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 /* 420 * Allocate new trace_probe and initialize it (including fprobe). 421 */ 422 static struct trace_fprobe *alloc_trace_fprobe(const char *group, 423 const char *event, 424 const char *symbol, 425 struct tracepoint *tpoint, 426 struct module *mod, 427 int maxactive, 428 int nargs, bool is_return) 429 { 430 struct trace_fprobe *tf; 431 int ret = -ENOMEM; 432 433 tf = kzalloc(struct_size(tf, tp.args, nargs), GFP_KERNEL); 434 if (!tf) 435 return ERR_PTR(ret); 436 437 tf->symbol = kstrdup(symbol, GFP_KERNEL); 438 if (!tf->symbol) 439 goto error; 440 441 if (is_return) 442 tf->fp.exit_handler = fexit_dispatcher; 443 else 444 tf->fp.entry_handler = fentry_dispatcher; 445 446 tf->tpoint = tpoint; 447 tf->mod = mod; 448 tf->fp.nr_maxactive = maxactive; 449 450 ret = trace_probe_init(&tf->tp, event, group, false, nargs); 451 if (ret < 0) 452 goto error; 453 454 dyn_event_init(&tf->devent, &trace_fprobe_ops); 455 return tf; 456 error: 457 free_trace_fprobe(tf); 458 return ERR_PTR(ret); 459 } 460 461 static struct trace_fprobe *find_trace_fprobe(const char *event, 462 const char *group) 463 { 464 struct dyn_event *pos; 465 struct trace_fprobe *tf; 466 467 for_each_trace_fprobe(tf, pos) 468 if (strcmp(trace_probe_name(&tf->tp), event) == 0 && 469 strcmp(trace_probe_group_name(&tf->tp), group) == 0) 470 return tf; 471 return NULL; 472 } 473 474 static inline int __enable_trace_fprobe(struct trace_fprobe *tf) 475 { 476 if (trace_fprobe_is_registered(tf)) 477 enable_fprobe(&tf->fp); 478 479 return 0; 480 } 481 482 static void __disable_trace_fprobe(struct trace_probe *tp) 483 { 484 struct trace_fprobe *tf; 485 486 list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) { 487 if (!trace_fprobe_is_registered(tf)) 488 continue; 489 disable_fprobe(&tf->fp); 490 } 491 } 492 493 /* 494 * Enable trace_probe 495 * if the file is NULL, enable "perf" handler, or enable "trace" handler. 496 */ 497 static int enable_trace_fprobe(struct trace_event_call *call, 498 struct trace_event_file *file) 499 { 500 struct trace_probe *tp; 501 struct trace_fprobe *tf; 502 bool enabled; 503 int ret = 0; 504 505 tp = trace_probe_primary_from_call(call); 506 if (WARN_ON_ONCE(!tp)) 507 return -ENODEV; 508 enabled = trace_probe_is_enabled(tp); 509 510 /* This also changes "enabled" state */ 511 if (file) { 512 ret = trace_probe_add_file(tp, file); 513 if (ret) 514 return ret; 515 } else 516 trace_probe_set_flag(tp, TP_FLAG_PROFILE); 517 518 if (!enabled) { 519 list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) { 520 /* TODO: check the fprobe is gone */ 521 __enable_trace_fprobe(tf); 522 } 523 } 524 525 return 0; 526 } 527 528 /* 529 * Disable trace_probe 530 * if the file is NULL, disable "perf" handler, or disable "trace" handler. 531 */ 532 static int disable_trace_fprobe(struct trace_event_call *call, 533 struct trace_event_file *file) 534 { 535 struct trace_probe *tp; 536 537 tp = trace_probe_primary_from_call(call); 538 if (WARN_ON_ONCE(!tp)) 539 return -ENODEV; 540 541 if (file) { 542 if (!trace_probe_get_file_link(tp, file)) 543 return -ENOENT; 544 if (!trace_probe_has_single_file(tp)) 545 goto out; 546 trace_probe_clear_flag(tp, TP_FLAG_TRACE); 547 } else 548 trace_probe_clear_flag(tp, TP_FLAG_PROFILE); 549 550 if (!trace_probe_is_enabled(tp)) 551 __disable_trace_fprobe(tp); 552 553 out: 554 if (file) 555 /* 556 * Synchronization is done in below function. For perf event, 557 * file == NULL and perf_trace_event_unreg() calls 558 * tracepoint_synchronize_unregister() to ensure synchronize 559 * event. We don't need to care about it. 560 */ 561 trace_probe_remove_file(tp, file); 562 563 return 0; 564 } 565 566 /* Event entry printers */ 567 static enum print_line_t 568 print_fentry_event(struct trace_iterator *iter, int flags, 569 struct trace_event *event) 570 { 571 struct fentry_trace_entry_head *field; 572 struct trace_seq *s = &iter->seq; 573 struct trace_probe *tp; 574 575 field = (struct fentry_trace_entry_head *)iter->ent; 576 tp = trace_probe_primary_from_call( 577 container_of(event, struct trace_event_call, event)); 578 if (WARN_ON_ONCE(!tp)) 579 goto out; 580 581 trace_seq_printf(s, "%s: (", trace_probe_name(tp)); 582 583 if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET)) 584 goto out; 585 586 trace_seq_putc(s, ')'); 587 588 if (trace_probe_print_args(s, tp->args, tp->nr_args, 589 (u8 *)&field[1], field) < 0) 590 goto out; 591 592 trace_seq_putc(s, '\n'); 593 out: 594 return trace_handle_return(s); 595 } 596 597 static enum print_line_t 598 print_fexit_event(struct trace_iterator *iter, int flags, 599 struct trace_event *event) 600 { 601 struct fexit_trace_entry_head *field; 602 struct trace_seq *s = &iter->seq; 603 struct trace_probe *tp; 604 605 field = (struct fexit_trace_entry_head *)iter->ent; 606 tp = trace_probe_primary_from_call( 607 container_of(event, struct trace_event_call, event)); 608 if (WARN_ON_ONCE(!tp)) 609 goto out; 610 611 trace_seq_printf(s, "%s: (", trace_probe_name(tp)); 612 613 if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET)) 614 goto out; 615 616 trace_seq_puts(s, " <- "); 617 618 if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET)) 619 goto out; 620 621 trace_seq_putc(s, ')'); 622 623 if (trace_probe_print_args(s, tp->args, tp->nr_args, 624 (u8 *)&field[1], field) < 0) 625 goto out; 626 627 trace_seq_putc(s, '\n'); 628 629 out: 630 return trace_handle_return(s); 631 } 632 633 static int fentry_event_define_fields(struct trace_event_call *event_call) 634 { 635 int ret; 636 struct fentry_trace_entry_head field; 637 struct trace_probe *tp; 638 639 tp = trace_probe_primary_from_call(event_call); 640 if (WARN_ON_ONCE(!tp)) 641 return -ENOENT; 642 643 DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0); 644 645 return traceprobe_define_arg_fields(event_call, sizeof(field), tp); 646 } 647 648 static int fexit_event_define_fields(struct trace_event_call *event_call) 649 { 650 int ret; 651 struct fexit_trace_entry_head field; 652 struct trace_probe *tp; 653 654 tp = trace_probe_primary_from_call(event_call); 655 if (WARN_ON_ONCE(!tp)) 656 return -ENOENT; 657 658 DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0); 659 DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0); 660 661 return traceprobe_define_arg_fields(event_call, sizeof(field), tp); 662 } 663 664 static struct trace_event_functions fentry_funcs = { 665 .trace = print_fentry_event 666 }; 667 668 static struct trace_event_functions fexit_funcs = { 669 .trace = print_fexit_event 670 }; 671 672 static struct trace_event_fields fentry_fields_array[] = { 673 { .type = TRACE_FUNCTION_TYPE, 674 .define_fields = fentry_event_define_fields }, 675 {} 676 }; 677 678 static struct trace_event_fields fexit_fields_array[] = { 679 { .type = TRACE_FUNCTION_TYPE, 680 .define_fields = fexit_event_define_fields }, 681 {} 682 }; 683 684 static int fprobe_register(struct trace_event_call *event, 685 enum trace_reg type, void *data); 686 687 static inline void init_trace_event_call(struct trace_fprobe *tf) 688 { 689 struct trace_event_call *call = trace_probe_event_call(&tf->tp); 690 691 if (trace_fprobe_is_return(tf)) { 692 call->event.funcs = &fexit_funcs; 693 call->class->fields_array = fexit_fields_array; 694 } else { 695 call->event.funcs = &fentry_funcs; 696 call->class->fields_array = fentry_fields_array; 697 } 698 699 call->flags = TRACE_EVENT_FL_FPROBE; 700 call->class->reg = fprobe_register; 701 } 702 703 static int register_fprobe_event(struct trace_fprobe *tf) 704 { 705 init_trace_event_call(tf); 706 707 return trace_probe_register_event_call(&tf->tp); 708 } 709 710 static int unregister_fprobe_event(struct trace_fprobe *tf) 711 { 712 return trace_probe_unregister_event_call(&tf->tp); 713 } 714 715 static int __regsiter_tracepoint_fprobe(struct trace_fprobe *tf) 716 { 717 struct tracepoint *tpoint = tf->tpoint; 718 unsigned long ip = (unsigned long)tpoint->probestub; 719 int ret; 720 721 /* 722 * Here, we do 2 steps to enable fprobe on a tracepoint. 723 * At first, put __probestub_##TP function on the tracepoint 724 * and put a fprobe on the stub function. 725 */ 726 ret = tracepoint_probe_register_prio_may_exist(tpoint, 727 tpoint->probestub, NULL, 0); 728 if (ret < 0) 729 return ret; 730 return register_fprobe_ips(&tf->fp, &ip, 1); 731 } 732 733 /* Internal register function - just handle fprobe and flags */ 734 static int __register_trace_fprobe(struct trace_fprobe *tf) 735 { 736 int i, ret; 737 738 /* Should we need new LOCKDOWN flag for fprobe? */ 739 ret = security_locked_down(LOCKDOWN_KPROBES); 740 if (ret) 741 return ret; 742 743 if (trace_fprobe_is_registered(tf)) 744 return -EINVAL; 745 746 for (i = 0; i < tf->tp.nr_args; i++) { 747 ret = traceprobe_update_arg(&tf->tp.args[i]); 748 if (ret) 749 return ret; 750 } 751 752 /* Set/clear disabled flag according to tp->flag */ 753 if (trace_probe_is_enabled(&tf->tp)) 754 tf->fp.flags &= ~FPROBE_FL_DISABLED; 755 else 756 tf->fp.flags |= FPROBE_FL_DISABLED; 757 758 if (trace_fprobe_is_tracepoint(tf)) { 759 760 /* This tracepoint is not loaded yet */ 761 if (tf->tpoint == TRACEPOINT_STUB) 762 return 0; 763 764 return __regsiter_tracepoint_fprobe(tf); 765 } 766 767 /* TODO: handle filter, nofilter or symbol list */ 768 return register_fprobe(&tf->fp, tf->symbol, NULL); 769 } 770 771 /* Internal unregister function - just handle fprobe and flags */ 772 static void __unregister_trace_fprobe(struct trace_fprobe *tf) 773 { 774 if (trace_fprobe_is_registered(tf)) { 775 unregister_fprobe(&tf->fp); 776 memset(&tf->fp, 0, sizeof(tf->fp)); 777 if (trace_fprobe_is_tracepoint(tf)) { 778 tracepoint_probe_unregister(tf->tpoint, 779 tf->tpoint->probestub, NULL); 780 tf->tpoint = NULL; 781 tf->mod = NULL; 782 } 783 } 784 } 785 786 /* TODO: make this trace_*probe common function */ 787 /* Unregister a trace_probe and probe_event */ 788 static int unregister_trace_fprobe(struct trace_fprobe *tf) 789 { 790 /* If other probes are on the event, just unregister fprobe */ 791 if (trace_probe_has_sibling(&tf->tp)) 792 goto unreg; 793 794 /* Enabled event can not be unregistered */ 795 if (trace_probe_is_enabled(&tf->tp)) 796 return -EBUSY; 797 798 /* If there's a reference to the dynamic event */ 799 if (trace_event_dyn_busy(trace_probe_event_call(&tf->tp))) 800 return -EBUSY; 801 802 /* Will fail if probe is being used by ftrace or perf */ 803 if (unregister_fprobe_event(tf)) 804 return -EBUSY; 805 806 unreg: 807 __unregister_trace_fprobe(tf); 808 dyn_event_remove(&tf->devent); 809 trace_probe_unlink(&tf->tp); 810 811 return 0; 812 } 813 814 static bool trace_fprobe_has_same_fprobe(struct trace_fprobe *orig, 815 struct trace_fprobe *comp) 816 { 817 struct trace_probe_event *tpe = orig->tp.event; 818 int i; 819 820 list_for_each_entry(orig, &tpe->probes, tp.list) { 821 if (strcmp(trace_fprobe_symbol(orig), 822 trace_fprobe_symbol(comp))) 823 continue; 824 825 /* 826 * trace_probe_compare_arg_type() ensured that nr_args and 827 * each argument name and type are same. Let's compare comm. 828 */ 829 for (i = 0; i < orig->tp.nr_args; i++) { 830 if (strcmp(orig->tp.args[i].comm, 831 comp->tp.args[i].comm)) 832 break; 833 } 834 835 if (i == orig->tp.nr_args) 836 return true; 837 } 838 839 return false; 840 } 841 842 static int append_trace_fprobe(struct trace_fprobe *tf, struct trace_fprobe *to) 843 { 844 int ret; 845 846 if (trace_fprobe_is_return(tf) != trace_fprobe_is_return(to) || 847 trace_fprobe_is_tracepoint(tf) != trace_fprobe_is_tracepoint(to)) { 848 trace_probe_log_set_index(0); 849 trace_probe_log_err(0, DIFF_PROBE_TYPE); 850 return -EEXIST; 851 } 852 ret = trace_probe_compare_arg_type(&tf->tp, &to->tp); 853 if (ret) { 854 /* Note that argument starts index = 2 */ 855 trace_probe_log_set_index(ret + 1); 856 trace_probe_log_err(0, DIFF_ARG_TYPE); 857 return -EEXIST; 858 } 859 if (trace_fprobe_has_same_fprobe(to, tf)) { 860 trace_probe_log_set_index(0); 861 trace_probe_log_err(0, SAME_PROBE); 862 return -EEXIST; 863 } 864 865 /* Append to existing event */ 866 ret = trace_probe_append(&tf->tp, &to->tp); 867 if (ret) 868 return ret; 869 870 ret = __register_trace_fprobe(tf); 871 if (ret) 872 trace_probe_unlink(&tf->tp); 873 else 874 dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp)); 875 876 return ret; 877 } 878 879 /* Register a trace_probe and probe_event */ 880 static int register_trace_fprobe(struct trace_fprobe *tf) 881 { 882 struct trace_fprobe *old_tf; 883 int ret; 884 885 mutex_lock(&event_mutex); 886 887 old_tf = find_trace_fprobe(trace_probe_name(&tf->tp), 888 trace_probe_group_name(&tf->tp)); 889 if (old_tf) { 890 ret = append_trace_fprobe(tf, old_tf); 891 goto end; 892 } 893 894 /* Register new event */ 895 ret = register_fprobe_event(tf); 896 if (ret) { 897 if (ret == -EEXIST) { 898 trace_probe_log_set_index(0); 899 trace_probe_log_err(0, EVENT_EXIST); 900 } else 901 pr_warn("Failed to register probe event(%d)\n", ret); 902 goto end; 903 } 904 905 /* Register fprobe */ 906 ret = __register_trace_fprobe(tf); 907 if (ret < 0) 908 unregister_fprobe_event(tf); 909 else 910 dyn_event_add(&tf->devent, trace_probe_event_call(&tf->tp)); 911 912 end: 913 mutex_unlock(&event_mutex); 914 return ret; 915 } 916 917 struct __find_tracepoint_cb_data { 918 const char *tp_name; 919 struct tracepoint *tpoint; 920 struct module *mod; 921 }; 922 923 static void __find_tracepoint_module_cb(struct tracepoint *tp, struct module *mod, void *priv) 924 { 925 struct __find_tracepoint_cb_data *data = priv; 926 927 if (!data->tpoint && !strcmp(data->tp_name, tp->name)) { 928 data->tpoint = tp; 929 if (!data->mod) { 930 data->mod = mod; 931 if (!try_module_get(data->mod)) { 932 data->tpoint = NULL; 933 data->mod = NULL; 934 } 935 } 936 } 937 } 938 939 static void __find_tracepoint_cb(struct tracepoint *tp, void *priv) 940 { 941 struct __find_tracepoint_cb_data *data = priv; 942 943 if (!data->tpoint && !strcmp(data->tp_name, tp->name)) 944 data->tpoint = tp; 945 } 946 947 /* 948 * Find a tracepoint from kernel and module. If the tracepoint is in a module, 949 * this increments the module refcount to prevent unloading until the 950 * trace_fprobe is registered to the list. After registering the trace_fprobe 951 * on the trace_fprobe list, the module refcount is decremented because 952 * tracepoint_probe_module_cb will handle it. 953 */ 954 static struct tracepoint *find_tracepoint(const char *tp_name, 955 struct module **tp_mod) 956 { 957 struct __find_tracepoint_cb_data data = { 958 .tp_name = tp_name, 959 .mod = NULL, 960 }; 961 962 for_each_kernel_tracepoint(__find_tracepoint_cb, &data); 963 964 if (!data.tpoint && IS_ENABLED(CONFIG_MODULES)) { 965 for_each_module_tracepoint(__find_tracepoint_module_cb, &data); 966 *tp_mod = data.mod; 967 } 968 969 return data.tpoint; 970 } 971 972 #ifdef CONFIG_MODULES 973 static void reenable_trace_fprobe(struct trace_fprobe *tf) 974 { 975 struct trace_probe *tp = &tf->tp; 976 977 list_for_each_entry(tf, trace_probe_probe_list(tp), tp.list) { 978 __enable_trace_fprobe(tf); 979 } 980 } 981 982 static struct tracepoint *find_tracepoint_in_module(struct module *mod, 983 const char *tp_name) 984 { 985 struct __find_tracepoint_cb_data data = { 986 .tp_name = tp_name, 987 .mod = mod, 988 }; 989 990 for_each_tracepoint_in_module(mod, __find_tracepoint_module_cb, &data); 991 return data.tpoint; 992 } 993 994 static int __tracepoint_probe_module_cb(struct notifier_block *self, 995 unsigned long val, void *data) 996 { 997 struct tp_module *tp_mod = data; 998 struct tracepoint *tpoint; 999 struct trace_fprobe *tf; 1000 struct dyn_event *pos; 1001 1002 if (val != MODULE_STATE_GOING && val != MODULE_STATE_COMING) 1003 return NOTIFY_DONE; 1004 1005 mutex_lock(&event_mutex); 1006 for_each_trace_fprobe(tf, pos) { 1007 if (val == MODULE_STATE_COMING && tf->tpoint == TRACEPOINT_STUB) { 1008 tpoint = find_tracepoint_in_module(tp_mod->mod, tf->symbol); 1009 if (tpoint) { 1010 tf->tpoint = tpoint; 1011 tf->mod = tp_mod->mod; 1012 if (!WARN_ON_ONCE(__regsiter_tracepoint_fprobe(tf)) && 1013 trace_probe_is_enabled(&tf->tp)) 1014 reenable_trace_fprobe(tf); 1015 } 1016 } else if (val == MODULE_STATE_GOING && tp_mod->mod == tf->mod) { 1017 tracepoint_probe_unregister(tf->tpoint, 1018 tf->tpoint->probestub, NULL); 1019 tf->tpoint = NULL; 1020 tf->mod = NULL; 1021 } 1022 } 1023 mutex_unlock(&event_mutex); 1024 1025 return NOTIFY_DONE; 1026 } 1027 1028 static struct notifier_block tracepoint_module_nb = { 1029 .notifier_call = __tracepoint_probe_module_cb, 1030 }; 1031 #endif /* CONFIG_MODULES */ 1032 1033 static int parse_symbol_and_return(int argc, const char *argv[], 1034 char **symbol, bool *is_return, 1035 bool is_tracepoint) 1036 { 1037 char *tmp = strchr(argv[1], '%'); 1038 int i; 1039 1040 if (tmp) { 1041 int len = tmp - argv[1]; 1042 1043 if (!is_tracepoint && !strcmp(tmp, "%return")) { 1044 *is_return = true; 1045 } else { 1046 trace_probe_log_err(len, BAD_ADDR_SUFFIX); 1047 return -EINVAL; 1048 } 1049 *symbol = kmemdup_nul(argv[1], len, GFP_KERNEL); 1050 } else 1051 *symbol = kstrdup(argv[1], GFP_KERNEL); 1052 if (!*symbol) 1053 return -ENOMEM; 1054 1055 if (*is_return) 1056 return 0; 1057 1058 /* If there is $retval, this should be a return fprobe. */ 1059 for (i = 2; i < argc; i++) { 1060 tmp = strstr(argv[i], "$retval"); 1061 if (tmp && !isalnum(tmp[7]) && tmp[7] != '_') { 1062 if (is_tracepoint) { 1063 trace_probe_log_set_index(i); 1064 trace_probe_log_err(tmp - argv[i], RETVAL_ON_PROBE); 1065 return -EINVAL; 1066 } 1067 *is_return = true; 1068 break; 1069 } 1070 } 1071 return 0; 1072 } 1073 1074 static int __trace_fprobe_create(int argc, const char *argv[]) 1075 { 1076 /* 1077 * Argument syntax: 1078 * - Add fentry probe: 1079 * f[:[GRP/][EVENT]] [MOD:]KSYM [FETCHARGS] 1080 * - Add fexit probe: 1081 * f[N][:[GRP/][EVENT]] [MOD:]KSYM%return [FETCHARGS] 1082 * - Add tracepoint probe: 1083 * t[:[GRP/][EVENT]] TRACEPOINT [FETCHARGS] 1084 * 1085 * Fetch args: 1086 * $retval : fetch return value 1087 * $stack : fetch stack address 1088 * $stackN : fetch Nth entry of stack (N:0-) 1089 * $argN : fetch Nth argument (N:1-) 1090 * $comm : fetch current task comm 1091 * @ADDR : fetch memory at ADDR (ADDR should be in kernel) 1092 * @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol) 1093 * Dereferencing memory fetch: 1094 * +|-offs(ARG) : fetch memory at ARG +|- offs address. 1095 * Alias name of args: 1096 * NAME=FETCHARG : set NAME as alias of FETCHARG. 1097 * Type of args: 1098 * FETCHARG:TYPE : use TYPE instead of unsigned long. 1099 */ 1100 struct trace_fprobe *tf = NULL; 1101 int i, len, new_argc = 0, ret = 0; 1102 bool is_return = false; 1103 char *symbol = NULL; 1104 const char *event = NULL, *group = FPROBE_EVENT_SYSTEM; 1105 const char **new_argv = NULL; 1106 int maxactive = 0; 1107 char buf[MAX_EVENT_NAME_LEN]; 1108 char gbuf[MAX_EVENT_NAME_LEN]; 1109 char sbuf[KSYM_NAME_LEN]; 1110 char abuf[MAX_BTF_ARGS_LEN]; 1111 char *dbuf = NULL; 1112 bool is_tracepoint = false; 1113 struct module *tp_mod = NULL; 1114 struct tracepoint *tpoint = NULL; 1115 struct traceprobe_parse_context ctx = { 1116 .flags = TPARG_FL_KERNEL | TPARG_FL_FPROBE, 1117 }; 1118 1119 if ((argv[0][0] != 'f' && argv[0][0] != 't') || argc < 2) 1120 return -ECANCELED; 1121 1122 if (argv[0][0] == 't') { 1123 is_tracepoint = true; 1124 group = TRACEPOINT_EVENT_SYSTEM; 1125 } 1126 1127 trace_probe_log_init("trace_fprobe", argc, argv); 1128 1129 event = strchr(&argv[0][1], ':'); 1130 if (event) 1131 event++; 1132 1133 if (isdigit(argv[0][1])) { 1134 if (event) 1135 len = event - &argv[0][1] - 1; 1136 else 1137 len = strlen(&argv[0][1]); 1138 if (len > MAX_EVENT_NAME_LEN - 1) { 1139 trace_probe_log_err(1, BAD_MAXACT); 1140 goto parse_error; 1141 } 1142 memcpy(buf, &argv[0][1], len); 1143 buf[len] = '\0'; 1144 ret = kstrtouint(buf, 0, &maxactive); 1145 if (ret || !maxactive) { 1146 trace_probe_log_err(1, BAD_MAXACT); 1147 goto parse_error; 1148 } 1149 /* fprobe rethook instances are iterated over via a list. The 1150 * maximum should stay reasonable. 1151 */ 1152 if (maxactive > RETHOOK_MAXACTIVE_MAX) { 1153 trace_probe_log_err(1, MAXACT_TOO_BIG); 1154 goto parse_error; 1155 } 1156 } 1157 1158 trace_probe_log_set_index(1); 1159 1160 /* a symbol(or tracepoint) must be specified */ 1161 ret = parse_symbol_and_return(argc, argv, &symbol, &is_return, is_tracepoint); 1162 if (ret < 0) 1163 goto parse_error; 1164 1165 if (!is_return && maxactive) { 1166 trace_probe_log_set_index(0); 1167 trace_probe_log_err(1, BAD_MAXACT_TYPE); 1168 goto parse_error; 1169 } 1170 1171 trace_probe_log_set_index(0); 1172 if (event) { 1173 ret = traceprobe_parse_event_name(&event, &group, gbuf, 1174 event - argv[0]); 1175 if (ret) 1176 goto parse_error; 1177 } 1178 1179 if (!event) { 1180 /* Make a new event name */ 1181 if (is_tracepoint) 1182 snprintf(buf, MAX_EVENT_NAME_LEN, "%s%s", 1183 isdigit(*symbol) ? "_" : "", symbol); 1184 else 1185 snprintf(buf, MAX_EVENT_NAME_LEN, "%s__%s", symbol, 1186 is_return ? "exit" : "entry"); 1187 sanitize_event_name(buf); 1188 event = buf; 1189 } 1190 1191 if (is_return) 1192 ctx.flags |= TPARG_FL_RETURN; 1193 else 1194 ctx.flags |= TPARG_FL_FENTRY; 1195 1196 if (is_tracepoint) { 1197 ctx.flags |= TPARG_FL_TPOINT; 1198 tpoint = find_tracepoint(symbol, &tp_mod); 1199 if (tpoint) { 1200 ctx.funcname = kallsyms_lookup( 1201 (unsigned long)tpoint->probestub, 1202 NULL, NULL, NULL, sbuf); 1203 } else if (IS_ENABLED(CONFIG_MODULES)) { 1204 /* This *may* be loaded afterwards */ 1205 tpoint = TRACEPOINT_STUB; 1206 ctx.funcname = symbol; 1207 } else { 1208 trace_probe_log_set_index(1); 1209 trace_probe_log_err(0, NO_TRACEPOINT); 1210 goto parse_error; 1211 } 1212 } else 1213 ctx.funcname = symbol; 1214 1215 argc -= 2; argv += 2; 1216 new_argv = traceprobe_expand_meta_args(argc, argv, &new_argc, 1217 abuf, MAX_BTF_ARGS_LEN, &ctx); 1218 if (IS_ERR(new_argv)) { 1219 ret = PTR_ERR(new_argv); 1220 new_argv = NULL; 1221 goto out; 1222 } 1223 if (new_argv) { 1224 argc = new_argc; 1225 argv = new_argv; 1226 } 1227 if (argc > MAX_TRACE_ARGS) { 1228 ret = -E2BIG; 1229 goto out; 1230 } 1231 1232 ret = traceprobe_expand_dentry_args(argc, argv, &dbuf); 1233 if (ret) 1234 goto out; 1235 1236 /* setup a probe */ 1237 tf = alloc_trace_fprobe(group, event, symbol, tpoint, tp_mod, 1238 maxactive, argc, is_return); 1239 if (IS_ERR(tf)) { 1240 ret = PTR_ERR(tf); 1241 /* This must return -ENOMEM, else there is a bug */ 1242 WARN_ON_ONCE(ret != -ENOMEM); 1243 goto out; /* We know tf is not allocated */ 1244 } 1245 1246 /* parse arguments */ 1247 for (i = 0; i < argc; i++) { 1248 trace_probe_log_set_index(i + 2); 1249 ctx.offset = 0; 1250 ret = traceprobe_parse_probe_arg(&tf->tp, i, argv[i], &ctx); 1251 if (ret) 1252 goto error; /* This can be -ENOMEM */ 1253 } 1254 1255 if (is_return && tf->tp.entry_arg) { 1256 tf->fp.entry_handler = trace_fprobe_entry_handler; 1257 tf->fp.entry_data_size = traceprobe_get_entry_data_size(&tf->tp); 1258 } 1259 1260 ret = traceprobe_set_print_fmt(&tf->tp, 1261 is_return ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL); 1262 if (ret < 0) 1263 goto error; 1264 1265 ret = register_trace_fprobe(tf); 1266 if (ret) { 1267 trace_probe_log_set_index(1); 1268 if (ret == -EILSEQ) 1269 trace_probe_log_err(0, BAD_INSN_BNDRY); 1270 else if (ret == -ENOENT) 1271 trace_probe_log_err(0, BAD_PROBE_ADDR); 1272 else if (ret != -ENOMEM && ret != -EEXIST) 1273 trace_probe_log_err(0, FAIL_REG_PROBE); 1274 goto error; 1275 } 1276 1277 out: 1278 if (tp_mod) 1279 module_put(tp_mod); 1280 traceprobe_finish_parse(&ctx); 1281 trace_probe_log_clear(); 1282 kfree(new_argv); 1283 kfree(symbol); 1284 kfree(dbuf); 1285 return ret; 1286 1287 parse_error: 1288 ret = -EINVAL; 1289 error: 1290 free_trace_fprobe(tf); 1291 goto out; 1292 } 1293 1294 static int trace_fprobe_create(const char *raw_command) 1295 { 1296 return trace_probe_create(raw_command, __trace_fprobe_create); 1297 } 1298 1299 static int trace_fprobe_release(struct dyn_event *ev) 1300 { 1301 struct trace_fprobe *tf = to_trace_fprobe(ev); 1302 int ret = unregister_trace_fprobe(tf); 1303 1304 if (!ret) 1305 free_trace_fprobe(tf); 1306 return ret; 1307 } 1308 1309 static int trace_fprobe_show(struct seq_file *m, struct dyn_event *ev) 1310 { 1311 struct trace_fprobe *tf = to_trace_fprobe(ev); 1312 int i; 1313 1314 if (trace_fprobe_is_tracepoint(tf)) 1315 seq_putc(m, 't'); 1316 else 1317 seq_putc(m, 'f'); 1318 if (trace_fprobe_is_return(tf) && tf->fp.nr_maxactive) 1319 seq_printf(m, "%d", tf->fp.nr_maxactive); 1320 seq_printf(m, ":%s/%s", trace_probe_group_name(&tf->tp), 1321 trace_probe_name(&tf->tp)); 1322 1323 seq_printf(m, " %s%s", trace_fprobe_symbol(tf), 1324 trace_fprobe_is_return(tf) ? "%return" : ""); 1325 1326 for (i = 0; i < tf->tp.nr_args; i++) 1327 seq_printf(m, " %s=%s", tf->tp.args[i].name, tf->tp.args[i].comm); 1328 seq_putc(m, '\n'); 1329 1330 return 0; 1331 } 1332 1333 /* 1334 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex. 1335 */ 1336 static int fprobe_register(struct trace_event_call *event, 1337 enum trace_reg type, void *data) 1338 { 1339 struct trace_event_file *file = data; 1340 1341 switch (type) { 1342 case TRACE_REG_REGISTER: 1343 return enable_trace_fprobe(event, file); 1344 case TRACE_REG_UNREGISTER: 1345 return disable_trace_fprobe(event, file); 1346 1347 #ifdef CONFIG_PERF_EVENTS 1348 case TRACE_REG_PERF_REGISTER: 1349 return enable_trace_fprobe(event, NULL); 1350 case TRACE_REG_PERF_UNREGISTER: 1351 return disable_trace_fprobe(event, NULL); 1352 case TRACE_REG_PERF_OPEN: 1353 case TRACE_REG_PERF_CLOSE: 1354 case TRACE_REG_PERF_ADD: 1355 case TRACE_REG_PERF_DEL: 1356 return 0; 1357 #endif 1358 } 1359 return 0; 1360 } 1361 1362 /* 1363 * Register dynevent at core_initcall. This allows kernel to setup fprobe 1364 * events in postcore_initcall without tracefs. 1365 */ 1366 static __init int init_fprobe_trace_early(void) 1367 { 1368 int ret; 1369 1370 ret = dyn_event_register(&trace_fprobe_ops); 1371 if (ret) 1372 return ret; 1373 1374 #ifdef CONFIG_MODULES 1375 ret = register_tracepoint_module_notifier(&tracepoint_module_nb); 1376 if (ret) 1377 return ret; 1378 #endif 1379 1380 return 0; 1381 } 1382 core_initcall(init_fprobe_trace_early); 1383