1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * uprobes-based tracing events 4 * 5 * Copyright (C) IBM Corporation, 2010-2012 6 * Author: Srikar Dronamraju <srikar@linux.vnet.ibm.com> 7 */ 8 #define pr_fmt(fmt) "trace_uprobe: " fmt 9 10 #include <linux/bpf-cgroup.h> 11 #include <linux/security.h> 12 #include <linux/ctype.h> 13 #include <linux/module.h> 14 #include <linux/uaccess.h> 15 #include <linux/uprobes.h> 16 #include <linux/namei.h> 17 #include <linux/string.h> 18 #include <linux/rculist.h> 19 #include <linux/filter.h> 20 21 #include "trace_dynevent.h" 22 #include "trace_probe.h" 23 #include "trace_probe_tmpl.h" 24 25 #define UPROBE_EVENT_SYSTEM "uprobes" 26 27 struct uprobe_trace_entry_head { 28 struct trace_entry ent; 29 unsigned long vaddr[]; 30 }; 31 32 #define SIZEOF_TRACE_ENTRY(is_return) \ 33 (sizeof(struct uprobe_trace_entry_head) + \ 34 sizeof(unsigned long) * (is_return ? 2 : 1)) 35 36 #define DATAOF_TRACE_ENTRY(entry, is_return) \ 37 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return)) 38 39 static int trace_uprobe_create(const char *raw_command); 40 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev); 41 static int trace_uprobe_release(struct dyn_event *ev); 42 static bool trace_uprobe_is_busy(struct dyn_event *ev); 43 static bool trace_uprobe_match(const char *system, const char *event, 44 int argc, const char **argv, struct dyn_event *ev); 45 46 static struct dyn_event_operations trace_uprobe_ops = { 47 .create = trace_uprobe_create, 48 .show = trace_uprobe_show, 49 .is_busy = trace_uprobe_is_busy, 50 .free = trace_uprobe_release, 51 .match = trace_uprobe_match, 52 }; 53 54 /* 55 * uprobe event core functions 56 */ 57 struct trace_uprobe { 58 struct dyn_event devent; 59 struct uprobe_consumer consumer; 60 struct path path; 61 struct inode *inode; 62 char *filename; 63 unsigned long offset; 64 unsigned long ref_ctr_offset; 65 unsigned long nhit; 66 struct trace_probe tp; 67 }; 68 69 static bool is_trace_uprobe(struct dyn_event *ev) 70 { 71 return ev->ops == &trace_uprobe_ops; 72 } 73 74 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev) 75 { 76 return container_of(ev, struct trace_uprobe, devent); 77 } 78 79 /** 80 * for_each_trace_uprobe - iterate over the trace_uprobe list 81 * @pos: the struct trace_uprobe * for each entry 82 * @dpos: the struct dyn_event * to use as a loop cursor 83 */ 84 #define for_each_trace_uprobe(pos, dpos) \ 85 for_each_dyn_event(dpos) \ 86 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos))) 87 88 static int register_uprobe_event(struct trace_uprobe *tu); 89 static int unregister_uprobe_event(struct trace_uprobe *tu); 90 91 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs); 92 static int uretprobe_dispatcher(struct uprobe_consumer *con, 93 unsigned long func, struct pt_regs *regs); 94 95 #ifdef CONFIG_STACK_GROWSUP 96 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n) 97 { 98 return addr - (n * sizeof(long)); 99 } 100 #else 101 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n) 102 { 103 return addr + (n * sizeof(long)); 104 } 105 #endif 106 107 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n) 108 { 109 unsigned long ret; 110 unsigned long addr = user_stack_pointer(regs); 111 112 addr = adjust_stack_addr(addr, n); 113 114 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret))) 115 return 0; 116 117 return ret; 118 } 119 120 /* 121 * Uprobes-specific fetch functions 122 */ 123 static nokprobe_inline int 124 probe_mem_read(void *dest, void *src, size_t size) 125 { 126 void __user *vaddr = (void __force __user *)src; 127 128 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0; 129 } 130 131 static nokprobe_inline int 132 probe_mem_read_user(void *dest, void *src, size_t size) 133 { 134 return probe_mem_read(dest, src, size); 135 } 136 137 /* 138 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max 139 * length and relative data location. 140 */ 141 static nokprobe_inline int 142 fetch_store_string(unsigned long addr, void *dest, void *base) 143 { 144 long ret; 145 u32 loc = *(u32 *)dest; 146 int maxlen = get_loc_len(loc); 147 u8 *dst = get_loc_data(dest, base); 148 void __user *src = (void __force __user *) addr; 149 150 if (unlikely(!maxlen)) 151 return -ENOMEM; 152 153 if (addr == FETCH_TOKEN_COMM) 154 ret = strscpy(dst, current->comm, maxlen); 155 else 156 ret = strncpy_from_user(dst, src, maxlen); 157 if (ret >= 0) { 158 if (ret == maxlen) 159 dst[ret - 1] = '\0'; 160 else 161 /* 162 * Include the terminating null byte. In this case it 163 * was copied by strncpy_from_user but not accounted 164 * for in ret. 165 */ 166 ret++; 167 *(u32 *)dest = make_data_loc(ret, (void *)dst - base); 168 } else 169 *(u32 *)dest = make_data_loc(0, (void *)dst - base); 170 171 return ret; 172 } 173 174 static nokprobe_inline int 175 fetch_store_string_user(unsigned long addr, void *dest, void *base) 176 { 177 return fetch_store_string(addr, dest, base); 178 } 179 180 /* Return the length of string -- including null terminal byte */ 181 static nokprobe_inline int 182 fetch_store_strlen(unsigned long addr) 183 { 184 int len; 185 void __user *vaddr = (void __force __user *) addr; 186 187 if (addr == FETCH_TOKEN_COMM) 188 len = strlen(current->comm) + 1; 189 else 190 len = strnlen_user(vaddr, MAX_STRING_SIZE); 191 192 return (len > MAX_STRING_SIZE) ? 0 : len; 193 } 194 195 static nokprobe_inline int 196 fetch_store_strlen_user(unsigned long addr) 197 { 198 return fetch_store_strlen(addr); 199 } 200 201 static unsigned long translate_user_vaddr(unsigned long file_offset) 202 { 203 unsigned long base_addr; 204 struct uprobe_dispatch_data *udd; 205 206 udd = (void *) current->utask->vaddr; 207 208 base_addr = udd->bp_addr - udd->tu->offset; 209 return base_addr + file_offset; 210 } 211 212 /* Note that we don't verify it, since the code does not come from user space */ 213 static int 214 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata, 215 void *dest, void *base) 216 { 217 struct pt_regs *regs = rec; 218 unsigned long val; 219 int ret; 220 221 /* 1st stage: get value from context */ 222 switch (code->op) { 223 case FETCH_OP_REG: 224 val = regs_get_register(regs, code->param); 225 break; 226 case FETCH_OP_STACK: 227 val = get_user_stack_nth(regs, code->param); 228 break; 229 case FETCH_OP_STACKP: 230 val = user_stack_pointer(regs); 231 break; 232 case FETCH_OP_RETVAL: 233 val = regs_return_value(regs); 234 break; 235 case FETCH_OP_COMM: 236 val = FETCH_TOKEN_COMM; 237 break; 238 case FETCH_OP_FOFFS: 239 val = translate_user_vaddr(code->immediate); 240 break; 241 default: 242 ret = process_common_fetch_insn(code, &val); 243 if (ret < 0) 244 return ret; 245 } 246 code++; 247 248 return process_fetch_insn_bottom(code, val, dest, base); 249 } 250 NOKPROBE_SYMBOL(process_fetch_insn) 251 252 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter) 253 { 254 rwlock_init(&filter->rwlock); 255 filter->nr_systemwide = 0; 256 INIT_LIST_HEAD(&filter->perf_events); 257 } 258 259 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter) 260 { 261 return !filter->nr_systemwide && list_empty(&filter->perf_events); 262 } 263 264 static inline bool is_ret_probe(struct trace_uprobe *tu) 265 { 266 return tu->consumer.ret_handler != NULL; 267 } 268 269 static bool trace_uprobe_is_busy(struct dyn_event *ev) 270 { 271 struct trace_uprobe *tu = to_trace_uprobe(ev); 272 273 return trace_probe_is_enabled(&tu->tp); 274 } 275 276 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu, 277 int argc, const char **argv) 278 { 279 char buf[MAX_ARGSTR_LEN + 1]; 280 int len; 281 282 if (!argc) 283 return true; 284 285 len = strlen(tu->filename); 286 if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':') 287 return false; 288 289 if (tu->ref_ctr_offset == 0) 290 snprintf(buf, sizeof(buf), "0x%0*lx", 291 (int)(sizeof(void *) * 2), tu->offset); 292 else 293 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)", 294 (int)(sizeof(void *) * 2), tu->offset, 295 tu->ref_ctr_offset); 296 if (strcmp(buf, &argv[0][len + 1])) 297 return false; 298 299 argc--; argv++; 300 301 return trace_probe_match_command_args(&tu->tp, argc, argv); 302 } 303 304 static bool trace_uprobe_match(const char *system, const char *event, 305 int argc, const char **argv, struct dyn_event *ev) 306 { 307 struct trace_uprobe *tu = to_trace_uprobe(ev); 308 309 return (event[0] == '\0' || 310 strcmp(trace_probe_name(&tu->tp), event) == 0) && 311 (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) && 312 trace_uprobe_match_command_head(tu, argc, argv); 313 } 314 315 static nokprobe_inline struct trace_uprobe * 316 trace_uprobe_primary_from_call(struct trace_event_call *call) 317 { 318 struct trace_probe *tp; 319 320 tp = trace_probe_primary_from_call(call); 321 if (WARN_ON_ONCE(!tp)) 322 return NULL; 323 324 return container_of(tp, struct trace_uprobe, tp); 325 } 326 327 /* 328 * Allocate new trace_uprobe and initialize it (including uprobes). 329 */ 330 static struct trace_uprobe * 331 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret) 332 { 333 struct trace_uprobe *tu; 334 int ret; 335 336 tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL); 337 if (!tu) 338 return ERR_PTR(-ENOMEM); 339 340 ret = trace_probe_init(&tu->tp, event, group, true, nargs); 341 if (ret < 0) 342 goto error; 343 344 dyn_event_init(&tu->devent, &trace_uprobe_ops); 345 tu->consumer.handler = uprobe_dispatcher; 346 if (is_ret) 347 tu->consumer.ret_handler = uretprobe_dispatcher; 348 init_trace_uprobe_filter(tu->tp.event->filter); 349 return tu; 350 351 error: 352 kfree(tu); 353 354 return ERR_PTR(ret); 355 } 356 357 static void free_trace_uprobe(struct trace_uprobe *tu) 358 { 359 if (!tu) 360 return; 361 362 path_put(&tu->path); 363 trace_probe_cleanup(&tu->tp); 364 kfree(tu->filename); 365 kfree(tu); 366 } 367 368 static struct trace_uprobe *find_probe_event(const char *event, const char *group) 369 { 370 struct dyn_event *pos; 371 struct trace_uprobe *tu; 372 373 for_each_trace_uprobe(tu, pos) 374 if (strcmp(trace_probe_name(&tu->tp), event) == 0 && 375 strcmp(trace_probe_group_name(&tu->tp), group) == 0) 376 return tu; 377 378 return NULL; 379 } 380 381 /* Unregister a trace_uprobe and probe_event */ 382 static int unregister_trace_uprobe(struct trace_uprobe *tu) 383 { 384 int ret; 385 386 if (trace_probe_has_sibling(&tu->tp)) 387 goto unreg; 388 389 /* If there's a reference to the dynamic event */ 390 if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp))) 391 return -EBUSY; 392 393 ret = unregister_uprobe_event(tu); 394 if (ret) 395 return ret; 396 397 unreg: 398 dyn_event_remove(&tu->devent); 399 trace_probe_unlink(&tu->tp); 400 free_trace_uprobe(tu); 401 return 0; 402 } 403 404 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig, 405 struct trace_uprobe *comp) 406 { 407 struct trace_probe_event *tpe = orig->tp.event; 408 struct inode *comp_inode = d_real_inode(comp->path.dentry); 409 int i; 410 411 list_for_each_entry(orig, &tpe->probes, tp.list) { 412 if (comp_inode != d_real_inode(orig->path.dentry) || 413 comp->offset != orig->offset) 414 continue; 415 416 /* 417 * trace_probe_compare_arg_type() ensured that nr_args and 418 * each argument name and type are same. Let's compare comm. 419 */ 420 for (i = 0; i < orig->tp.nr_args; i++) { 421 if (strcmp(orig->tp.args[i].comm, 422 comp->tp.args[i].comm)) 423 break; 424 } 425 426 if (i == orig->tp.nr_args) 427 return true; 428 } 429 430 return false; 431 } 432 433 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to) 434 { 435 int ret; 436 437 ret = trace_probe_compare_arg_type(&tu->tp, &to->tp); 438 if (ret) { 439 /* Note that argument starts index = 2 */ 440 trace_probe_log_set_index(ret + 1); 441 trace_probe_log_err(0, DIFF_ARG_TYPE); 442 return -EEXIST; 443 } 444 if (trace_uprobe_has_same_uprobe(to, tu)) { 445 trace_probe_log_set_index(0); 446 trace_probe_log_err(0, SAME_PROBE); 447 return -EEXIST; 448 } 449 450 /* Append to existing event */ 451 ret = trace_probe_append(&tu->tp, &to->tp); 452 if (!ret) 453 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp)); 454 455 return ret; 456 } 457 458 /* 459 * Uprobe with multiple reference counter is not allowed. i.e. 460 * If inode and offset matches, reference counter offset *must* 461 * match as well. Though, there is one exception: If user is 462 * replacing old trace_uprobe with new one(same group/event), 463 * then we allow same uprobe with new reference counter as far 464 * as the new one does not conflict with any other existing 465 * ones. 466 */ 467 static int validate_ref_ctr_offset(struct trace_uprobe *new) 468 { 469 struct dyn_event *pos; 470 struct trace_uprobe *tmp; 471 struct inode *new_inode = d_real_inode(new->path.dentry); 472 473 for_each_trace_uprobe(tmp, pos) { 474 if (new_inode == d_real_inode(tmp->path.dentry) && 475 new->offset == tmp->offset && 476 new->ref_ctr_offset != tmp->ref_ctr_offset) { 477 pr_warn("Reference counter offset mismatch."); 478 return -EINVAL; 479 } 480 } 481 return 0; 482 } 483 484 /* Register a trace_uprobe and probe_event */ 485 static int register_trace_uprobe(struct trace_uprobe *tu) 486 { 487 struct trace_uprobe *old_tu; 488 int ret; 489 490 mutex_lock(&event_mutex); 491 492 ret = validate_ref_ctr_offset(tu); 493 if (ret) 494 goto end; 495 496 /* register as an event */ 497 old_tu = find_probe_event(trace_probe_name(&tu->tp), 498 trace_probe_group_name(&tu->tp)); 499 if (old_tu) { 500 if (is_ret_probe(tu) != is_ret_probe(old_tu)) { 501 trace_probe_log_set_index(0); 502 trace_probe_log_err(0, DIFF_PROBE_TYPE); 503 ret = -EEXIST; 504 } else { 505 ret = append_trace_uprobe(tu, old_tu); 506 } 507 goto end; 508 } 509 510 ret = register_uprobe_event(tu); 511 if (ret) { 512 if (ret == -EEXIST) { 513 trace_probe_log_set_index(0); 514 trace_probe_log_err(0, EVENT_EXIST); 515 } else 516 pr_warn("Failed to register probe event(%d)\n", ret); 517 goto end; 518 } 519 520 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp)); 521 522 end: 523 mutex_unlock(&event_mutex); 524 525 return ret; 526 } 527 528 /* 529 * Argument syntax: 530 * - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS] 531 */ 532 static int __trace_uprobe_create(int argc, const char **argv) 533 { 534 struct trace_uprobe *tu; 535 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM; 536 char *arg, *filename, *rctr, *rctr_end, *tmp; 537 char buf[MAX_EVENT_NAME_LEN]; 538 char gbuf[MAX_EVENT_NAME_LEN]; 539 enum probe_print_type ptype; 540 struct path path; 541 unsigned long offset, ref_ctr_offset; 542 bool is_return = false; 543 int i, ret; 544 545 ref_ctr_offset = 0; 546 547 switch (argv[0][0]) { 548 case 'r': 549 is_return = true; 550 break; 551 case 'p': 552 break; 553 default: 554 return -ECANCELED; 555 } 556 557 if (argc < 2) 558 return -ECANCELED; 559 560 if (argv[0][1] == ':') 561 event = &argv[0][2]; 562 563 if (!strchr(argv[1], '/')) 564 return -ECANCELED; 565 566 filename = kstrdup(argv[1], GFP_KERNEL); 567 if (!filename) 568 return -ENOMEM; 569 570 /* Find the last occurrence, in case the path contains ':' too. */ 571 arg = strrchr(filename, ':'); 572 if (!arg || !isdigit(arg[1])) { 573 kfree(filename); 574 return -ECANCELED; 575 } 576 577 trace_probe_log_init("trace_uprobe", argc, argv); 578 trace_probe_log_set_index(1); /* filename is the 2nd argument */ 579 580 *arg++ = '\0'; 581 ret = kern_path(filename, LOOKUP_FOLLOW, &path); 582 if (ret) { 583 trace_probe_log_err(0, FILE_NOT_FOUND); 584 kfree(filename); 585 trace_probe_log_clear(); 586 return ret; 587 } 588 if (!d_is_reg(path.dentry)) { 589 trace_probe_log_err(0, NO_REGULAR_FILE); 590 ret = -EINVAL; 591 goto fail_address_parse; 592 } 593 594 /* Parse reference counter offset if specified. */ 595 rctr = strchr(arg, '('); 596 if (rctr) { 597 rctr_end = strchr(rctr, ')'); 598 if (!rctr_end) { 599 ret = -EINVAL; 600 rctr_end = rctr + strlen(rctr); 601 trace_probe_log_err(rctr_end - filename, 602 REFCNT_OPEN_BRACE); 603 goto fail_address_parse; 604 } else if (rctr_end[1] != '\0') { 605 ret = -EINVAL; 606 trace_probe_log_err(rctr_end + 1 - filename, 607 BAD_REFCNT_SUFFIX); 608 goto fail_address_parse; 609 } 610 611 *rctr++ = '\0'; 612 *rctr_end = '\0'; 613 ret = kstrtoul(rctr, 0, &ref_ctr_offset); 614 if (ret) { 615 trace_probe_log_err(rctr - filename, BAD_REFCNT); 616 goto fail_address_parse; 617 } 618 } 619 620 /* Check if there is %return suffix */ 621 tmp = strchr(arg, '%'); 622 if (tmp) { 623 if (!strcmp(tmp, "%return")) { 624 *tmp = '\0'; 625 is_return = true; 626 } else { 627 trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX); 628 ret = -EINVAL; 629 goto fail_address_parse; 630 } 631 } 632 633 /* Parse uprobe offset. */ 634 ret = kstrtoul(arg, 0, &offset); 635 if (ret) { 636 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS); 637 goto fail_address_parse; 638 } 639 640 /* setup a probe */ 641 trace_probe_log_set_index(0); 642 if (event) { 643 ret = traceprobe_parse_event_name(&event, &group, gbuf, 644 event - argv[0]); 645 if (ret) 646 goto fail_address_parse; 647 } 648 649 if (!event) { 650 char *tail; 651 char *ptr; 652 653 tail = kstrdup(kbasename(filename), GFP_KERNEL); 654 if (!tail) { 655 ret = -ENOMEM; 656 goto fail_address_parse; 657 } 658 659 ptr = strpbrk(tail, ".-_"); 660 if (ptr) 661 *ptr = '\0'; 662 663 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset); 664 event = buf; 665 kfree(tail); 666 } 667 668 argc -= 2; 669 argv += 2; 670 671 tu = alloc_trace_uprobe(group, event, argc, is_return); 672 if (IS_ERR(tu)) { 673 ret = PTR_ERR(tu); 674 /* This must return -ENOMEM otherwise there is a bug */ 675 WARN_ON_ONCE(ret != -ENOMEM); 676 goto fail_address_parse; 677 } 678 tu->offset = offset; 679 tu->ref_ctr_offset = ref_ctr_offset; 680 tu->path = path; 681 tu->filename = filename; 682 683 /* parse arguments */ 684 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) { 685 struct traceprobe_parse_context ctx = { 686 .flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER, 687 }; 688 689 trace_probe_log_set_index(i + 2); 690 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx); 691 traceprobe_finish_parse(&ctx); 692 if (ret) 693 goto error; 694 } 695 696 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL; 697 ret = traceprobe_set_print_fmt(&tu->tp, ptype); 698 if (ret < 0) 699 goto error; 700 701 ret = register_trace_uprobe(tu); 702 if (!ret) 703 goto out; 704 705 error: 706 free_trace_uprobe(tu); 707 out: 708 trace_probe_log_clear(); 709 return ret; 710 711 fail_address_parse: 712 trace_probe_log_clear(); 713 path_put(&path); 714 kfree(filename); 715 716 return ret; 717 } 718 719 int trace_uprobe_create(const char *raw_command) 720 { 721 return trace_probe_create(raw_command, __trace_uprobe_create); 722 } 723 724 static int create_or_delete_trace_uprobe(const char *raw_command) 725 { 726 int ret; 727 728 if (raw_command[0] == '-') 729 return dyn_event_release(raw_command, &trace_uprobe_ops); 730 731 ret = trace_uprobe_create(raw_command); 732 return ret == -ECANCELED ? -EINVAL : ret; 733 } 734 735 static int trace_uprobe_release(struct dyn_event *ev) 736 { 737 struct trace_uprobe *tu = to_trace_uprobe(ev); 738 739 return unregister_trace_uprobe(tu); 740 } 741 742 /* Probes listing interfaces */ 743 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev) 744 { 745 struct trace_uprobe *tu = to_trace_uprobe(ev); 746 char c = is_ret_probe(tu) ? 'r' : 'p'; 747 int i; 748 749 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp), 750 trace_probe_name(&tu->tp), tu->filename, 751 (int)(sizeof(void *) * 2), tu->offset); 752 753 if (tu->ref_ctr_offset) 754 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset); 755 756 for (i = 0; i < tu->tp.nr_args; i++) 757 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm); 758 759 seq_putc(m, '\n'); 760 return 0; 761 } 762 763 static int probes_seq_show(struct seq_file *m, void *v) 764 { 765 struct dyn_event *ev = v; 766 767 if (!is_trace_uprobe(ev)) 768 return 0; 769 770 return trace_uprobe_show(m, ev); 771 } 772 773 static const struct seq_operations probes_seq_op = { 774 .start = dyn_event_seq_start, 775 .next = dyn_event_seq_next, 776 .stop = dyn_event_seq_stop, 777 .show = probes_seq_show 778 }; 779 780 static int probes_open(struct inode *inode, struct file *file) 781 { 782 int ret; 783 784 ret = security_locked_down(LOCKDOWN_TRACEFS); 785 if (ret) 786 return ret; 787 788 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) { 789 ret = dyn_events_release_all(&trace_uprobe_ops); 790 if (ret) 791 return ret; 792 } 793 794 return seq_open(file, &probes_seq_op); 795 } 796 797 static ssize_t probes_write(struct file *file, const char __user *buffer, 798 size_t count, loff_t *ppos) 799 { 800 return trace_parse_run_command(file, buffer, count, ppos, 801 create_or_delete_trace_uprobe); 802 } 803 804 static const struct file_operations uprobe_events_ops = { 805 .owner = THIS_MODULE, 806 .open = probes_open, 807 .read = seq_read, 808 .llseek = seq_lseek, 809 .release = seq_release, 810 .write = probes_write, 811 }; 812 813 /* Probes profiling interfaces */ 814 static int probes_profile_seq_show(struct seq_file *m, void *v) 815 { 816 struct dyn_event *ev = v; 817 struct trace_uprobe *tu; 818 819 if (!is_trace_uprobe(ev)) 820 return 0; 821 822 tu = to_trace_uprobe(ev); 823 seq_printf(m, " %s %-44s %15lu\n", tu->filename, 824 trace_probe_name(&tu->tp), tu->nhit); 825 return 0; 826 } 827 828 static const struct seq_operations profile_seq_op = { 829 .start = dyn_event_seq_start, 830 .next = dyn_event_seq_next, 831 .stop = dyn_event_seq_stop, 832 .show = probes_profile_seq_show 833 }; 834 835 static int profile_open(struct inode *inode, struct file *file) 836 { 837 int ret; 838 839 ret = security_locked_down(LOCKDOWN_TRACEFS); 840 if (ret) 841 return ret; 842 843 return seq_open(file, &profile_seq_op); 844 } 845 846 static const struct file_operations uprobe_profile_ops = { 847 .owner = THIS_MODULE, 848 .open = profile_open, 849 .read = seq_read, 850 .llseek = seq_lseek, 851 .release = seq_release, 852 }; 853 854 struct uprobe_cpu_buffer { 855 struct mutex mutex; 856 void *buf; 857 int dsize; 858 }; 859 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer; 860 static int uprobe_buffer_refcnt; 861 862 static int uprobe_buffer_init(void) 863 { 864 int cpu, err_cpu; 865 866 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer); 867 if (uprobe_cpu_buffer == NULL) 868 return -ENOMEM; 869 870 for_each_possible_cpu(cpu) { 871 struct page *p = alloc_pages_node(cpu_to_node(cpu), 872 GFP_KERNEL, 0); 873 if (p == NULL) { 874 err_cpu = cpu; 875 goto err; 876 } 877 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p); 878 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex); 879 } 880 881 return 0; 882 883 err: 884 for_each_possible_cpu(cpu) { 885 if (cpu == err_cpu) 886 break; 887 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf); 888 } 889 890 free_percpu(uprobe_cpu_buffer); 891 return -ENOMEM; 892 } 893 894 static int uprobe_buffer_enable(void) 895 { 896 int ret = 0; 897 898 BUG_ON(!mutex_is_locked(&event_mutex)); 899 900 if (uprobe_buffer_refcnt++ == 0) { 901 ret = uprobe_buffer_init(); 902 if (ret < 0) 903 uprobe_buffer_refcnt--; 904 } 905 906 return ret; 907 } 908 909 static void uprobe_buffer_disable(void) 910 { 911 int cpu; 912 913 BUG_ON(!mutex_is_locked(&event_mutex)); 914 915 if (--uprobe_buffer_refcnt == 0) { 916 for_each_possible_cpu(cpu) 917 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, 918 cpu)->buf); 919 920 free_percpu(uprobe_cpu_buffer); 921 uprobe_cpu_buffer = NULL; 922 } 923 } 924 925 static struct uprobe_cpu_buffer *uprobe_buffer_get(void) 926 { 927 struct uprobe_cpu_buffer *ucb; 928 int cpu; 929 930 cpu = raw_smp_processor_id(); 931 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu); 932 933 /* 934 * Use per-cpu buffers for fastest access, but we might migrate 935 * so the mutex makes sure we have sole access to it. 936 */ 937 mutex_lock(&ucb->mutex); 938 939 return ucb; 940 } 941 942 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb) 943 { 944 if (!ucb) 945 return; 946 mutex_unlock(&ucb->mutex); 947 } 948 949 static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu, 950 struct pt_regs *regs, 951 struct uprobe_cpu_buffer **ucbp) 952 { 953 struct uprobe_cpu_buffer *ucb; 954 int dsize, esize; 955 956 if (*ucbp) 957 return *ucbp; 958 959 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 960 dsize = __get_data_size(&tu->tp, regs, NULL); 961 962 ucb = uprobe_buffer_get(); 963 ucb->dsize = tu->tp.size + dsize; 964 965 store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize); 966 967 *ucbp = ucb; 968 return ucb; 969 } 970 971 static void __uprobe_trace_func(struct trace_uprobe *tu, 972 unsigned long func, struct pt_regs *regs, 973 struct uprobe_cpu_buffer **ucbp, 974 struct trace_event_file *trace_file) 975 { 976 struct uprobe_trace_entry_head *entry; 977 struct trace_event_buffer fbuffer; 978 struct uprobe_cpu_buffer *ucb; 979 void *data; 980 int size, esize; 981 struct trace_event_call *call = trace_probe_event_call(&tu->tp); 982 983 WARN_ON(call != trace_file->event_call); 984 985 ucb = prepare_uprobe_buffer(tu, regs, ucbp); 986 if (WARN_ON_ONCE(ucb->dsize > PAGE_SIZE)) 987 return; 988 989 if (trace_trigger_soft_disabled(trace_file)) 990 return; 991 992 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 993 size = esize + ucb->dsize; 994 entry = trace_event_buffer_reserve(&fbuffer, trace_file, size); 995 if (!entry) 996 return; 997 998 if (is_ret_probe(tu)) { 999 entry->vaddr[0] = func; 1000 entry->vaddr[1] = instruction_pointer(regs); 1001 data = DATAOF_TRACE_ENTRY(entry, true); 1002 } else { 1003 entry->vaddr[0] = instruction_pointer(regs); 1004 data = DATAOF_TRACE_ENTRY(entry, false); 1005 } 1006 1007 memcpy(data, ucb->buf, ucb->dsize); 1008 1009 trace_event_buffer_commit(&fbuffer); 1010 } 1011 1012 /* uprobe handler */ 1013 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs, 1014 struct uprobe_cpu_buffer **ucbp) 1015 { 1016 struct event_file_link *link; 1017 1018 if (is_ret_probe(tu)) 1019 return 0; 1020 1021 rcu_read_lock(); 1022 trace_probe_for_each_link_rcu(link, &tu->tp) 1023 __uprobe_trace_func(tu, 0, regs, ucbp, link->file); 1024 rcu_read_unlock(); 1025 1026 return 0; 1027 } 1028 1029 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func, 1030 struct pt_regs *regs, 1031 struct uprobe_cpu_buffer **ucbp) 1032 { 1033 struct event_file_link *link; 1034 1035 rcu_read_lock(); 1036 trace_probe_for_each_link_rcu(link, &tu->tp) 1037 __uprobe_trace_func(tu, func, regs, ucbp, link->file); 1038 rcu_read_unlock(); 1039 } 1040 1041 /* Event entry printers */ 1042 static enum print_line_t 1043 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event) 1044 { 1045 struct uprobe_trace_entry_head *entry; 1046 struct trace_seq *s = &iter->seq; 1047 struct trace_uprobe *tu; 1048 u8 *data; 1049 1050 entry = (struct uprobe_trace_entry_head *)iter->ent; 1051 tu = trace_uprobe_primary_from_call( 1052 container_of(event, struct trace_event_call, event)); 1053 if (unlikely(!tu)) 1054 goto out; 1055 1056 if (is_ret_probe(tu)) { 1057 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)", 1058 trace_probe_name(&tu->tp), 1059 entry->vaddr[1], entry->vaddr[0]); 1060 data = DATAOF_TRACE_ENTRY(entry, true); 1061 } else { 1062 trace_seq_printf(s, "%s: (0x%lx)", 1063 trace_probe_name(&tu->tp), 1064 entry->vaddr[0]); 1065 data = DATAOF_TRACE_ENTRY(entry, false); 1066 } 1067 1068 if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0) 1069 goto out; 1070 1071 trace_seq_putc(s, '\n'); 1072 1073 out: 1074 return trace_handle_return(s); 1075 } 1076 1077 typedef bool (*filter_func_t)(struct uprobe_consumer *self, 1078 enum uprobe_filter_ctx ctx, 1079 struct mm_struct *mm); 1080 1081 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter) 1082 { 1083 int ret; 1084 1085 tu->consumer.filter = filter; 1086 tu->inode = d_real_inode(tu->path.dentry); 1087 1088 if (tu->ref_ctr_offset) 1089 ret = uprobe_register_refctr(tu->inode, tu->offset, 1090 tu->ref_ctr_offset, &tu->consumer); 1091 else 1092 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer); 1093 1094 if (ret) 1095 tu->inode = NULL; 1096 1097 return ret; 1098 } 1099 1100 static void __probe_event_disable(struct trace_probe *tp) 1101 { 1102 struct trace_uprobe *tu; 1103 1104 tu = container_of(tp, struct trace_uprobe, tp); 1105 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter)); 1106 1107 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) { 1108 if (!tu->inode) 1109 continue; 1110 1111 uprobe_unregister(tu->inode, tu->offset, &tu->consumer); 1112 tu->inode = NULL; 1113 } 1114 } 1115 1116 static int probe_event_enable(struct trace_event_call *call, 1117 struct trace_event_file *file, filter_func_t filter) 1118 { 1119 struct trace_probe *tp; 1120 struct trace_uprobe *tu; 1121 bool enabled; 1122 int ret; 1123 1124 tp = trace_probe_primary_from_call(call); 1125 if (WARN_ON_ONCE(!tp)) 1126 return -ENODEV; 1127 enabled = trace_probe_is_enabled(tp); 1128 1129 /* This may also change "enabled" state */ 1130 if (file) { 1131 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE)) 1132 return -EINTR; 1133 1134 ret = trace_probe_add_file(tp, file); 1135 if (ret < 0) 1136 return ret; 1137 } else { 1138 if (trace_probe_test_flag(tp, TP_FLAG_TRACE)) 1139 return -EINTR; 1140 1141 trace_probe_set_flag(tp, TP_FLAG_PROFILE); 1142 } 1143 1144 tu = container_of(tp, struct trace_uprobe, tp); 1145 WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter)); 1146 1147 if (enabled) 1148 return 0; 1149 1150 ret = uprobe_buffer_enable(); 1151 if (ret) 1152 goto err_flags; 1153 1154 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) { 1155 ret = trace_uprobe_enable(tu, filter); 1156 if (ret) { 1157 __probe_event_disable(tp); 1158 goto err_buffer; 1159 } 1160 } 1161 1162 return 0; 1163 1164 err_buffer: 1165 uprobe_buffer_disable(); 1166 1167 err_flags: 1168 if (file) 1169 trace_probe_remove_file(tp, file); 1170 else 1171 trace_probe_clear_flag(tp, TP_FLAG_PROFILE); 1172 1173 return ret; 1174 } 1175 1176 static void probe_event_disable(struct trace_event_call *call, 1177 struct trace_event_file *file) 1178 { 1179 struct trace_probe *tp; 1180 1181 tp = trace_probe_primary_from_call(call); 1182 if (WARN_ON_ONCE(!tp)) 1183 return; 1184 1185 if (!trace_probe_is_enabled(tp)) 1186 return; 1187 1188 if (file) { 1189 if (trace_probe_remove_file(tp, file) < 0) 1190 return; 1191 1192 if (trace_probe_is_enabled(tp)) 1193 return; 1194 } else 1195 trace_probe_clear_flag(tp, TP_FLAG_PROFILE); 1196 1197 __probe_event_disable(tp); 1198 uprobe_buffer_disable(); 1199 } 1200 1201 static int uprobe_event_define_fields(struct trace_event_call *event_call) 1202 { 1203 int ret, size; 1204 struct uprobe_trace_entry_head field; 1205 struct trace_uprobe *tu; 1206 1207 tu = trace_uprobe_primary_from_call(event_call); 1208 if (unlikely(!tu)) 1209 return -ENODEV; 1210 1211 if (is_ret_probe(tu)) { 1212 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0); 1213 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0); 1214 size = SIZEOF_TRACE_ENTRY(true); 1215 } else { 1216 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0); 1217 size = SIZEOF_TRACE_ENTRY(false); 1218 } 1219 1220 return traceprobe_define_arg_fields(event_call, size, &tu->tp); 1221 } 1222 1223 #ifdef CONFIG_PERF_EVENTS 1224 static bool 1225 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm) 1226 { 1227 struct perf_event *event; 1228 1229 list_for_each_entry(event, &filter->perf_events, hw.tp_list) { 1230 if (event->hw.target->mm == mm) 1231 return true; 1232 } 1233 1234 return false; 1235 } 1236 1237 static inline bool 1238 trace_uprobe_filter_event(struct trace_uprobe_filter *filter, 1239 struct perf_event *event) 1240 { 1241 return __uprobe_perf_filter(filter, event->hw.target->mm); 1242 } 1243 1244 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter, 1245 struct perf_event *event) 1246 { 1247 bool done; 1248 1249 write_lock(&filter->rwlock); 1250 if (event->hw.target) { 1251 list_del(&event->hw.tp_list); 1252 done = filter->nr_systemwide || 1253 (event->hw.target->flags & PF_EXITING) || 1254 trace_uprobe_filter_event(filter, event); 1255 } else { 1256 filter->nr_systemwide--; 1257 done = filter->nr_systemwide; 1258 } 1259 write_unlock(&filter->rwlock); 1260 1261 return done; 1262 } 1263 1264 /* This returns true if the filter always covers target mm */ 1265 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter, 1266 struct perf_event *event) 1267 { 1268 bool done; 1269 1270 write_lock(&filter->rwlock); 1271 if (event->hw.target) { 1272 /* 1273 * event->parent != NULL means copy_process(), we can avoid 1274 * uprobe_apply(). current->mm must be probed and we can rely 1275 * on dup_mmap() which preserves the already installed bp's. 1276 * 1277 * attr.enable_on_exec means that exec/mmap will install the 1278 * breakpoints we need. 1279 */ 1280 done = filter->nr_systemwide || 1281 event->parent || event->attr.enable_on_exec || 1282 trace_uprobe_filter_event(filter, event); 1283 list_add(&event->hw.tp_list, &filter->perf_events); 1284 } else { 1285 done = filter->nr_systemwide; 1286 filter->nr_systemwide++; 1287 } 1288 write_unlock(&filter->rwlock); 1289 1290 return done; 1291 } 1292 1293 static int uprobe_perf_close(struct trace_event_call *call, 1294 struct perf_event *event) 1295 { 1296 struct trace_probe *tp; 1297 struct trace_uprobe *tu; 1298 int ret = 0; 1299 1300 tp = trace_probe_primary_from_call(call); 1301 if (WARN_ON_ONCE(!tp)) 1302 return -ENODEV; 1303 1304 tu = container_of(tp, struct trace_uprobe, tp); 1305 if (trace_uprobe_filter_remove(tu->tp.event->filter, event)) 1306 return 0; 1307 1308 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) { 1309 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false); 1310 if (ret) 1311 break; 1312 } 1313 1314 return ret; 1315 } 1316 1317 static int uprobe_perf_open(struct trace_event_call *call, 1318 struct perf_event *event) 1319 { 1320 struct trace_probe *tp; 1321 struct trace_uprobe *tu; 1322 int err = 0; 1323 1324 tp = trace_probe_primary_from_call(call); 1325 if (WARN_ON_ONCE(!tp)) 1326 return -ENODEV; 1327 1328 tu = container_of(tp, struct trace_uprobe, tp); 1329 if (trace_uprobe_filter_add(tu->tp.event->filter, event)) 1330 return 0; 1331 1332 list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) { 1333 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true); 1334 if (err) { 1335 uprobe_perf_close(call, event); 1336 break; 1337 } 1338 } 1339 1340 return err; 1341 } 1342 1343 static bool uprobe_perf_filter(struct uprobe_consumer *uc, 1344 enum uprobe_filter_ctx ctx, struct mm_struct *mm) 1345 { 1346 struct trace_uprobe_filter *filter; 1347 struct trace_uprobe *tu; 1348 int ret; 1349 1350 tu = container_of(uc, struct trace_uprobe, consumer); 1351 filter = tu->tp.event->filter; 1352 1353 /* 1354 * speculative short-circuiting check to avoid unnecessarily taking 1355 * filter->rwlock below, if the uprobe has system-wide consumer 1356 */ 1357 if (READ_ONCE(filter->nr_systemwide)) 1358 return true; 1359 1360 read_lock(&filter->rwlock); 1361 ret = __uprobe_perf_filter(filter, mm); 1362 read_unlock(&filter->rwlock); 1363 1364 return ret; 1365 } 1366 1367 static void __uprobe_perf_func(struct trace_uprobe *tu, 1368 unsigned long func, struct pt_regs *regs, 1369 struct uprobe_cpu_buffer **ucbp) 1370 { 1371 struct trace_event_call *call = trace_probe_event_call(&tu->tp); 1372 struct uprobe_trace_entry_head *entry; 1373 struct uprobe_cpu_buffer *ucb; 1374 struct hlist_head *head; 1375 void *data; 1376 int size, esize; 1377 int rctx; 1378 1379 #ifdef CONFIG_BPF_EVENTS 1380 if (bpf_prog_array_valid(call)) { 1381 u32 ret; 1382 1383 ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run); 1384 if (!ret) 1385 return; 1386 } 1387 #endif /* CONFIG_BPF_EVENTS */ 1388 1389 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 1390 1391 ucb = prepare_uprobe_buffer(tu, regs, ucbp); 1392 size = esize + ucb->dsize; 1393 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32); 1394 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough")) 1395 return; 1396 1397 preempt_disable(); 1398 head = this_cpu_ptr(call->perf_events); 1399 if (hlist_empty(head)) 1400 goto out; 1401 1402 entry = perf_trace_buf_alloc(size, NULL, &rctx); 1403 if (!entry) 1404 goto out; 1405 1406 if (is_ret_probe(tu)) { 1407 entry->vaddr[0] = func; 1408 entry->vaddr[1] = instruction_pointer(regs); 1409 data = DATAOF_TRACE_ENTRY(entry, true); 1410 } else { 1411 entry->vaddr[0] = instruction_pointer(regs); 1412 data = DATAOF_TRACE_ENTRY(entry, false); 1413 } 1414 1415 memcpy(data, ucb->buf, ucb->dsize); 1416 1417 if (size - esize > ucb->dsize) 1418 memset(data + ucb->dsize, 0, size - esize - ucb->dsize); 1419 1420 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs, 1421 head, NULL); 1422 out: 1423 preempt_enable(); 1424 } 1425 1426 /* uprobe profile handler */ 1427 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs, 1428 struct uprobe_cpu_buffer **ucbp) 1429 { 1430 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm)) 1431 return UPROBE_HANDLER_REMOVE; 1432 1433 if (!is_ret_probe(tu)) 1434 __uprobe_perf_func(tu, 0, regs, ucbp); 1435 return 0; 1436 } 1437 1438 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func, 1439 struct pt_regs *regs, 1440 struct uprobe_cpu_buffer **ucbp) 1441 { 1442 __uprobe_perf_func(tu, func, regs, ucbp); 1443 } 1444 1445 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type, 1446 const char **filename, u64 *probe_offset, 1447 u64 *probe_addr, bool perf_type_tracepoint) 1448 { 1449 const char *pevent = trace_event_name(event->tp_event); 1450 const char *group = event->tp_event->class->system; 1451 struct trace_uprobe *tu; 1452 1453 if (perf_type_tracepoint) 1454 tu = find_probe_event(pevent, group); 1455 else 1456 tu = trace_uprobe_primary_from_call(event->tp_event); 1457 if (!tu) 1458 return -EINVAL; 1459 1460 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE 1461 : BPF_FD_TYPE_UPROBE; 1462 *filename = tu->filename; 1463 *probe_offset = tu->offset; 1464 *probe_addr = 0; 1465 return 0; 1466 } 1467 #endif /* CONFIG_PERF_EVENTS */ 1468 1469 static int 1470 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type, 1471 void *data) 1472 { 1473 struct trace_event_file *file = data; 1474 1475 switch (type) { 1476 case TRACE_REG_REGISTER: 1477 return probe_event_enable(event, file, NULL); 1478 1479 case TRACE_REG_UNREGISTER: 1480 probe_event_disable(event, file); 1481 return 0; 1482 1483 #ifdef CONFIG_PERF_EVENTS 1484 case TRACE_REG_PERF_REGISTER: 1485 return probe_event_enable(event, NULL, uprobe_perf_filter); 1486 1487 case TRACE_REG_PERF_UNREGISTER: 1488 probe_event_disable(event, NULL); 1489 return 0; 1490 1491 case TRACE_REG_PERF_OPEN: 1492 return uprobe_perf_open(event, data); 1493 1494 case TRACE_REG_PERF_CLOSE: 1495 return uprobe_perf_close(event, data); 1496 1497 #endif 1498 default: 1499 return 0; 1500 } 1501 } 1502 1503 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs) 1504 { 1505 struct trace_uprobe *tu; 1506 struct uprobe_dispatch_data udd; 1507 struct uprobe_cpu_buffer *ucb = NULL; 1508 int ret = 0; 1509 1510 tu = container_of(con, struct trace_uprobe, consumer); 1511 tu->nhit++; 1512 1513 udd.tu = tu; 1514 udd.bp_addr = instruction_pointer(regs); 1515 1516 current->utask->vaddr = (unsigned long) &udd; 1517 1518 if (WARN_ON_ONCE(!uprobe_cpu_buffer)) 1519 return 0; 1520 1521 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE)) 1522 ret |= uprobe_trace_func(tu, regs, &ucb); 1523 1524 #ifdef CONFIG_PERF_EVENTS 1525 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE)) 1526 ret |= uprobe_perf_func(tu, regs, &ucb); 1527 #endif 1528 uprobe_buffer_put(ucb); 1529 return ret; 1530 } 1531 1532 static int uretprobe_dispatcher(struct uprobe_consumer *con, 1533 unsigned long func, struct pt_regs *regs) 1534 { 1535 struct trace_uprobe *tu; 1536 struct uprobe_dispatch_data udd; 1537 struct uprobe_cpu_buffer *ucb = NULL; 1538 1539 tu = container_of(con, struct trace_uprobe, consumer); 1540 1541 udd.tu = tu; 1542 udd.bp_addr = func; 1543 1544 current->utask->vaddr = (unsigned long) &udd; 1545 1546 if (WARN_ON_ONCE(!uprobe_cpu_buffer)) 1547 return 0; 1548 1549 if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE)) 1550 uretprobe_trace_func(tu, func, regs, &ucb); 1551 1552 #ifdef CONFIG_PERF_EVENTS 1553 if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE)) 1554 uretprobe_perf_func(tu, func, regs, &ucb); 1555 #endif 1556 uprobe_buffer_put(ucb); 1557 return 0; 1558 } 1559 1560 static struct trace_event_functions uprobe_funcs = { 1561 .trace = print_uprobe_event 1562 }; 1563 1564 static struct trace_event_fields uprobe_fields_array[] = { 1565 { .type = TRACE_FUNCTION_TYPE, 1566 .define_fields = uprobe_event_define_fields }, 1567 {} 1568 }; 1569 1570 static inline void init_trace_event_call(struct trace_uprobe *tu) 1571 { 1572 struct trace_event_call *call = trace_probe_event_call(&tu->tp); 1573 call->event.funcs = &uprobe_funcs; 1574 call->class->fields_array = uprobe_fields_array; 1575 1576 call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY; 1577 call->class->reg = trace_uprobe_register; 1578 } 1579 1580 static int register_uprobe_event(struct trace_uprobe *tu) 1581 { 1582 init_trace_event_call(tu); 1583 1584 return trace_probe_register_event_call(&tu->tp); 1585 } 1586 1587 static int unregister_uprobe_event(struct trace_uprobe *tu) 1588 { 1589 return trace_probe_unregister_event_call(&tu->tp); 1590 } 1591 1592 #ifdef CONFIG_PERF_EVENTS 1593 struct trace_event_call * 1594 create_local_trace_uprobe(char *name, unsigned long offs, 1595 unsigned long ref_ctr_offset, bool is_return) 1596 { 1597 enum probe_print_type ptype; 1598 struct trace_uprobe *tu; 1599 struct path path; 1600 int ret; 1601 1602 ret = kern_path(name, LOOKUP_FOLLOW, &path); 1603 if (ret) 1604 return ERR_PTR(ret); 1605 1606 if (!d_is_reg(path.dentry)) { 1607 path_put(&path); 1608 return ERR_PTR(-EINVAL); 1609 } 1610 1611 /* 1612 * local trace_kprobes are not added to dyn_event, so they are never 1613 * searched in find_trace_kprobe(). Therefore, there is no concern of 1614 * duplicated name "DUMMY_EVENT" here. 1615 */ 1616 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0, 1617 is_return); 1618 1619 if (IS_ERR(tu)) { 1620 pr_info("Failed to allocate trace_uprobe.(%d)\n", 1621 (int)PTR_ERR(tu)); 1622 path_put(&path); 1623 return ERR_CAST(tu); 1624 } 1625 1626 tu->offset = offs; 1627 tu->path = path; 1628 tu->ref_ctr_offset = ref_ctr_offset; 1629 tu->filename = kstrdup(name, GFP_KERNEL); 1630 if (!tu->filename) { 1631 ret = -ENOMEM; 1632 goto error; 1633 } 1634 1635 init_trace_event_call(tu); 1636 1637 ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL; 1638 if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) { 1639 ret = -ENOMEM; 1640 goto error; 1641 } 1642 1643 return trace_probe_event_call(&tu->tp); 1644 error: 1645 free_trace_uprobe(tu); 1646 return ERR_PTR(ret); 1647 } 1648 1649 void destroy_local_trace_uprobe(struct trace_event_call *event_call) 1650 { 1651 struct trace_uprobe *tu; 1652 1653 tu = trace_uprobe_primary_from_call(event_call); 1654 1655 free_trace_uprobe(tu); 1656 } 1657 #endif /* CONFIG_PERF_EVENTS */ 1658 1659 /* Make a trace interface for controlling probe points */ 1660 static __init int init_uprobe_trace(void) 1661 { 1662 int ret; 1663 1664 ret = dyn_event_register(&trace_uprobe_ops); 1665 if (ret) 1666 return ret; 1667 1668 ret = tracing_init_dentry(); 1669 if (ret) 1670 return 0; 1671 1672 trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL, 1673 NULL, &uprobe_events_ops); 1674 /* Profile interface */ 1675 trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL, 1676 NULL, &uprobe_profile_ops); 1677 return 0; 1678 } 1679 1680 fs_initcall(init_uprobe_trace); 1681