops = { 1524 .start = user_seq_start, 1525 .next = user_seq_next, 1526 .stop = user_seq_stop, 1527 .show = user_seq_show, 1528 }; 1529 1530 static int user_status_open(struct inode *node, struct file *file) 1531 { 1532 return seq_open(file, &user_seq_ops); 1533 } 1534 1535 static const struct file_operations user_status_fops = { 1536 .open = user_status_open, 1537 .mmap = user_status_mmap, 1538 .read = seq_read, 1539 .llseek = seq_lseek, 1540 .release = seq_release, 1541 }; 1542 1543 /* 1544 * Creates a set of tracefs files to allow user mode interactions. 1545 */ 1546 static int create_user_tracefs(void) 1547 { 1548 struct dentry *edata, *emmap; 1549 1550 edata = tracefs_create_file("user_events_data", TRACE_MODE_WRITE, 1551 NULL, NULL, &user_data_fops); 1552 1553 if (!edata) { 1554 pr_warn("Could not create tracefs 'user_events_data' entry\n"); 1555 goto err; 1556 } 1557 1558 /* mmap with MAP_SHARED requires writable fd */ 1559 emmap = tracefs_create_file("user_events_status", TRACE_MODE_WRITE, 1560 NULL, NULL, &user_status_fops); 1561 1562 if (!emmap) { 1563 tracefs_remove(edata); 1564 pr_warn("Could not create tracefs 'user_events_mmap' entry\n"); 1565 goto err; 1566 } 1567 1568 return 0; 1569 err: 1570 return -ENODEV; 1571 } 1572 1573 static void set_page_reservations(bool set) 1574 { 1575 int page; 1576 1577 for (page = 0; page < MAX_PAGES; ++page) { 1578 void *addr = register_page_data + (PAGE_SIZE * page); 1579 1580 if (set) 1581 SetPageReserved(virt_to_page(addr)); 1582 else 1583 ClearPageReserved(virt_to_page(addr)); 1584 } 1585 } 1586 1587 static int __init trace_events_user_init(void) 1588 { 1589 int ret; 1590 1591 /* Zero all bits beside 0 (which is reserved for failures) */ 1592 bitmap_zero(page_bitmap, MAX_EVENTS); 1593 set_bit(0, page_bitmap); 1594 1595 register_page_data = kzalloc(MAX_EVENTS, GFP_KERNEL); 1596 1597 if (!register_page_data) 1598 return -ENOMEM; 1599 1600 set_page_reservations(true); 1601 1602 ret = create_user_tracefs(); 1603 1604 if (ret) { 1605 pr_warn("user_events could not register with tracefs\n"); 1606 set_page_reservations(false); 1607 kfree(register_page_data); 1608 return ret; 1609 } 1610 1611 if (dyn_event_register(&user_event_dops)) 1612 pr_warn("user_events could not register with dyn_events\n"); 1613 1614 return 0; 1615 } 1616 1617 fs_initcall(trace_events_user_init); 1618