1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Fprobe-based tracing events 4 * Copyright (C) 2022 Google LLC. 5 */ 6 #define pr_fmt(fmt) "trace_fprobe: " fmt 7 #include <asm/ptrace.h> 8 9 #include <linux/fprobe.h> 10 #include <linux/module.h> 11 #include <linux/rculist.h> 12 #include <linux/security.h> 13 #include <linux/tracepoint.h> 14 #include <linux/uaccess.h> 15 16 #include "trace_dynevent.h" 17 #include "trace_probe.h" 18 #include "trace_probe_kernel.h" 19 #include "trace_probe_tmpl.h" 20 21 #define FPROBE_EVENT_SYSTEM "fprobes" 22 #define TRACEPOINT_EVENT_SYSTEM "tracepoints" 23 #define RETHOOK_MAXACTIVE_MAX 4096 24 #define TRACEPOINT_STUB ERR_PTR(-ENOENT) 25 26 static int trace_fprobe_create(const char *raw_command); 27 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev); 28 static int trace_fprobe_release(struct dyn_event *ev); 29 static bool trace_fprobe_is_busy(struct dyn_event *ev); 30 static bool trace_fprobe_match(const char *system, const char *event, 31 int argc, const char **argv, struct dyn_event *ev); 32 33 static struct dyn_event_operations trace_fprobe_ops = { 34 .create = trace_fprobe_create, 35 .show = trace_fprobe_show, 36 .is_busy = trace_fprobe_is_busy, 37 .free = trace_fprobe_release, 38 .match = trace_fprobe_match, 39 }; 40 41 /* 42 * Fprobe event core functions 43 */ 44 struct trace_fprobe { 45 struct dyn_event devent; 46 struct fprobe fp; 47 const char *symbol; 48 struct tracepoint *tpoint; 49 struct module *mod; 50 struct trace_probe tp; 51 }; 52 53 static bool is_trace_fprobe(struct dyn_event *ev) 54 { 55 return ev->ops == &trace_fprobe_ops; 56 } 57 58 static struct trace_fprobe *to_trace_fprobe(struct dyn_event *ev) 59 { 60 return container_of(ev, struct trace_fprobe, devent); 61 } 62 63 /** 64 * for_each_trace_fprobe - iterate over the trace_fprobe list 65 * @pos: the struct trace_fprobe * for each entry 66 * @dpos: the struct dyn_event * to use as a loop cursor 67 */ 68 #define for_each_trace_fprobe(pos, dpos) \ 69 for_each_dyn_event(dpos) \ 70 if (is_trace_fprobe(dpos) && (pos = to_trace_fprobe(dpos))) 71 72 static bool trace_fprobe_is_return(struct trace_fprobe *tf) 73 { 74 return tf->fp.exit_handler != NULL; 75 } 76 77 static bool trace_fprobe_is_tracepoint(struct trace_fprobe *tf) 78 { 79 return tf->tpoint != NULL; 80 } 81 82 static const char *trace_fprobe_symbol(struct trace_fprobe *tf) 83 { 84 return tf->symbol ? tf->symbol : "unknown"; 85 } 86 87 static bool trace_fprobe_is_busy(struct dyn_event *ev) 88 { 89 struct trace_fprobe *tf = to_trace_fprobe(ev); 90 91 return trace_probe_is_enabled(&tf->tp); 92 } 93 94 static bool trace_fprobe_match_command_head(struct trace_fprobe *tf, 95 int argc, const char **argv) 96 { 97 char buf[MAX_ARGSTR_LEN + 1]; 98 99 if (!argc) 100 return true; 101 102 snprintf(buf, sizeof(buf), "%s", trace_fprobe_symbol(tf)); 103 if (strcmp(buf, argv[0])) 104 return false; 105 argc--; argv++; 106 107 return trace_probe_match_command_args(&tf->tp, argc, argv); 108 } 109 110 static bool trace_fprobe_match(const char *system, const char *event, 111 int argc, const char **argv, struct dyn_event *ev) 112 { 113 struct trace_fprobe *tf = to_trace_fprobe(ev); 114 115 if (event[0] != '\0' && strcmp(trace_probe_name(&tf->tp), event)) 116 return false; 117 118 if (system && strcmp(trace_probe_group_name(&tf->tp), system)) 119 return false; 120 121 return trace_fprobe_match_command_head(tf, argc, argv); 122 } 123 124 static bool trace_fprobe_is_registered(struct trace_fprobe *tf) 125 { 126 return fprobe_is_registered(&tf->fp); 127 } 128 129 /* 130 * Note that we don't verify the fetch_insn code, since it does not come 131 * from user space. 132 */ 133 static int 134 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata, 135 void *dest, void *base) 136 { 137 struct ftrace_regs *fregs = rec; 138 unsigned long val; 139 int ret; 140 141 retry: 142 /* 1st stage: get value from context */ 143 switch (code->op) { 144 case FETCH_OP_STACK: 145 val = ftrace_regs_get_kernel_stack_nth(fregs, code->param); 146 break; 147 case FETCH_OP_STACKP: 148 val = ftrace_regs_get_stack_pointer(fregs); 149 break; 150 case FETCH_OP_RETVAL: 151 val = ftrace_regs_get_return_value(fregs); 152 break; 153 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API 154 case FETCH_OP_ARG: 155 val = ftrace_regs_get_argument(fregs, code->param); 156 break; 157 case FETCH_OP_EDATA: 158 val = *(unsigned long *)((unsigned long)edata + code->offset); 159 break; 160 #endif 161 case FETCH_NOP_SYMBOL: /* Ignore a place holder */ 162 code++; 163 goto retry; 164 default: 165 ret = process_common_fetch_insn(code, &val); 166 if (ret < 0) 167 return ret; 168 } 169 code++; 170 171 return process_fetch_insn_bottom(code, val, dest, base); 172 } 173 NOKPROBE_SYMBOL(process_fetch_insn) 174 175 /* function entry handler */ 176 static nokprobe_inline void 177 __fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip, 178 struct ftrace_regs *fregs, 179 struct trace_event_file *trace_file) 180 { 181 struct fentry_trace_entry_head *entry; 182 struct trace_event_call *call = trace_probe_event_call(&tf->tp); 183 struct trace_event_buffer fbuffer; 184 int dsize; 185 186 if (WARN_ON_ONCE(call != trace_file->event_call)) 187 return; 188 189 if (trace_trigger_soft_disabled(trace_file)) 190 return; 191 192 dsize = __get_data_size(&tf->tp, fregs, NULL); 193 194 entry = trace_event_buffer_reserve(&fbuffer, trace_file, 195 sizeof(*entry) + tf->tp.size + dsize); 196 if (!entry) 197 return; 198 199 fbuffer.regs = ftrace_get_regs(fregs); 200 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event); 201 entry->ip = entry_ip; 202 store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize); 203 204 trace_event_buffer_commit(&fbuffer); 205 } 206 207 static void 208 fentry_trace_func(struct trace_fprobe *tf, unsigned long entry_ip, 209 struct ftrace_regs *fregs) 210 { 211 struct event_file_link *link; 212 213 trace_probe_for_each_link_rcu(link, &tf->tp) 214 __fentry_trace_func(tf, entry_ip, fregs, link->file); 215 } 216 NOKPROBE_SYMBOL(fentry_trace_func); 217 218 static nokprobe_inline 219 void store_fprobe_entry_data(void *edata, struct trace_probe *tp, struct ftrace_regs *fregs) 220 { 221 struct probe_entry_arg *earg = tp->entry_arg; 222 unsigned long val = 0; 223 int i; 224 225 if (!earg) 226 return; 227 228 for (i = 0; i < earg->size; i++) { 229 struct fetch_insn *code = &earg->code[i]; 230 231 switch (code->op) { 232 case FETCH_OP_ARG: 233 val = ftrace_regs_get_argument(fregs, code->param); 234 break; 235 case FETCH_OP_ST_EDATA: 236 *(unsigned long *)((unsigned long)edata + code->offset) = val; 237 break; 238 case FETCH_OP_END: 239 goto end; 240 default: 241 break; 242 } 243 } 244 end: 245 return; 246 } 247 248 /* function exit handler */ 249 static int trace_fprobe_entry_handler(struct fprobe *fp, unsigned long entry_ip, 250 unsigned long ret_ip, struct ftrace_regs *fregs, 251 void *entry_data) 252 { 253 struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp); 254 255 if (tf->tp.entry_arg) 256 store_fprobe_entry_data(entry_data, &tf->tp, fregs); 257 258 return 0; 259 } 260 NOKPROBE_SYMBOL(trace_fprobe_entry_handler) 261 262 static nokprobe_inline void 263 __fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip, 264 unsigned long ret_ip, struct ftrace_regs *fregs, 265 void *entry_data, struct trace_event_file *trace_file) 266 { 267 struct fexit_trace_entry_head *entry; 268 struct trace_event_buffer fbuffer; 269 struct trace_event_call *call = trace_probe_event_call(&tf->tp); 270 int dsize; 271 272 if (WARN_ON_ONCE(call != trace_file->event_call)) 273 return; 274 275 if (trace_trigger_soft_disabled(trace_file)) 276 return; 277 278 dsize = __get_data_size(&tf->tp, fregs, entry_data); 279 280 entry = trace_event_buffer_reserve(&fbuffer, trace_file, 281 sizeof(*entry) + tf->tp.size + dsize); 282 if (!entry) 283 return; 284 285 fbuffer.regs = ftrace_get_regs(fregs); 286 entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event); 287 entry->func = entry_ip; 288 entry->ret_ip = ret_ip; 289 store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize); 290 291 trace_event_buffer_commit(&fbuffer); 292 } 293 294 static void 295 fexit_trace_func(struct trace_fprobe *tf, unsigned long entry_ip, 296 unsigned long ret_ip, struct ftrace_regs *fregs, void *entry_data) 297 { 298 struct event_file_link *link; 299 300 trace_probe_for_each_link_rcu(link, &tf->tp) 301 __fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data, link->file); 302 } 303 NOKPROBE_SYMBOL(fexit_trace_func); 304 305 #ifdef CONFIG_PERF_EVENTS 306 307 static int fentry_perf_func(struct trace_fprobe *tf, unsigned long entry_ip, 308 struct ftrace_regs *fregs) 309 { 310 struct trace_event_call *call = trace_probe_event_call(&tf->tp); 311 struct fentry_trace_entry_head *entry; 312 struct hlist_head *head; 313 int size, __size, dsize; 314 struct pt_regs *regs; 315 int rctx; 316 317 head = this_cpu_ptr(call->perf_events); 318 if (hlist_empty(head)) 319 return 0; 320 321 dsize = __get_data_size(&tf->tp, fregs, NULL); 322 __size = sizeof(*entry) + tf->tp.size + dsize; 323 size = ALIGN(__size + sizeof(u32), sizeof(u64)); 324 size -= sizeof(u32); 325 326 entry = perf_trace_buf_alloc(size, ®s, &rctx); 327 if (!entry) 328 return 0; 329 330 regs = ftrace_fill_perf_regs(fregs, regs); 331 332 entry->ip = entry_ip; 333 memset(&entry[1], 0, dsize); 334 store_trace_args(&entry[1], &tf->tp, fregs, NULL, sizeof(*entry), dsize); 335 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs, 336 head, NULL); 337 return 0; 338 } 339 NOKPROBE_SYMBOL(fentry_perf_func); 340 341 static void 342 fexit_perf_func(struct trace_fprobe *tf, unsigned long entry_ip, 343 unsigned long ret_ip, struct ftrace_regs *fregs, 344 void *entry_data) 345 { 346 struct trace_event_call *call = trace_probe_event_call(&tf->tp); 347 struct fexit_trace_entry_head *entry; 348 struct hlist_head *head; 349 int size, __size, dsize; 350 struct pt_regs *regs; 351 int rctx; 352 353 head = this_cpu_ptr(call->perf_events); 354 if (hlist_empty(head)) 355 return; 356 357 dsize = __get_data_size(&tf->tp, fregs, entry_data); 358 __size = sizeof(*entry) + tf->tp.size + dsize; 359 size = ALIGN(__size + sizeof(u32), sizeof(u64)); 360 size -= sizeof(u32); 361 362 entry = perf_trace_buf_alloc(size, ®s, &rctx); 363 if (!entry) 364 return; 365 366 regs = ftrace_fill_perf_regs(fregs, regs); 367 368 entry->func = entry_ip; 369 entry->ret_ip = ret_ip; 370 store_trace_args(&entry[1], &tf->tp, fregs, entry_data, sizeof(*entry), dsize); 371 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs, 372 head, NULL); 373 } 374 NOKPROBE_SYMBOL(fexit_perf_func); 375 #endif /* CONFIG_PERF_EVENTS */ 376 377 static int fentry_dispatcher(struct fprobe *fp, unsigned long entry_ip, 378 unsigned long ret_ip, struct ftrace_regs *fregs, 379 void *entry_data) 380 { 381 struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp); 382 int ret = 0; 383 384 if (trace_probe_test_flag(&tf->tp, TP_FLAG_TRACE)) 385 fentry_trace_func(tf, entry_ip, fregs); 386 387 #ifdef CONFIG_PERF_EVENTS 388 if (trace_probe_test_flag(&tf->tp, TP_FLAG_PROFILE)) 389 ret = fentry_perf_func(tf, entry_ip, fregs); 390 #endif 391 return ret; 392 } 393 NOKPROBE_SYMBOL(fentry_dispatcher); 394 395 static void fexit_dispatcher(struct fprobe *fp, unsigned long entry_ip, 396 unsigned long ret_ip, struct ftrace_regs *fregs, 397 void *entry_data) 398 { 399 struct trace_fprobe *tf = container_of(fp, struct trace_fprobe, fp); 400 401 if (trace_probe_test_flag(&tf->tp, TP_FLAG_TRACE)) 402 fexit_trace_func(tf, entry_ip, ret_ip, fregs, entry_data); 403 #ifdef CONFIG_PERF_EVENTS 404 if (trace_probe_test_flag(&tf->tp, TP_FLAG_PROFILE)) 405 fexit_perf_func(tf, entry_ip, ret_ip, fregs, entry_data); 406 #endif 407 } 408 NOKPROBE_SYMBOL(fexit_dispatcher); 409 410 static void free_trace_fprobe(struct trace_fprobe *tf) 411 { 412 if (tf) { 413 trace_probe_cleanup(&tf->tp); 414 kfree(tf->symbol); 415 kfree(tf); 416 } 417 } 418 419 /* Since alloc_trace_fprobe() can return error, check the pointer is ERR too. */ 420 DEFINE_FREE(free_trace_fprobe, struct trace_fprobe *, if (!IS_ERR_OR_NULL(_T)) free_trace_fprobe(_T)) 421 422 /* 423 * Allocate new trace_probe and initialize it (including fprobe). 424 */ 425 static struct trace_fprobe *alloc_trace_fprobe(const char *group, 426 const char *event, 427 const char *symbol, 428 struct tracepoint *tpoint, 429 struct module *mod, 430 int nargs, bool is_return) 431 { 432 struct trace_fprobe *tf __free(free_trace_fprobe) = NULL; 433 int ret = -ENOMEM; 434 435 tf = kzalloc(struct_size(tf, tp.args, nargs), GFP_KERNEL); 436 if (!tf) 437 return ERR_PTR(ret); 438 439 tf->symbol = kstrdup(symbol, GFP_KERNEL); 440 if (!tf->symbol) 441 return ERR_PTR(-ENOMEM); 442 443 if (is_return) 444 tf->fp.exit_handler = fexit_dispatcher; 445 else 446 tf->fp.entry_handler = fentry_dispatcher; 447 448 tf->tpoint = tpoint; 449 tf->mod = mod; 450 451 ret = trace_probe_init(&tf->tp, event, group, false, nargs); 452 if (ret < 0) 453 return ERR_PTR(ret); 454 455 dyn_event_init(&tf->devent, &trace_fprobe_ops); 456 return_ptr(tf); 457 } 458 459 static struct trace_fprobe *find_trace_fprobe(const char *event, 460 const char *group) 461 { 462 struct dyn_event *pos; 463 struct trace_fprobe *tf; 464 465 for_each_trace_fprobe(tf, pos) 466 if (strcmp(trace_probe_name(&tf->tp), event) == 0 && 467 strcmp(trace_probe_group_name(&tf->tp), group) == 0) 468 return tf; 469 return NULL; 470 } 471 472 static inline int __enable_trace_fprobe(struct trace_fprobe *tf) 473 { 474 if (trace_fprobe_is_registered(tf)) 475 enable_fprobe(&tf->fp); 476 477 return 0; 478 } 479 480 static void __disable_trace_fprobe(struct trace_probe *tp) 481 { 482 struct trace_fprobe *tf; 483 484 list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) { 485 if (!trace_fprobe_is_registered(tf)) 486 continue; 487 disable_fprobe(&tf->fp); 488 } 489 } 490 491 /* 492 * Enable trace_probe 493 * if the file is NULL, enable "perf" handler, or enable "trace" handler. 494 */ 495 static int enable_trace_fprobe(struct trace_event_call *call, 496 struct trace_event_file *file) 497 { 498 struct trace_probe *tp; 499 struct trace_fprobe *tf; 500 bool enabled; 501 int ret = 0; 502 503 tp = trace_probe_primary_from_call(call); 504 if (WARN_ON_ONCE(!tp)) 505 return -ENODEV; 506 enabled = trace_probe_is_enabled(tp); 507 508 /* This also changes "enabled" state */ 509 if (file) { 510 ret = trace_probe_add_file(tp, file); 511 if (ret) 512 return ret; 513 } else 514 trace_probe_set_flag(tp, TP_FLAG_PROFILE); 515 516 if (!enabled) { 517 list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) { 518 /* TODO: check the fprobe is gone */ 519 __enable_trace_fprobe(tf); 520 } 521 } 522 523 return 0; 524 } 525 526 /* 527 * Disable trace_probe 528 * if the file is NULL, disable "perf" handler, or disable "trace" handler. 529 */ 530 static int disable_trace_fprobe(struct trace_event_call *call, 531 struct trace_event_file *file) 532 { 533 struct trace_probe *tp; 534 535 tp = trace_probe_primary_from_call(call); 536 if (WARN_ON_ONCE(!tp)) 537 return -ENODEV; 538 539 if (file) { 540 if (!trace_probe_get_file_link(tp, file)) 541 return -ENOENT; 542 if (!trace_probe_has_single_file(tp)) 543 goto out; 544 trace_probe_clear_flag(tp, TP_FLAG_TRACE); 545 } else 546 trace_probe_clear_flag(tp, TP_FLAG_PROFILE); 547 548 if (!trace_probe_is_enabled(tp)) 549 __disable_trace_fprobe(tp); 550 551 out: 552 if (file) 553 /* 554 * Synchronization is done in below function. For perf event, 555 * file == NULL and perf_trace_event_unreg() calls 556 * tracepoint_synchronize_unregister() to ensure synchronize 557 * event. We don't need to care about it. 558 */ 559 trace_probe_remove_file(tp, file); 560 561 return 0; 562 } 563 564 /* Event entry printers */ 565 static enum print_line_t 566 print_fentry_event(struct trace_iterator *iter, int flags, 567 struct trace_event *event) 568 { 569 struct fentry_trace_entry_head *field; 570 struct trace_seq *s = &iter->seq; 571 struct trace_probe *tp; 572 573 field = (struct fentry_trace_entry_head *)iter->ent; 574 tp = trace_probe_primary_from_call( 575 container_of(event, struct trace_event_call, event)); 576 if (WARN_ON_ONCE(!tp)) 577 goto out; 578 579 trace_seq_printf(s, "%s: (", trace_probe_name(tp)); 580 581 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET)) 582 goto out; 583 584 trace_seq_putc(s, ')'); 585 586 if (trace_probe_print_args(s, tp->args, tp->nr_args, 587 (u8 *)&field[1], field) < 0) 588 goto out; 589 590 trace_seq_putc(s, '\n'); 591 out: 592 return trace_handle_return(s); 593 } 594 595 static enum print_line_t 596 print_fexit_event(struct trace_iterator *iter, int flags, 597 struct trace_event *event) 598 { 599 struct fexit_trace_entry_head *field; 600 struct trace_seq *s = &iter->seq; 601 struct trace_probe *tp; 602 603 field = (struct fexit_trace_entry_head *)iter->ent; 604 tp = trace_probe_primary_from_call( 605 container_of(event, struct trace_event_call, event)); 606 if (WARN_ON_ONCE(!tp)) 607 goto out; 608 609 trace_seq_printf(s, "%s: (", trace_probe_name(tp)); 610 611 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET)) 612 goto out; 613 614 trace_seq_puts(s, " <- "); 615 616 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET)) 617 goto out; 618 619 trace_seq_putc(s, ')'); 620 621 if (trace_probe_print_args(s, tp->args, tp->nr_args, 622 (u8 *)&field[1], field) < 0) 623 goto out; 624 625 trace_seq_putc(s, '\n'); 626 627 out: 628 return trace_handle_return(s); 629 } 630 631 static int fentry_event_define_fields(struct trace_event_call *event_call) 632 { 633 int ret; 634 struct fentry_trace_entry_head field; 635 struct trace_probe *tp; 636 637 tp = trace_probe_primary_from_call(event_call); 638 if (WARN_ON_ONCE(!tp)) 639 return -ENOENT; 640 641 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0); 642 643 return traceprobe_define_arg_fields(event_call, sizeof(field), tp); 644 } 645 646 static int fexit_event_define_fields(struct trace_event_call *event_call) 647 { 648 int ret; 649 struct fexit_trace_entry_head field; 650 struct trace_probe *tp; 651 652 tp = trace_probe_primary_from_call(event_call); 653 if (WARN_ON_ONCE(!tp)) 654 return -ENOENT; 655 656 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0); 657 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0); 658 659 return traceprobe_define_arg_fields(event_call, sizeof(field), tp); 660 } 661 662 static struct trace_event_functions fentry_funcs = { 663 .trace = print_fentry_event 664 }; 665 666 static struct trace_event_functions fexit_funcs = { 667 .trace = print_fexit_event 668 }; 669 670 static struct trace_event_fields fentry_fields_array[] = { 671 { .type = TRACE_FUNCTION_TYPE, 672 .define_fields = fentry_event_define_fields }, 673 {} 674 }; 675 676 static struct trace_event_fields fexit_fields_array[] = { 677 { .type = TRACE_FUNCTION_TYPE, 678 .define_fields = fexit_event_define_fields }, 679 {} 680 }; 681 682 static int fprobe_register(struct trace_event_call *event, 683 enum trace_reg type, void *data); 684 685 static inline void init_trace_event_call(struct trace_fprobe *tf) 686 { 687 struct trace_event_call *call = trace_probe_event_call(&tf->tp); 688 689 if (trace_fprobe_is_return(tf)) { 690 call->event.funcs = &fexit_funcs; 691 call->class->fields_array = fexit_fields_array; 692 } else { 693 call->event.funcs = &fentry_funcs; 694 call->class->fields_array = fentry_fields_array; 695 } 696 697 call->flags = TRACE_EVENT_FL_FPROBE; 698 call->class->reg = fprobe_register; 699 } 700 701 static int register_fprobe_event(struct trace_fprobe *tf) 702 { 703 init_trace_event_call(tf); 704 705 return trace_probe_register_event_call(&tf->tp); 706 } 707 708 static int unregister_fprobe_event(struct trace_fprobe *tf) 709 { 710 return trace_probe_unregister_event_call(&tf->tp); 711 } 712 713 static int __regsiter_tracepoint_fprobe(struct trace_fprobe *tf) 714 { 715 struct tracepoint *tpoint = tf->tpoint; 716 unsigned long ip = (unsigned long)tpoint->probestub; 717 int ret; 718 719 /* 720 * Here, we do 2 steps to enable fprobe on a tracepoint. 721 * At first, put __probestub_##TP function on the tracepoint 722 * and put a fprobe on the stub function. 723 */ 724 ret = tracepoint_probe_register_prio_may_exist(tpoint, 725 tpoint->probestub, NULL, 0); 726 if (ret < 0) 727 return ret; 728 return register_fprobe_ips(&tf->fp, &ip, 1); 729 } 730 731 /* Internal register function - just handle fprobe and flags */ 732 static int __register_trace_fprobe(struct trace_fprobe *tf) 733 { 734 int i, ret; 735 736 /* Should we need new LOCKDOWN flag for fprobe? */ 737 ret = security_locked_down(LOCKDOWN_KPROBES); 738 if (ret) 739 return ret; 740 741 if (trace_fprobe_is_registered(tf)) 742 return -EINVAL; 743 744 for (i = 0; i < tf->tp.nr_args; i++) { 745 ret = traceprobe_update_arg(&tf->tp.args[i]); 746 if (ret) 747 return ret; 748 } 749 750 /* Set/clear disabled flag according to tp->flag */ 751 if (trace_probe_is_enabled(&tf->tp)) 752 tf->fp.flags &= ~FPROBE_FL_DISABLED; 753 else 754 tf->fp.flags |= FPROBE_FL_DISABLED; 755 756 if (trace_fprobe_is_tracepoint(tf)) { 757 758 /* This tracepoint is not loaded yet */ 759 if (tf->tpoint == TRACEPOINT_STUB) 760 return 0; 761 762 return __regsiter_tracepoint_fprobe(tf); 763 } 764 765 /* TODO: handle filter, nofilter or symbol list */ 766 return register_fprobe(&tf->fp, tf->symbol, NULL); 767 } 768 769 /* Internal unregister function - just handle fprobe and flags */ 770 static void __unregister_trace_fprobe(struct trace_fprobe *tf) 771 { 772 if (trace_fprobe_is_registered(tf)) { 773 unregister_fprobe(&tf->fp); 774 memset(&tf->fp, 0, sizeof(tf->fp)); 775 if (trace_fprobe_is_tracepoint(tf)) { 776 tracepoint_probe_unregister(tf->tpoint, 777 tf->tpoint->probestub, NULL); 778 tf->tpoint = NULL; 779 tf->mod = NULL; 780 } 781 } 782 } 783 784 /* TODO: make this trace_*probe common function */ 785 /* Unregister a trace_probe and probe_event */ 786 static int unregister_trace_fprobe(struct trace_fprobe *tf) 787 { 788 /* If other probes are on the event, just unregister fprobe */ 789 if (trace_probe_has_sibling(&tf->tp)) 790 goto unreg; 791 792 /* Enabled event can not be unregistered */ 793 if (trace_probe_is_enabled(&tf->tp)) 794 return -EBUSY; 795 796 /* If there's a reference to the dynamic event */ 797 if (trace_event_dyn_busy(trace_probe_event_call(&tf->tp))) 798 return -EBUSY; 799 800 /* Will fail if probe is being used by ftrace or perf */ 801 if (unregister_fprobe_event(tf)) 802 return -EBUSY; 803 804 unreg: 805 __unregister_trace_fprobe(tf); 806 dyn_event_remove(&tf->devent); 807 trace_probe_unlink(&tf->tp); 808 809 return 0; 810 } 811 812 static bool trace_fprobe_has_same_fprobe(struct trace_fprobe *orig, 813 struct trace_fprobe *comp) 814 { 815 struct trace_probe_event *tpe = orig->tp.event; 816 int i; 817 818 list_for_each_entry(orig, &tpe->probes, tp.list) { 819 if (strcmp(trace_fprobe_symbol(orig), 820 trace_fprobe_symbol(comp))) 821 continue; 822 823 /* 824 * trace_probe_compare_arg_type() ensured that nr_args and 825 * each argument name and type are same. Let's compare comm. 826 */ 827 for (i = 0; i < orig->tp.nr_args; i++) { 828 if (strcmp(orig->tp.args[i].comm, 829 comp->tp.args[i].comm)) 830 break; 831 } 832 833 if (i == orig->tp.nr_args) 834 return true; 835 } 836 837 return false; 838 } 839 840 static int append_trace_fprobe(struct trace_fprobe *tf, struct trace_fprobe *to) 841 { 842 int ret; 843 844 if (trace_fprobe_is_return(tf) != trace_fprobe_is_return(to) || 845 trace_fprobe_is_tracepoint(tf) != trace_fprobe_is_tracepoint(to)) { 846 trace_probe_log_set_index(0); 847 trace_probe_log_err(0, DIFF_PROBE_TYPE); 848 return -EEXIST; 849 } 850 ret = trace_probe_compare_arg_type(&tf->tp, &to->tp); 851 if (ret) { 852 /* Note that argument starts index = 2 */ 853 trace_probe_log_set_index(ret + 1); 854 trace_probe_log_err(0, DIFF_ARG_TYPE); 855 return -EEXIST; 856 } 857 if (trace_fprobe_has_same_fprobe(to, tf)) { 858 trace_probe_log_set_index(0); 859 trace_probe_log_err(0, SAME_PROBE); 860 return -EEXIST; 861 } 862 863 /* Append to existing event */ 864 ret = trace_probe_append(&tf->tp, &to->tp); 865 if (ret) 866 return ret; 867 868 ret = __register_trace_fprobe(tf); 869 if (ret) 870 trace_probe_unlink(&tf->tp); 871 else 872 dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp)); 873 874 return ret; 875 } 876 877 /* Register a trace_probe and probe_event */ 878 static int register_trace_fprobe(struct trace_fprobe *tf) 879 { 880 struct trace_fprobe *old_tf; 881 int ret; 882 883 guard(mutex)(&event_mutex); 884 885 old_tf = find_trace_fprobe(trace_probe_name(&tf->tp), 886 trace_probe_group_name(&tf->tp)); 887 if (old_tf) 888 return append_trace_fprobe(tf, old_tf); 889 890 /* Register new event */ 891 ret = register_fprobe_event(tf); 892 if (ret) { 893 if (ret == -EEXIST) { 894 trace_probe_log_set_index(0); 895 trace_probe_log_err(0, EVENT_EXIST); 896 } else 897 pr_warn("Failed to register probe event(%d)\n", ret); 898 return ret; 899 } 900 901 /* Register fprobe */ 902 ret = __register_trace_fprobe(tf); 903 if (ret < 0) 904 unregister_fprobe_event(tf); 905 else 906 dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp)); 907 908 return ret; 909 } 910 911 struct __find_tracepoint_cb_data { 912 const char *tp_name; 913 struct tracepoint *tpoint; 914 struct module *mod; 915 }; 916 917 static void __find_tracepoint_module_cb(struct tracepoint *tp, struct module *mod, void *priv) 918 { 919 struct __find_tracepoint_cb_data *data = priv; 920 921 if (!data->tpoint && !strcmp(data->tp_name, tp->name)) { 922 data->tpoint = tp; 923 if (!data->mod) { 924 data->mod = mod; 925 if (!try_module_get(data->mod)) { 926 data->tpoint = NULL; 927 data->mod = NULL; 928 } 929 } 930 } 931 } 932 933 static void __find_tracepoint_cb(struct tracepoint *tp, void *priv) 934 { 935 struct __find_tracepoint_cb_data *data = priv; 936 937 if (!data->tpoint && !strcmp(data->tp_name, tp->name)) 938 data->tpoint = tp; 939 } 940 941 /* 942 * Find a tracepoint from kernel and module. If the tracepoint is in a module, 943 * this increments the module refcount to prevent unloading until the 944 * trace_fprobe is registered to the list. After registering the trace_fprobe 945 * on the trace_fprobe list, the module refcount is decremented because 946 * tracepoint_probe_module_cb will handle it. 947 */ 948 static struct tracepoint *find_tracepoint(const char *tp_name, 949 struct module **tp_mod) 950 { 951 struct __find_tracepoint_cb_data data = { 952 .tp_name = tp_name, 953 .mod = NULL, 954 }; 955 956 for_each_kernel_tracepoint(__find_tracepoint_cb, &data); 957 958 if (!data.tpoint && IS_ENABLED(CONFIG_MODULES)) { 959 for_each_module_tracepoint(__find_tracepoint_module_cb, &data); 960 *tp_mod = data.mod; 961 } 962 963 return data.tpoint; 964 } 965 966 #ifdef CONFIG_MODULES 967 static void reenable_trace_fprobe(struct trace_fprobe *tf) 968 { 969 struct trace_probe *tp = &tf->tp; 970 971 list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) { 972 __enable_trace_fprobe(tf); 973 } 974 } 975 976 static struct tracepoint *find_tracepoint_in_module(struct module *mod, 977 const char *tp_name) 978 { 979 struct __find_tracepoint_cb_data data = { 980 .tp_name = tp_name, 981 .mod = mod, 982 }; 983 984 for_each_tracepoint_in_module(mod, __find_tracepoint_module_cb, &data); 985 return data.tpoint; 986 } 987 988 static int __tracepoint_probe_module_cb(struct notifier_block *self, 989 unsigned long val, void *data) 990 { 991 struct tp_module *tp_mod = data; 992 struct tracepoint *tpoint; 993 struct trace_fprobe *tf; 994 struct dyn_event *pos; 995 996 if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING) 997 return NOTIFY_DONE; 998 999 mutex_lock(&event_mutex); 1000 for_each_trace_fprobe(tf, pos) { 1001 if (val == MODULE_STATE_COMING && tf->tpoint == TRACEPOINT_STUB) { 1002 tpoint = find_tracepoint_in_module(tp_mod->mod, tf->symbol); 1003 if (tpoint) { 1004 tf->tpoint = tpoint; 1005 tf->mod = tp_mod->mod; 1006 if (!WARN_ON_ONCE(__regsiter_tracepoint_fprobe(tf)) && 1007 trace_probe_is_enabled(&tf->tp)) 1008 reenable_trace_fprobe(tf); 1009 } 1010 } else if (val == MODULE_STATE_GOING && tp_mod->mod == tf->mod) { 1011 tracepoint_probe_unregister(tf->tpoint, 1012 tf->tpoint->probestub, NULL); 1013 tf->tpoint = NULL; 1014 tf->mod = NULL; 1015 } 1016 } 1017 mutex_unlock(&event_mutex); 1018 1019 return NOTIFY_DONE; 1020 } 1021 1022 static struct notifier_block tracepoint_module_nb = { 1023 .notifier_call = __tracepoint_probe_module_cb, 1024 }; 1025 #endif /* CONFIG_MODULES */ 1026 1027 static int parse_symbol_and_return(int argc, const char *argv[], 1028 char **symbol, bool *is_return, 1029 bool is_tracepoint) 1030 { 1031 char *tmp = strchr(argv[1], '%'); 1032 int i; 1033 1034 if (tmp) { 1035 int len = tmp - argv[1]; 1036 1037 if (!is_tracepoint && !strcmp(tmp, "%return")) { 1038 *is_return = true; 1039 } else { 1040 trace_probe_log_err(len, BAD_ADDR_SUFFIX); 1041 return -EINVAL; 1042 } 1043 *symbol = kmemdup_nul(argv[1], len, GFP_KERNEL); 1044 } else 1045 *symbol = kstrdup(argv[1], GFP_KERNEL); 1046 if (!*symbol) 1047 return -ENOMEM; 1048 1049 if (*is_return) 1050 return 0; 1051 1052 if (is_tracepoint) { 1053 tmp = *symbol; 1054 while (*tmp && (isalnum(*tmp) || *tmp == '_')) 1055 tmp++; 1056 if (*tmp) { 1057 /* find a wrong character. */ 1058 trace_probe_log_err(tmp - *symbol, BAD_TP_NAME); 1059 kfree(*symbol); 1060 *symbol = NULL; 1061 return -EINVAL; 1062 } 1063 } 1064 1065 /* If there is $retval, this should be a return fprobe. */ 1066 for (i = 2; i < argc; i++) { 1067 tmp = strstr(argv[i], "$retval"); 1068 if (tmp && !isalnum(tmp[7]) && tmp[7] != '_') { 1069 if (is_tracepoint) { 1070 trace_probe_log_set_index(i); 1071 trace_probe_log_err(tmp - argv[i], RETVAL_ON_PROBE); 1072 kfree(*symbol); 1073 *symbol = NULL; 1074 return -EINVAL; 1075 } 1076 *is_return = true; 1077 break; 1078 } 1079 } 1080 return 0; 1081 } 1082 1083 DEFINE_FREE(module_put, struct module *, if (_T) module_put(_T)) 1084 1085 static int trace_fprobe_create_internal(int argc, const char *argv[], 1086 struct traceprobe_parse_context *ctx) 1087 { 1088 /* 1089 * Argument syntax: 1090 * - Add fentry probe: 1091 * f[:[GRP/][EVENT]] [MOD:]KSYM [FETCHARGS] 1092 * - Add fexit probe: 1093 * f[N][:[GRP/][EVENT]] [MOD:]KSYM%return [FETCHARGS] 1094 * - Add tracepoint probe: 1095 * t[:[GRP/][EVENT]] TRACEPOINT [FETCHARGS] 1096 * 1097 * Fetch args: 1098 * $retval : fetch return value 1099 * $stack : fetch stack address 1100 * $stackN : fetch Nth entry of stack (N:0-) 1101 * $argN : fetch Nth argument (N:1-) 1102 * $comm : fetch current task comm 1103 * @ADDR : fetch memory at ADDR (ADDR should be in kernel) 1104 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol) 1105 * Dereferencing memory fetch: 1106 * +|-offs(ARG) : fetch memory at ARG +|- offs address. 1107 * Alias name of args: 1108 * NAME=FETCHARG : set NAME as alias of FETCHARG. 1109 * Type of args: 1110 * FETCHARG:TYPE : use TYPE instead of unsigned long. 1111 */ 1112 struct trace_fprobe *tf __free(free_trace_fprobe) = NULL; 1113 int i, new_argc = 0, ret = 0; 1114 bool is_return = false; 1115 char *symbol __free(kfree) = NULL; 1116 const char *event = NULL, *group = FPROBE_EVENT_SYSTEM; 1117 const char **new_argv __free(kfree) = NULL; 1118 char buf[MAX_EVENT_NAME_LEN]; 1119 char gbuf[MAX_EVENT_NAME_LEN]; 1120 char sbuf[KSYM_NAME_LEN]; 1121 char abuf[MAX_BTF_ARGS_LEN]; 1122 char *dbuf __free(kfree) = NULL; 1123 bool is_tracepoint = false; 1124 struct module *tp_mod __free(module_put) = NULL; 1125 struct tracepoint *tpoint = NULL; 1126 1127 if ((argv[0][0] != 'f' && argv[0][0] != 't') || argc < 2) 1128 return -ECANCELED; 1129 1130 if (argv[0][0] == 't') { 1131 is_tracepoint = true; 1132 group = TRACEPOINT_EVENT_SYSTEM; 1133 } 1134 1135 if (argv[0][1] != '\0') { 1136 if (argv[0][1] != ':') { 1137 trace_probe_log_set_index(0); 1138 trace_probe_log_err(1, BAD_MAXACT); 1139 return -EINVAL; 1140 } 1141 event = &argv[0][2]; 1142 } 1143 1144 trace_probe_log_set_index(1); 1145 1146 /* a symbol(or tracepoint) must be specified */ 1147 ret = parse_symbol_and_return(argc, argv, &symbol, &is_return, is_tracepoint); 1148 if (ret < 0) 1149 return -EINVAL; 1150 1151 trace_probe_log_set_index(0); 1152 if (event) { 1153 ret = traceprobe_parse_event_name(&event, &group, gbuf, 1154 event - argv[0]); 1155 if (ret) 1156 return -EINVAL; 1157 } 1158 1159 if (!event) { 1160 /* Make a new event name */ 1161 if (is_tracepoint) 1162 snprintf(buf, MAX_EVENT_NAME_LEN, "%s%s", 1163 isdigit(*symbol) ? "_" : "", symbol); 1164 else 1165 snprintf(buf, MAX_EVENT_NAME_LEN, "%s__%s", symbol, 1166 is_return ? "exit" : "entry"); 1167 sanitize_event_name(buf); 1168 event = buf; 1169 } 1170 1171 if (is_return) 1172 ctx->flags |= TPARG_FL_RETURN; 1173 else 1174 ctx->flags |= TPARG_FL_FENTRY; 1175 1176 if (is_tracepoint) { 1177 ctx->flags |= TPARG_FL_TPOINT; 1178 tpoint = find_tracepoint(symbol, &tp_mod); 1179 if (tpoint) { 1180 ctx->funcname = kallsyms_lookup( 1181 (unsigned long)tpoint->probestub, 1182 NULL, NULL, NULL, sbuf); 1183 } else if (IS_ENABLED(CONFIG_MODULES)) { 1184 /* This *may* be loaded afterwards */ 1185 tpoint = TRACEPOINT_STUB; 1186 ctx->funcname = symbol; 1187 } else { 1188 trace_probe_log_set_index(1); 1189 trace_probe_log_err(0, NO_TRACEPOINT); 1190 return -EINVAL; 1191 } 1192 } else 1193 ctx->funcname = symbol; 1194 1195 argc -= 2; argv += 2; 1196 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc, 1197 abuf, MAX_BTF_ARGS_LEN, ctx); 1198 if (IS_ERR(new_argv)) 1199 return PTR_ERR(new_argv); 1200 if (new_argv) { 1201 argc = new_argc; 1202 argv = new_argv; 1203 } 1204 if (argc > MAX_TRACE_ARGS) 1205 return -E2BIG; 1206 1207 ret = traceprobe_expand_dentry_args(argc, argv, &dbuf); 1208 if (ret) 1209 return ret; 1210 1211 /* setup a probe */ 1212 tf = alloc_trace_fprobe(group, event, symbol, tpoint, tp_mod, 1213 argc, is_return); 1214 if (IS_ERR(tf)) { 1215 ret = PTR_ERR(tf); 1216 /* This must return -ENOMEM, else there is a bug */ 1217 WARN_ON_ONCE(ret != -ENOMEM); 1218 return ret; 1219 } 1220 1221 /* parse arguments */ 1222 for (i = 0; i < argc; i++) { 1223 trace_probe_log_set_index(i + 2); 1224 ctx->offset = 0; 1225 ret = traceprobe_parse_probe_arg(&tf->tp, i, argv[i], ctx); 1226 if (ret) 1227 return ret; /* This can be -ENOMEM */ 1228 } 1229 1230 if (is_return && tf->tp.entry_arg) { 1231 tf->fp.entry_handler = trace_fprobe_entry_handler; 1232 tf->fp.entry_data_size = traceprobe_get_entry_data_size(&tf->tp); 1233 if (ALIGN(tf->fp.entry_data_size, sizeof(long)) > MAX_FPROBE_DATA_SIZE) { 1234 trace_probe_log_set_index(2); 1235 trace_probe_log_err(0, TOO_MANY_EARGS); 1236 return -E2BIG; 1237 } 1238 } 1239 1240 ret = traceprobe_set_print_fmt(&tf->tp, 1241 is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL); 1242 if (ret < 0) 1243 return ret; 1244 1245 ret = register_trace_fprobe(tf); 1246 if (ret) { 1247 trace_probe_log_set_index(1); 1248 if (ret == -EILSEQ) 1249 trace_probe_log_err(0, BAD_INSN_BNDRY); 1250 else if (ret == -ENOENT) 1251 trace_probe_log_err(0, BAD_PROBE_ADDR); 1252 else if (ret != -ENOMEM && ret != -EEXIST) 1253 trace_probe_log_err(0, FAIL_REG_PROBE); 1254 return -EINVAL; 1255 } 1256 1257 /* 'tf' is successfully registered. To avoid freeing, assign NULL. */ 1258 tf = NULL; 1259 1260 return 0; 1261 } 1262 1263 static int trace_fprobe_create_cb(int argc, const char *argv[]) 1264 { 1265 struct traceprobe_parse_context ctx = { 1266 .flags = TPARG_FL_KERNEL | TPARG_FL_FPROBE, 1267 }; 1268 int ret; 1269 1270 trace_probe_log_init("trace_fprobe", argc, argv); 1271 ret = trace_fprobe_create_internal(argc, argv, &ctx); 1272 traceprobe_finish_parse(&ctx); 1273 trace_probe_log_clear(); 1274 return ret; 1275 } 1276 1277 static int trace_fprobe_create(const char *raw_command) 1278 { 1279 return trace_probe_create(raw_command, trace_fprobe_create_cb); 1280 } 1281 1282 static int trace_fprobe_release(struct dyn_event *ev) 1283 { 1284 struct trace_fprobe *tf = to_trace_fprobe(ev); 1285 int ret = unregister_trace_fprobe(tf); 1286 1287 if (!ret) 1288 free_trace_fprobe(tf); 1289 return ret; 1290 } 1291 1292 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev) 1293 { 1294 struct trace_fprobe *tf = to_trace_fprobe(ev); 1295 int i; 1296 1297 if (trace_fprobe_is_tracepoint(tf)) 1298 seq_putc(m, 't'); 1299 else 1300 seq_putc(m, 'f'); 1301 seq_printf(m, ":%s/%s", trace_probe_group_name(&tf->tp), 1302 trace_probe_name(&tf->tp)); 1303 1304 seq_printf(m, " %s%s", trace_fprobe_symbol(tf), 1305 trace_fprobe_is_return(tf) ? "%return" : ""); 1306 1307 for (i = 0; i < tf->tp.nr_args; i++) 1308 seq_printf(m, " %s=%s", tf->tp.args[i].name, tf->tp.args[i].comm); 1309 seq_putc(m, '\n'); 1310 1311 return 0; 1312 } 1313 1314 /* 1315 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex. 1316 */ 1317 static int fprobe_register(struct trace_event_call *event, 1318 enum trace_reg type, void *data) 1319 { 1320 struct trace_event_file *file = data; 1321 1322 switch (type) { 1323 case TRACE_REG_REGISTER: 1324 return enable_trace_fprobe(event, file); 1325 case TRACE_REG_UNREGISTER: 1326 return disable_trace_fprobe(event, file); 1327 1328 #ifdef CONFIG_PERF_EVENTS 1329 case TRACE_REG_PERF_REGISTER: 1330 return enable_trace_fprobe(event, NULL); 1331 case TRACE_REG_PERF_UNREGISTER: 1332 return disable_trace_fprobe(event, NULL); 1333 case TRACE_REG_PERF_OPEN: 1334 case TRACE_REG_PERF_CLOSE: 1335 case TRACE_REG_PERF_ADD: 1336 case TRACE_REG_PERF_DEL: 1337 return 0; 1338 #endif 1339 } 1340 return 0; 1341 } 1342 1343 /* 1344 * Register dynevent at core_initcall. This allows kernel to setup fprobe 1345 * events in postcore_initcall without tracefs. 1346 */ 1347 static __init int init_fprobe_trace_early(void) 1348 { 1349 int ret; 1350 1351 ret = dyn_event_register(&trace_fprobe_ops); 1352 if (ret) 1353 return ret; 1354 1355 #ifdef CONFIG_MODULES 1356 ret = register_tracepoint_module_notifier(&tracepoint_module_nb); 1357 if (ret) 1358 return ret; 1359 #endif 1360 1361 return 0; 1362 } 1363 core_initcall(init_fprobe_trace_early); 1364