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_kprobe: " fmt 9 10 #include <linux/ctype.h> 11 #include <linux/module.h> 12 #include <linux/uaccess.h> 13 #include <linux/uprobes.h> 14 #include <linux/namei.h> 15 #include <linux/string.h> 16 #include <linux/rculist.h> 17 18 #include "trace_dynevent.h" 19 #include "trace_probe.h" 20 #include "trace_probe_tmpl.h" 21 22 #define UPROBE_EVENT_SYSTEM "uprobes" 23 24 struct uprobe_trace_entry_head { 25 struct trace_entry ent; 26 unsigned long vaddr[]; 27 }; 28 29 #define SIZEOF_TRACE_ENTRY(is_return) \ 30 (sizeof(struct uprobe_trace_entry_head) + \ 31 sizeof(unsigned long) * (is_return ? 2 : 1)) 32 33 #define DATAOF_TRACE_ENTRY(entry, is_return) \ 34 ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return)) 35 36 struct trace_uprobe_filter { 37 rwlock_t rwlock; 38 int nr_systemwide; 39 struct list_head perf_events; 40 }; 41 42 static int trace_uprobe_create(int argc, const char **argv); 43 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev); 44 static int trace_uprobe_release(struct dyn_event *ev); 45 static bool trace_uprobe_is_busy(struct dyn_event *ev); 46 static bool trace_uprobe_match(const char *system, const char *event, 47 struct dyn_event *ev); 48 49 static struct dyn_event_operations trace_uprobe_ops = { 50 .create = trace_uprobe_create, 51 .show = trace_uprobe_show, 52 .is_busy = trace_uprobe_is_busy, 53 .free = trace_uprobe_release, 54 .match = trace_uprobe_match, 55 }; 56 57 /* 58 * uprobe event core functions 59 */ 60 struct trace_uprobe { 61 struct dyn_event devent; 62 struct trace_uprobe_filter filter; 63 struct uprobe_consumer consumer; 64 struct path path; 65 struct inode *inode; 66 char *filename; 67 unsigned long offset; 68 unsigned long ref_ctr_offset; 69 unsigned long nhit; 70 struct trace_probe tp; 71 }; 72 73 static bool is_trace_uprobe(struct dyn_event *ev) 74 { 75 return ev->ops == &trace_uprobe_ops; 76 } 77 78 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev) 79 { 80 return container_of(ev, struct trace_uprobe, devent); 81 } 82 83 /** 84 * for_each_trace_uprobe - iterate over the trace_uprobe list 85 * @pos: the struct trace_uprobe * for each entry 86 * @dpos: the struct dyn_event * to use as a loop cursor 87 */ 88 #define for_each_trace_uprobe(pos, dpos) \ 89 for_each_dyn_event(dpos) \ 90 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos))) 91 92 #define SIZEOF_TRACE_UPROBE(n) \ 93 (offsetof(struct trace_uprobe, tp.args) + \ 94 (sizeof(struct probe_arg) * (n))) 95 96 static int register_uprobe_event(struct trace_uprobe *tu); 97 static int unregister_uprobe_event(struct trace_uprobe *tu); 98 99 struct uprobe_dispatch_data { 100 struct trace_uprobe *tu; 101 unsigned long bp_addr; 102 }; 103 104 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs); 105 static int uretprobe_dispatcher(struct uprobe_consumer *con, 106 unsigned long func, struct pt_regs *regs); 107 108 #ifdef CONFIG_STACK_GROWSUP 109 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n) 110 { 111 return addr - (n * sizeof(long)); 112 } 113 #else 114 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n) 115 { 116 return addr + (n * sizeof(long)); 117 } 118 #endif 119 120 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n) 121 { 122 unsigned long ret; 123 unsigned long addr = user_stack_pointer(regs); 124 125 addr = adjust_stack_addr(addr, n); 126 127 if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret))) 128 return 0; 129 130 return ret; 131 } 132 133 /* 134 * Uprobes-specific fetch functions 135 */ 136 static nokprobe_inline int 137 probe_mem_read(void *dest, void *src, size_t size) 138 { 139 void __user *vaddr = (void __force __user *)src; 140 141 return copy_from_user(dest, vaddr, size) ? -EFAULT : 0; 142 } 143 /* 144 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max 145 * length and relative data location. 146 */ 147 static nokprobe_inline int 148 fetch_store_string(unsigned long addr, void *dest, void *base) 149 { 150 long ret; 151 u32 loc = *(u32 *)dest; 152 int maxlen = get_loc_len(loc); 153 u8 *dst = get_loc_data(dest, base); 154 void __user *src = (void __force __user *) addr; 155 156 if (unlikely(!maxlen)) 157 return -ENOMEM; 158 159 ret = strncpy_from_user(dst, src, maxlen); 160 if (ret >= 0) { 161 if (ret == maxlen) 162 dst[ret - 1] = '\0'; 163 *(u32 *)dest = make_data_loc(ret, (void *)dst - base); 164 } 165 166 return ret; 167 } 168 169 /* Return the length of string -- including null terminal byte */ 170 static nokprobe_inline int 171 fetch_store_strlen(unsigned long addr) 172 { 173 int len; 174 void __user *vaddr = (void __force __user *) addr; 175 176 len = strnlen_user(vaddr, MAX_STRING_SIZE); 177 178 return (len > MAX_STRING_SIZE) ? 0 : len; 179 } 180 181 static unsigned long translate_user_vaddr(unsigned long file_offset) 182 { 183 unsigned long base_addr; 184 struct uprobe_dispatch_data *udd; 185 186 udd = (void *) current->utask->vaddr; 187 188 base_addr = udd->bp_addr - udd->tu->offset; 189 return base_addr + file_offset; 190 } 191 192 /* Note that we don't verify it, since the code does not come from user space */ 193 static int 194 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest, 195 void *base) 196 { 197 unsigned long val; 198 199 /* 1st stage: get value from context */ 200 switch (code->op) { 201 case FETCH_OP_REG: 202 val = regs_get_register(regs, code->param); 203 break; 204 case FETCH_OP_STACK: 205 val = get_user_stack_nth(regs, code->param); 206 break; 207 case FETCH_OP_STACKP: 208 val = user_stack_pointer(regs); 209 break; 210 case FETCH_OP_RETVAL: 211 val = regs_return_value(regs); 212 break; 213 case FETCH_OP_IMM: 214 val = code->immediate; 215 break; 216 case FETCH_OP_FOFFS: 217 val = translate_user_vaddr(code->immediate); 218 break; 219 default: 220 return -EILSEQ; 221 } 222 code++; 223 224 return process_fetch_insn_bottom(code, val, dest, base); 225 } 226 NOKPROBE_SYMBOL(process_fetch_insn) 227 228 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter) 229 { 230 rwlock_init(&filter->rwlock); 231 filter->nr_systemwide = 0; 232 INIT_LIST_HEAD(&filter->perf_events); 233 } 234 235 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter) 236 { 237 return !filter->nr_systemwide && list_empty(&filter->perf_events); 238 } 239 240 static inline bool is_ret_probe(struct trace_uprobe *tu) 241 { 242 return tu->consumer.ret_handler != NULL; 243 } 244 245 static bool trace_uprobe_is_busy(struct dyn_event *ev) 246 { 247 struct trace_uprobe *tu = to_trace_uprobe(ev); 248 249 return trace_probe_is_enabled(&tu->tp); 250 } 251 252 static bool trace_uprobe_match(const char *system, const char *event, 253 struct dyn_event *ev) 254 { 255 struct trace_uprobe *tu = to_trace_uprobe(ev); 256 257 return strcmp(trace_event_name(&tu->tp.call), event) == 0 && 258 (!system || strcmp(tu->tp.call.class->system, system) == 0); 259 } 260 261 /* 262 * Allocate new trace_uprobe and initialize it (including uprobes). 263 */ 264 static struct trace_uprobe * 265 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret) 266 { 267 struct trace_uprobe *tu; 268 269 if (!event || !is_good_name(event)) 270 return ERR_PTR(-EINVAL); 271 272 if (!group || !is_good_name(group)) 273 return ERR_PTR(-EINVAL); 274 275 tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL); 276 if (!tu) 277 return ERR_PTR(-ENOMEM); 278 279 tu->tp.call.class = &tu->tp.class; 280 tu->tp.call.name = kstrdup(event, GFP_KERNEL); 281 if (!tu->tp.call.name) 282 goto error; 283 284 tu->tp.class.system = kstrdup(group, GFP_KERNEL); 285 if (!tu->tp.class.system) 286 goto error; 287 288 dyn_event_init(&tu->devent, &trace_uprobe_ops); 289 INIT_LIST_HEAD(&tu->tp.files); 290 tu->consumer.handler = uprobe_dispatcher; 291 if (is_ret) 292 tu->consumer.ret_handler = uretprobe_dispatcher; 293 init_trace_uprobe_filter(&tu->filter); 294 return tu; 295 296 error: 297 kfree(tu->tp.call.name); 298 kfree(tu); 299 300 return ERR_PTR(-ENOMEM); 301 } 302 303 static void free_trace_uprobe(struct trace_uprobe *tu) 304 { 305 int i; 306 307 if (!tu) 308 return; 309 310 for (i = 0; i < tu->tp.nr_args; i++) 311 traceprobe_free_probe_arg(&tu->tp.args[i]); 312 313 path_put(&tu->path); 314 kfree(tu->tp.call.class->system); 315 kfree(tu->tp.call.name); 316 kfree(tu->filename); 317 kfree(tu); 318 } 319 320 static struct trace_uprobe *find_probe_event(const char *event, const char *group) 321 { 322 struct dyn_event *pos; 323 struct trace_uprobe *tu; 324 325 for_each_trace_uprobe(tu, pos) 326 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 && 327 strcmp(tu->tp.call.class->system, group) == 0) 328 return tu; 329 330 return NULL; 331 } 332 333 /* Unregister a trace_uprobe and probe_event */ 334 static int unregister_trace_uprobe(struct trace_uprobe *tu) 335 { 336 int ret; 337 338 ret = unregister_uprobe_event(tu); 339 if (ret) 340 return ret; 341 342 dyn_event_remove(&tu->devent); 343 free_trace_uprobe(tu); 344 return 0; 345 } 346 347 /* 348 * Uprobe with multiple reference counter is not allowed. i.e. 349 * If inode and offset matches, reference counter offset *must* 350 * match as well. Though, there is one exception: If user is 351 * replacing old trace_uprobe with new one(same group/event), 352 * then we allow same uprobe with new reference counter as far 353 * as the new one does not conflict with any other existing 354 * ones. 355 */ 356 static struct trace_uprobe *find_old_trace_uprobe(struct trace_uprobe *new) 357 { 358 struct dyn_event *pos; 359 struct trace_uprobe *tmp, *old = NULL; 360 struct inode *new_inode = d_real_inode(new->path.dentry); 361 362 old = find_probe_event(trace_event_name(&new->tp.call), 363 new->tp.call.class->system); 364 365 for_each_trace_uprobe(tmp, pos) { 366 if ((old ? old != tmp : true) && 367 new_inode == d_real_inode(tmp->path.dentry) && 368 new->offset == tmp->offset && 369 new->ref_ctr_offset != tmp->ref_ctr_offset) { 370 pr_warn("Reference counter offset mismatch."); 371 return ERR_PTR(-EINVAL); 372 } 373 } 374 return old; 375 } 376 377 /* Register a trace_uprobe and probe_event */ 378 static int register_trace_uprobe(struct trace_uprobe *tu) 379 { 380 struct trace_uprobe *old_tu; 381 int ret; 382 383 mutex_lock(&event_mutex); 384 385 /* register as an event */ 386 old_tu = find_old_trace_uprobe(tu); 387 if (IS_ERR(old_tu)) { 388 ret = PTR_ERR(old_tu); 389 goto end; 390 } 391 392 if (old_tu) { 393 /* delete old event */ 394 ret = unregister_trace_uprobe(old_tu); 395 if (ret) 396 goto end; 397 } 398 399 ret = register_uprobe_event(tu); 400 if (ret) { 401 pr_warn("Failed to register probe event(%d)\n", ret); 402 goto end; 403 } 404 405 dyn_event_add(&tu->devent); 406 407 end: 408 mutex_unlock(&event_mutex); 409 410 return ret; 411 } 412 413 /* 414 * Argument syntax: 415 * - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] 416 * 417 * - Remove uprobe: -:[GRP/]EVENT 418 */ 419 static int trace_uprobe_create(int argc, const char **argv) 420 { 421 struct trace_uprobe *tu; 422 const char *event = NULL, *group = UPROBE_EVENT_SYSTEM; 423 char *arg, *filename, *rctr, *rctr_end, *tmp; 424 char buf[MAX_EVENT_NAME_LEN]; 425 struct path path; 426 unsigned long offset, ref_ctr_offset; 427 bool is_return = false; 428 int i, ret; 429 430 ret = 0; 431 ref_ctr_offset = 0; 432 433 /* argc must be >= 1 */ 434 if (argv[0][0] == 'r') 435 is_return = true; 436 else if (argv[0][0] != 'p' || argc < 2) 437 return -ECANCELED; 438 439 if (argv[0][1] == ':') 440 event = &argv[0][2]; 441 442 if (!strchr(argv[1], '/')) 443 return -ECANCELED; 444 445 filename = kstrdup(argv[1], GFP_KERNEL); 446 if (!filename) 447 return -ENOMEM; 448 449 /* Find the last occurrence, in case the path contains ':' too. */ 450 arg = strrchr(filename, ':'); 451 if (!arg || !isdigit(arg[1])) { 452 kfree(filename); 453 return -ECANCELED; 454 } 455 456 *arg++ = '\0'; 457 ret = kern_path(filename, LOOKUP_FOLLOW, &path); 458 if (ret) { 459 kfree(filename); 460 return ret; 461 } 462 if (!d_is_reg(path.dentry)) { 463 ret = -EINVAL; 464 goto fail_address_parse; 465 } 466 467 /* Parse reference counter offset if specified. */ 468 rctr = strchr(arg, '('); 469 if (rctr) { 470 rctr_end = strchr(rctr, ')'); 471 if (rctr > rctr_end || *(rctr_end + 1) != 0) { 472 ret = -EINVAL; 473 pr_info("Invalid reference counter offset.\n"); 474 goto fail_address_parse; 475 } 476 477 *rctr++ = '\0'; 478 *rctr_end = '\0'; 479 ret = kstrtoul(rctr, 0, &ref_ctr_offset); 480 if (ret) { 481 pr_info("Invalid reference counter offset.\n"); 482 goto fail_address_parse; 483 } 484 } 485 486 /* Parse uprobe offset. */ 487 ret = kstrtoul(arg, 0, &offset); 488 if (ret) 489 goto fail_address_parse; 490 491 argc -= 2; 492 argv += 2; 493 494 /* setup a probe */ 495 if (event) { 496 ret = traceprobe_parse_event_name(&event, &group, buf); 497 if (ret) 498 goto fail_address_parse; 499 } else { 500 char *tail; 501 char *ptr; 502 503 tail = kstrdup(kbasename(filename), GFP_KERNEL); 504 if (!tail) { 505 ret = -ENOMEM; 506 goto fail_address_parse; 507 } 508 509 ptr = strpbrk(tail, ".-_"); 510 if (ptr) 511 *ptr = '\0'; 512 513 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset); 514 event = buf; 515 kfree(tail); 516 } 517 518 tu = alloc_trace_uprobe(group, event, argc, is_return); 519 if (IS_ERR(tu)) { 520 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu)); 521 ret = PTR_ERR(tu); 522 goto fail_address_parse; 523 } 524 tu->offset = offset; 525 tu->ref_ctr_offset = ref_ctr_offset; 526 tu->path = path; 527 tu->filename = filename; 528 529 /* parse arguments */ 530 for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) { 531 tmp = kstrdup(argv[i], GFP_KERNEL); 532 if (!tmp) { 533 ret = -ENOMEM; 534 goto error; 535 } 536 537 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp, 538 is_return ? TPARG_FL_RETURN : 0); 539 kfree(tmp); 540 if (ret) 541 goto error; 542 } 543 544 ret = register_trace_uprobe(tu); 545 if (ret) 546 goto error; 547 return 0; 548 549 error: 550 free_trace_uprobe(tu); 551 return ret; 552 553 fail_address_parse: 554 path_put(&path); 555 kfree(filename); 556 557 pr_info("Failed to parse address or file.\n"); 558 559 return ret; 560 } 561 562 static int create_or_delete_trace_uprobe(int argc, char **argv) 563 { 564 int ret; 565 566 if (argv[0][0] == '-') 567 return dyn_event_release(argc, argv, &trace_uprobe_ops); 568 569 ret = trace_uprobe_create(argc, (const char **)argv); 570 return ret == -ECANCELED ? -EINVAL : ret; 571 } 572 573 static int trace_uprobe_release(struct dyn_event *ev) 574 { 575 struct trace_uprobe *tu = to_trace_uprobe(ev); 576 577 return unregister_trace_uprobe(tu); 578 } 579 580 /* Probes listing interfaces */ 581 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev) 582 { 583 struct trace_uprobe *tu = to_trace_uprobe(ev); 584 char c = is_ret_probe(tu) ? 'r' : 'p'; 585 int i; 586 587 seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, tu->tp.call.class->system, 588 trace_event_name(&tu->tp.call), tu->filename, 589 (int)(sizeof(void *) * 2), tu->offset); 590 591 if (tu->ref_ctr_offset) 592 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset); 593 594 for (i = 0; i < tu->tp.nr_args; i++) 595 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm); 596 597 seq_putc(m, '\n'); 598 return 0; 599 } 600 601 static int probes_seq_show(struct seq_file *m, void *v) 602 { 603 struct dyn_event *ev = v; 604 605 if (!is_trace_uprobe(ev)) 606 return 0; 607 608 return trace_uprobe_show(m, ev); 609 } 610 611 static const struct seq_operations probes_seq_op = { 612 .start = dyn_event_seq_start, 613 .next = dyn_event_seq_next, 614 .stop = dyn_event_seq_stop, 615 .show = probes_seq_show 616 }; 617 618 static int probes_open(struct inode *inode, struct file *file) 619 { 620 int ret; 621 622 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) { 623 ret = dyn_events_release_all(&trace_uprobe_ops); 624 if (ret) 625 return ret; 626 } 627 628 return seq_open(file, &probes_seq_op); 629 } 630 631 static ssize_t probes_write(struct file *file, const char __user *buffer, 632 size_t count, loff_t *ppos) 633 { 634 return trace_parse_run_command(file, buffer, count, ppos, 635 create_or_delete_trace_uprobe); 636 } 637 638 static const struct file_operations uprobe_events_ops = { 639 .owner = THIS_MODULE, 640 .open = probes_open, 641 .read = seq_read, 642 .llseek = seq_lseek, 643 .release = seq_release, 644 .write = probes_write, 645 }; 646 647 /* Probes profiling interfaces */ 648 static int probes_profile_seq_show(struct seq_file *m, void *v) 649 { 650 struct dyn_event *ev = v; 651 struct trace_uprobe *tu; 652 653 if (!is_trace_uprobe(ev)) 654 return 0; 655 656 tu = to_trace_uprobe(ev); 657 seq_printf(m, " %s %-44s %15lu\n", tu->filename, 658 trace_event_name(&tu->tp.call), tu->nhit); 659 return 0; 660 } 661 662 static const struct seq_operations profile_seq_op = { 663 .start = dyn_event_seq_start, 664 .next = dyn_event_seq_next, 665 .stop = dyn_event_seq_stop, 666 .show = probes_profile_seq_show 667 }; 668 669 static int profile_open(struct inode *inode, struct file *file) 670 { 671 return seq_open(file, &profile_seq_op); 672 } 673 674 static const struct file_operations uprobe_profile_ops = { 675 .owner = THIS_MODULE, 676 .open = profile_open, 677 .read = seq_read, 678 .llseek = seq_lseek, 679 .release = seq_release, 680 }; 681 682 struct uprobe_cpu_buffer { 683 struct mutex mutex; 684 void *buf; 685 }; 686 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer; 687 static int uprobe_buffer_refcnt; 688 689 static int uprobe_buffer_init(void) 690 { 691 int cpu, err_cpu; 692 693 uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer); 694 if (uprobe_cpu_buffer == NULL) 695 return -ENOMEM; 696 697 for_each_possible_cpu(cpu) { 698 struct page *p = alloc_pages_node(cpu_to_node(cpu), 699 GFP_KERNEL, 0); 700 if (p == NULL) { 701 err_cpu = cpu; 702 goto err; 703 } 704 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p); 705 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex); 706 } 707 708 return 0; 709 710 err: 711 for_each_possible_cpu(cpu) { 712 if (cpu == err_cpu) 713 break; 714 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf); 715 } 716 717 free_percpu(uprobe_cpu_buffer); 718 return -ENOMEM; 719 } 720 721 static int uprobe_buffer_enable(void) 722 { 723 int ret = 0; 724 725 BUG_ON(!mutex_is_locked(&event_mutex)); 726 727 if (uprobe_buffer_refcnt++ == 0) { 728 ret = uprobe_buffer_init(); 729 if (ret < 0) 730 uprobe_buffer_refcnt--; 731 } 732 733 return ret; 734 } 735 736 static void uprobe_buffer_disable(void) 737 { 738 int cpu; 739 740 BUG_ON(!mutex_is_locked(&event_mutex)); 741 742 if (--uprobe_buffer_refcnt == 0) { 743 for_each_possible_cpu(cpu) 744 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, 745 cpu)->buf); 746 747 free_percpu(uprobe_cpu_buffer); 748 uprobe_cpu_buffer = NULL; 749 } 750 } 751 752 static struct uprobe_cpu_buffer *uprobe_buffer_get(void) 753 { 754 struct uprobe_cpu_buffer *ucb; 755 int cpu; 756 757 cpu = raw_smp_processor_id(); 758 ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu); 759 760 /* 761 * Use per-cpu buffers for fastest access, but we might migrate 762 * so the mutex makes sure we have sole access to it. 763 */ 764 mutex_lock(&ucb->mutex); 765 766 return ucb; 767 } 768 769 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb) 770 { 771 mutex_unlock(&ucb->mutex); 772 } 773 774 static void __uprobe_trace_func(struct trace_uprobe *tu, 775 unsigned long func, struct pt_regs *regs, 776 struct uprobe_cpu_buffer *ucb, int dsize, 777 struct trace_event_file *trace_file) 778 { 779 struct uprobe_trace_entry_head *entry; 780 struct ring_buffer_event *event; 781 struct ring_buffer *buffer; 782 void *data; 783 int size, esize; 784 struct trace_event_call *call = &tu->tp.call; 785 786 WARN_ON(call != trace_file->event_call); 787 788 if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE)) 789 return; 790 791 if (trace_trigger_soft_disabled(trace_file)) 792 return; 793 794 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 795 size = esize + tu->tp.size + dsize; 796 event = trace_event_buffer_lock_reserve(&buffer, trace_file, 797 call->event.type, size, 0, 0); 798 if (!event) 799 return; 800 801 entry = ring_buffer_event_data(event); 802 if (is_ret_probe(tu)) { 803 entry->vaddr[0] = func; 804 entry->vaddr[1] = instruction_pointer(regs); 805 data = DATAOF_TRACE_ENTRY(entry, true); 806 } else { 807 entry->vaddr[0] = instruction_pointer(regs); 808 data = DATAOF_TRACE_ENTRY(entry, false); 809 } 810 811 memcpy(data, ucb->buf, tu->tp.size + dsize); 812 813 event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0); 814 } 815 816 /* uprobe handler */ 817 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs, 818 struct uprobe_cpu_buffer *ucb, int dsize) 819 { 820 struct event_file_link *link; 821 822 if (is_ret_probe(tu)) 823 return 0; 824 825 rcu_read_lock(); 826 list_for_each_entry_rcu(link, &tu->tp.files, list) 827 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file); 828 rcu_read_unlock(); 829 830 return 0; 831 } 832 833 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func, 834 struct pt_regs *regs, 835 struct uprobe_cpu_buffer *ucb, int dsize) 836 { 837 struct event_file_link *link; 838 839 rcu_read_lock(); 840 list_for_each_entry_rcu(link, &tu->tp.files, list) 841 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file); 842 rcu_read_unlock(); 843 } 844 845 /* Event entry printers */ 846 static enum print_line_t 847 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event) 848 { 849 struct uprobe_trace_entry_head *entry; 850 struct trace_seq *s = &iter->seq; 851 struct trace_uprobe *tu; 852 u8 *data; 853 854 entry = (struct uprobe_trace_entry_head *)iter->ent; 855 tu = container_of(event, struct trace_uprobe, tp.call.event); 856 857 if (is_ret_probe(tu)) { 858 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)", 859 trace_event_name(&tu->tp.call), 860 entry->vaddr[1], entry->vaddr[0]); 861 data = DATAOF_TRACE_ENTRY(entry, true); 862 } else { 863 trace_seq_printf(s, "%s: (0x%lx)", 864 trace_event_name(&tu->tp.call), 865 entry->vaddr[0]); 866 data = DATAOF_TRACE_ENTRY(entry, false); 867 } 868 869 if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0) 870 goto out; 871 872 trace_seq_putc(s, '\n'); 873 874 out: 875 return trace_handle_return(s); 876 } 877 878 typedef bool (*filter_func_t)(struct uprobe_consumer *self, 879 enum uprobe_filter_ctx ctx, 880 struct mm_struct *mm); 881 882 static int 883 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file, 884 filter_func_t filter) 885 { 886 bool enabled = trace_probe_is_enabled(&tu->tp); 887 struct event_file_link *link = NULL; 888 int ret; 889 890 if (file) { 891 if (tu->tp.flags & TP_FLAG_PROFILE) 892 return -EINTR; 893 894 link = kmalloc(sizeof(*link), GFP_KERNEL); 895 if (!link) 896 return -ENOMEM; 897 898 link->file = file; 899 list_add_tail_rcu(&link->list, &tu->tp.files); 900 901 tu->tp.flags |= TP_FLAG_TRACE; 902 } else { 903 if (tu->tp.flags & TP_FLAG_TRACE) 904 return -EINTR; 905 906 tu->tp.flags |= TP_FLAG_PROFILE; 907 } 908 909 WARN_ON(!uprobe_filter_is_empty(&tu->filter)); 910 911 if (enabled) 912 return 0; 913 914 ret = uprobe_buffer_enable(); 915 if (ret) 916 goto err_flags; 917 918 tu->consumer.filter = filter; 919 tu->inode = d_real_inode(tu->path.dentry); 920 if (tu->ref_ctr_offset) { 921 ret = uprobe_register_refctr(tu->inode, tu->offset, 922 tu->ref_ctr_offset, &tu->consumer); 923 } else { 924 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer); 925 } 926 927 if (ret) 928 goto err_buffer; 929 930 return 0; 931 932 err_buffer: 933 uprobe_buffer_disable(); 934 935 err_flags: 936 if (file) { 937 list_del(&link->list); 938 kfree(link); 939 tu->tp.flags &= ~TP_FLAG_TRACE; 940 } else { 941 tu->tp.flags &= ~TP_FLAG_PROFILE; 942 } 943 return ret; 944 } 945 946 static void 947 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file) 948 { 949 if (!trace_probe_is_enabled(&tu->tp)) 950 return; 951 952 if (file) { 953 struct event_file_link *link; 954 955 link = find_event_file_link(&tu->tp, file); 956 if (!link) 957 return; 958 959 list_del_rcu(&link->list); 960 /* synchronize with u{,ret}probe_trace_func */ 961 synchronize_rcu(); 962 kfree(link); 963 964 if (!list_empty(&tu->tp.files)) 965 return; 966 } 967 968 WARN_ON(!uprobe_filter_is_empty(&tu->filter)); 969 970 uprobe_unregister(tu->inode, tu->offset, &tu->consumer); 971 tu->inode = NULL; 972 tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE; 973 974 uprobe_buffer_disable(); 975 } 976 977 static int uprobe_event_define_fields(struct trace_event_call *event_call) 978 { 979 int ret, size; 980 struct uprobe_trace_entry_head field; 981 struct trace_uprobe *tu = event_call->data; 982 983 if (is_ret_probe(tu)) { 984 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0); 985 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0); 986 size = SIZEOF_TRACE_ENTRY(true); 987 } else { 988 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0); 989 size = SIZEOF_TRACE_ENTRY(false); 990 } 991 992 return traceprobe_define_arg_fields(event_call, size, &tu->tp); 993 } 994 995 #ifdef CONFIG_PERF_EVENTS 996 static bool 997 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm) 998 { 999 struct perf_event *event; 1000 1001 if (filter->nr_systemwide) 1002 return true; 1003 1004 list_for_each_entry(event, &filter->perf_events, hw.tp_list) { 1005 if (event->hw.target->mm == mm) 1006 return true; 1007 } 1008 1009 return false; 1010 } 1011 1012 static inline bool 1013 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) 1014 { 1015 return __uprobe_perf_filter(&tu->filter, event->hw.target->mm); 1016 } 1017 1018 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event) 1019 { 1020 bool done; 1021 1022 write_lock(&tu->filter.rwlock); 1023 if (event->hw.target) { 1024 list_del(&event->hw.tp_list); 1025 done = tu->filter.nr_systemwide || 1026 (event->hw.target->flags & PF_EXITING) || 1027 uprobe_filter_event(tu, event); 1028 } else { 1029 tu->filter.nr_systemwide--; 1030 done = tu->filter.nr_systemwide; 1031 } 1032 write_unlock(&tu->filter.rwlock); 1033 1034 if (!done) 1035 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false); 1036 1037 return 0; 1038 } 1039 1040 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event) 1041 { 1042 bool done; 1043 int err; 1044 1045 write_lock(&tu->filter.rwlock); 1046 if (event->hw.target) { 1047 /* 1048 * event->parent != NULL means copy_process(), we can avoid 1049 * uprobe_apply(). current->mm must be probed and we can rely 1050 * on dup_mmap() which preserves the already installed bp's. 1051 * 1052 * attr.enable_on_exec means that exec/mmap will install the 1053 * breakpoints we need. 1054 */ 1055 done = tu->filter.nr_systemwide || 1056 event->parent || event->attr.enable_on_exec || 1057 uprobe_filter_event(tu, event); 1058 list_add(&event->hw.tp_list, &tu->filter.perf_events); 1059 } else { 1060 done = tu->filter.nr_systemwide; 1061 tu->filter.nr_systemwide++; 1062 } 1063 write_unlock(&tu->filter.rwlock); 1064 1065 err = 0; 1066 if (!done) { 1067 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true); 1068 if (err) 1069 uprobe_perf_close(tu, event); 1070 } 1071 return err; 1072 } 1073 1074 static bool uprobe_perf_filter(struct uprobe_consumer *uc, 1075 enum uprobe_filter_ctx ctx, struct mm_struct *mm) 1076 { 1077 struct trace_uprobe *tu; 1078 int ret; 1079 1080 tu = container_of(uc, struct trace_uprobe, consumer); 1081 read_lock(&tu->filter.rwlock); 1082 ret = __uprobe_perf_filter(&tu->filter, mm); 1083 read_unlock(&tu->filter.rwlock); 1084 1085 return ret; 1086 } 1087 1088 static void __uprobe_perf_func(struct trace_uprobe *tu, 1089 unsigned long func, struct pt_regs *regs, 1090 struct uprobe_cpu_buffer *ucb, int dsize) 1091 { 1092 struct trace_event_call *call = &tu->tp.call; 1093 struct uprobe_trace_entry_head *entry; 1094 struct hlist_head *head; 1095 void *data; 1096 int size, esize; 1097 int rctx; 1098 1099 if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs)) 1100 return; 1101 1102 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 1103 1104 size = esize + tu->tp.size + dsize; 1105 size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32); 1106 if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough")) 1107 return; 1108 1109 preempt_disable(); 1110 head = this_cpu_ptr(call->perf_events); 1111 if (hlist_empty(head)) 1112 goto out; 1113 1114 entry = perf_trace_buf_alloc(size, NULL, &rctx); 1115 if (!entry) 1116 goto out; 1117 1118 if (is_ret_probe(tu)) { 1119 entry->vaddr[0] = func; 1120 entry->vaddr[1] = instruction_pointer(regs); 1121 data = DATAOF_TRACE_ENTRY(entry, true); 1122 } else { 1123 entry->vaddr[0] = instruction_pointer(regs); 1124 data = DATAOF_TRACE_ENTRY(entry, false); 1125 } 1126 1127 memcpy(data, ucb->buf, tu->tp.size + dsize); 1128 1129 if (size - esize > tu->tp.size + dsize) { 1130 int len = tu->tp.size + dsize; 1131 1132 memset(data + len, 0, size - esize - len); 1133 } 1134 1135 perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs, 1136 head, NULL); 1137 out: 1138 preempt_enable(); 1139 } 1140 1141 /* uprobe profile handler */ 1142 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs, 1143 struct uprobe_cpu_buffer *ucb, int dsize) 1144 { 1145 if (!uprobe_perf_filter(&tu->consumer, 0, current->mm)) 1146 return UPROBE_HANDLER_REMOVE; 1147 1148 if (!is_ret_probe(tu)) 1149 __uprobe_perf_func(tu, 0, regs, ucb, dsize); 1150 return 0; 1151 } 1152 1153 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func, 1154 struct pt_regs *regs, 1155 struct uprobe_cpu_buffer *ucb, int dsize) 1156 { 1157 __uprobe_perf_func(tu, func, regs, ucb, dsize); 1158 } 1159 1160 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type, 1161 const char **filename, u64 *probe_offset, 1162 bool perf_type_tracepoint) 1163 { 1164 const char *pevent = trace_event_name(event->tp_event); 1165 const char *group = event->tp_event->class->system; 1166 struct trace_uprobe *tu; 1167 1168 if (perf_type_tracepoint) 1169 tu = find_probe_event(pevent, group); 1170 else 1171 tu = event->tp_event->data; 1172 if (!tu) 1173 return -EINVAL; 1174 1175 *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE 1176 : BPF_FD_TYPE_UPROBE; 1177 *filename = tu->filename; 1178 *probe_offset = tu->offset; 1179 return 0; 1180 } 1181 #endif /* CONFIG_PERF_EVENTS */ 1182 1183 static int 1184 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type, 1185 void *data) 1186 { 1187 struct trace_uprobe *tu = event->data; 1188 struct trace_event_file *file = data; 1189 1190 switch (type) { 1191 case TRACE_REG_REGISTER: 1192 return probe_event_enable(tu, file, NULL); 1193 1194 case TRACE_REG_UNREGISTER: 1195 probe_event_disable(tu, file); 1196 return 0; 1197 1198 #ifdef CONFIG_PERF_EVENTS 1199 case TRACE_REG_PERF_REGISTER: 1200 return probe_event_enable(tu, NULL, uprobe_perf_filter); 1201 1202 case TRACE_REG_PERF_UNREGISTER: 1203 probe_event_disable(tu, NULL); 1204 return 0; 1205 1206 case TRACE_REG_PERF_OPEN: 1207 return uprobe_perf_open(tu, data); 1208 1209 case TRACE_REG_PERF_CLOSE: 1210 return uprobe_perf_close(tu, data); 1211 1212 #endif 1213 default: 1214 return 0; 1215 } 1216 return 0; 1217 } 1218 1219 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs) 1220 { 1221 struct trace_uprobe *tu; 1222 struct uprobe_dispatch_data udd; 1223 struct uprobe_cpu_buffer *ucb; 1224 int dsize, esize; 1225 int ret = 0; 1226 1227 1228 tu = container_of(con, struct trace_uprobe, consumer); 1229 tu->nhit++; 1230 1231 udd.tu = tu; 1232 udd.bp_addr = instruction_pointer(regs); 1233 1234 current->utask->vaddr = (unsigned long) &udd; 1235 1236 if (WARN_ON_ONCE(!uprobe_cpu_buffer)) 1237 return 0; 1238 1239 dsize = __get_data_size(&tu->tp, regs); 1240 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 1241 1242 ucb = uprobe_buffer_get(); 1243 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize); 1244 1245 if (tu->tp.flags & TP_FLAG_TRACE) 1246 ret |= uprobe_trace_func(tu, regs, ucb, dsize); 1247 1248 #ifdef CONFIG_PERF_EVENTS 1249 if (tu->tp.flags & TP_FLAG_PROFILE) 1250 ret |= uprobe_perf_func(tu, regs, ucb, dsize); 1251 #endif 1252 uprobe_buffer_put(ucb); 1253 return ret; 1254 } 1255 1256 static int uretprobe_dispatcher(struct uprobe_consumer *con, 1257 unsigned long func, struct pt_regs *regs) 1258 { 1259 struct trace_uprobe *tu; 1260 struct uprobe_dispatch_data udd; 1261 struct uprobe_cpu_buffer *ucb; 1262 int dsize, esize; 1263 1264 tu = container_of(con, struct trace_uprobe, consumer); 1265 1266 udd.tu = tu; 1267 udd.bp_addr = func; 1268 1269 current->utask->vaddr = (unsigned long) &udd; 1270 1271 if (WARN_ON_ONCE(!uprobe_cpu_buffer)) 1272 return 0; 1273 1274 dsize = __get_data_size(&tu->tp, regs); 1275 esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu)); 1276 1277 ucb = uprobe_buffer_get(); 1278 store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize); 1279 1280 if (tu->tp.flags & TP_FLAG_TRACE) 1281 uretprobe_trace_func(tu, func, regs, ucb, dsize); 1282 1283 #ifdef CONFIG_PERF_EVENTS 1284 if (tu->tp.flags & TP_FLAG_PROFILE) 1285 uretprobe_perf_func(tu, func, regs, ucb, dsize); 1286 #endif 1287 uprobe_buffer_put(ucb); 1288 return 0; 1289 } 1290 1291 static struct trace_event_functions uprobe_funcs = { 1292 .trace = print_uprobe_event 1293 }; 1294 1295 static inline void init_trace_event_call(struct trace_uprobe *tu, 1296 struct trace_event_call *call) 1297 { 1298 INIT_LIST_HEAD(&call->class->fields); 1299 call->event.funcs = &uprobe_funcs; 1300 call->class->define_fields = uprobe_event_define_fields; 1301 1302 call->flags = TRACE_EVENT_FL_UPROBE; 1303 call->class->reg = trace_uprobe_register; 1304 call->data = tu; 1305 } 1306 1307 static int register_uprobe_event(struct trace_uprobe *tu) 1308 { 1309 struct trace_event_call *call = &tu->tp.call; 1310 int ret = 0; 1311 1312 init_trace_event_call(tu, call); 1313 1314 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) 1315 return -ENOMEM; 1316 1317 ret = register_trace_event(&call->event); 1318 if (!ret) { 1319 kfree(call->print_fmt); 1320 return -ENODEV; 1321 } 1322 1323 ret = trace_add_event_call(call); 1324 1325 if (ret) { 1326 pr_info("Failed to register uprobe event: %s\n", 1327 trace_event_name(call)); 1328 kfree(call->print_fmt); 1329 unregister_trace_event(&call->event); 1330 } 1331 1332 return ret; 1333 } 1334 1335 static int unregister_uprobe_event(struct trace_uprobe *tu) 1336 { 1337 int ret; 1338 1339 /* tu->event is unregistered in trace_remove_event_call() */ 1340 ret = trace_remove_event_call(&tu->tp.call); 1341 if (ret) 1342 return ret; 1343 kfree(tu->tp.call.print_fmt); 1344 tu->tp.call.print_fmt = NULL; 1345 return 0; 1346 } 1347 1348 #ifdef CONFIG_PERF_EVENTS 1349 struct trace_event_call * 1350 create_local_trace_uprobe(char *name, unsigned long offs, 1351 unsigned long ref_ctr_offset, bool is_return) 1352 { 1353 struct trace_uprobe *tu; 1354 struct path path; 1355 int ret; 1356 1357 ret = kern_path(name, LOOKUP_FOLLOW, &path); 1358 if (ret) 1359 return ERR_PTR(ret); 1360 1361 if (!d_is_reg(path.dentry)) { 1362 path_put(&path); 1363 return ERR_PTR(-EINVAL); 1364 } 1365 1366 /* 1367 * local trace_kprobes are not added to dyn_event, so they are never 1368 * searched in find_trace_kprobe(). Therefore, there is no concern of 1369 * duplicated name "DUMMY_EVENT" here. 1370 */ 1371 tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0, 1372 is_return); 1373 1374 if (IS_ERR(tu)) { 1375 pr_info("Failed to allocate trace_uprobe.(%d)\n", 1376 (int)PTR_ERR(tu)); 1377 path_put(&path); 1378 return ERR_CAST(tu); 1379 } 1380 1381 tu->offset = offs; 1382 tu->path = path; 1383 tu->ref_ctr_offset = ref_ctr_offset; 1384 tu->filename = kstrdup(name, GFP_KERNEL); 1385 init_trace_event_call(tu, &tu->tp.call); 1386 1387 if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) { 1388 ret = -ENOMEM; 1389 goto error; 1390 } 1391 1392 return &tu->tp.call; 1393 error: 1394 free_trace_uprobe(tu); 1395 return ERR_PTR(ret); 1396 } 1397 1398 void destroy_local_trace_uprobe(struct trace_event_call *event_call) 1399 { 1400 struct trace_uprobe *tu; 1401 1402 tu = container_of(event_call, struct trace_uprobe, tp.call); 1403 1404 kfree(tu->tp.call.print_fmt); 1405 tu->tp.call.print_fmt = NULL; 1406 1407 free_trace_uprobe(tu); 1408 } 1409 #endif /* CONFIG_PERF_EVENTS */ 1410 1411 /* Make a trace interface for controling probe points */ 1412 static __init int init_uprobe_trace(void) 1413 { 1414 struct dentry *d_tracer; 1415 int ret; 1416 1417 ret = dyn_event_register(&trace_uprobe_ops); 1418 if (ret) 1419 return ret; 1420 1421 d_tracer = tracing_init_dentry(); 1422 if (IS_ERR(d_tracer)) 1423 return 0; 1424 1425 trace_create_file("uprobe_events", 0644, d_tracer, 1426 NULL, &uprobe_events_ops); 1427 /* Profile interface */ 1428 trace_create_file("uprobe_profile", 0444, d_tracer, 1429 NULL, &uprobe_profile_ops); 1430 return 0; 1431 } 1432 1433 fs_initcall(init_uprobe_trace); 1434