Linux Audio

Check our new training course

Loading...
v6.2
   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
  27struct 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
  39static int trace_uprobe_create(const char *raw_command);
  40static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  41static int trace_uprobe_release(struct dyn_event *ev);
  42static bool trace_uprobe_is_busy(struct dyn_event *ev);
  43static bool trace_uprobe_match(const char *system, const char *event,
  44			int argc, const char **argv, struct dyn_event *ev);
  45
  46static 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 */
  57struct 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
  69static bool is_trace_uprobe(struct dyn_event *ev)
  70{
  71	return ev->ops == &trace_uprobe_ops;
  72}
  73
  74static 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
 
 
 
 
  88static int register_uprobe_event(struct trace_uprobe *tu);
  89static int unregister_uprobe_event(struct trace_uprobe *tu);
  90
  91struct uprobe_dispatch_data {
  92	struct trace_uprobe	*tu;
  93	unsigned long		bp_addr;
  94};
  95
  96static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  97static int uretprobe_dispatcher(struct uprobe_consumer *con,
  98				unsigned long func, struct pt_regs *regs);
  99
 100#ifdef CONFIG_STACK_GROWSUP
 101static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 102{
 103	return addr - (n * sizeof(long));
 104}
 105#else
 106static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 107{
 108	return addr + (n * sizeof(long));
 109}
 110#endif
 111
 112static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 113{
 114	unsigned long ret;
 115	unsigned long addr = user_stack_pointer(regs);
 116
 117	addr = adjust_stack_addr(addr, n);
 118
 119	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 120		return 0;
 121
 122	return ret;
 123}
 124
 125/*
 126 * Uprobes-specific fetch functions
 127 */
 128static nokprobe_inline int
 129probe_mem_read(void *dest, void *src, size_t size)
 130{
 131	void __user *vaddr = (void __force __user *)src;
 132
 133	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 134}
 135
 136static nokprobe_inline int
 137probe_mem_read_user(void *dest, void *src, size_t size)
 138{
 139	return probe_mem_read(dest, src, size);
 140}
 141
 142/*
 143 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 144 * length and relative data location.
 145 */
 146static nokprobe_inline int
 147fetch_store_string(unsigned long addr, void *dest, void *base)
 148{
 149	long ret;
 150	u32 loc = *(u32 *)dest;
 151	int maxlen  = get_loc_len(loc);
 152	u8 *dst = get_loc_data(dest, base);
 153	void __user *src = (void __force __user *) addr;
 154
 155	if (unlikely(!maxlen))
 156		return -ENOMEM;
 157
 158	if (addr == FETCH_TOKEN_COMM)
 159		ret = strlcpy(dst, current->comm, maxlen);
 160	else
 161		ret = strncpy_from_user(dst, src, maxlen);
 162	if (ret >= 0) {
 163		if (ret == maxlen)
 164			dst[ret - 1] = '\0';
 165		else
 166			/*
 167			 * Include the terminating null byte. In this case it
 168			 * was copied by strncpy_from_user but not accounted
 169			 * for in ret.
 170			 */
 171			ret++;
 172		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 173	}
 174
 175	return ret;
 176}
 177
 178static nokprobe_inline int
 179fetch_store_string_user(unsigned long addr, void *dest, void *base)
 180{
 181	return fetch_store_string(addr, dest, base);
 182}
 183
 184/* Return the length of string -- including null terminal byte */
 185static nokprobe_inline int
 186fetch_store_strlen(unsigned long addr)
 187{
 188	int len;
 189	void __user *vaddr = (void __force __user *) addr;
 190
 191	if (addr == FETCH_TOKEN_COMM)
 192		len = strlen(current->comm) + 1;
 193	else
 194		len = strnlen_user(vaddr, MAX_STRING_SIZE);
 195
 196	return (len > MAX_STRING_SIZE) ? 0 : len;
 197}
 198
 199static nokprobe_inline int
 200fetch_store_strlen_user(unsigned long addr)
 201{
 202	return fetch_store_strlen(addr);
 203}
 204
 205static unsigned long translate_user_vaddr(unsigned long file_offset)
 206{
 207	unsigned long base_addr;
 208	struct uprobe_dispatch_data *udd;
 209
 210	udd = (void *) current->utask->vaddr;
 211
 212	base_addr = udd->bp_addr - udd->tu->offset;
 213	return base_addr + file_offset;
 214}
 215
 216/* Note that we don't verify it, since the code does not come from user space */
 217static int
 218process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 219		   void *base)
 220{
 221	struct pt_regs *regs = rec;
 222	unsigned long val;
 223
 224	/* 1st stage: get value from context */
 225	switch (code->op) {
 226	case FETCH_OP_REG:
 227		val = regs_get_register(regs, code->param);
 228		break;
 229	case FETCH_OP_STACK:
 230		val = get_user_stack_nth(regs, code->param);
 231		break;
 232	case FETCH_OP_STACKP:
 233		val = user_stack_pointer(regs);
 234		break;
 235	case FETCH_OP_RETVAL:
 236		val = regs_return_value(regs);
 237		break;
 238	case FETCH_OP_IMM:
 239		val = code->immediate;
 240		break;
 241	case FETCH_OP_COMM:
 242		val = FETCH_TOKEN_COMM;
 243		break;
 244	case FETCH_OP_DATA:
 245		val = (unsigned long)code->data;
 246		break;
 247	case FETCH_OP_FOFFS:
 248		val = translate_user_vaddr(code->immediate);
 249		break;
 250	default:
 251		return -EILSEQ;
 252	}
 253	code++;
 254
 255	return process_fetch_insn_bottom(code, val, dest, base);
 256}
 257NOKPROBE_SYMBOL(process_fetch_insn)
 258
 259static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 260{
 261	rwlock_init(&filter->rwlock);
 262	filter->nr_systemwide = 0;
 263	INIT_LIST_HEAD(&filter->perf_events);
 264}
 265
 266static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 267{
 268	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 269}
 270
 271static inline bool is_ret_probe(struct trace_uprobe *tu)
 272{
 273	return tu->consumer.ret_handler != NULL;
 274}
 275
 276static bool trace_uprobe_is_busy(struct dyn_event *ev)
 277{
 278	struct trace_uprobe *tu = to_trace_uprobe(ev);
 279
 280	return trace_probe_is_enabled(&tu->tp);
 281}
 282
 283static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 284					    int argc, const char **argv)
 285{
 286	char buf[MAX_ARGSTR_LEN + 1];
 287	int len;
 288
 289	if (!argc)
 290		return true;
 291
 292	len = strlen(tu->filename);
 293	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 294		return false;
 295
 296	if (tu->ref_ctr_offset == 0)
 297		snprintf(buf, sizeof(buf), "0x%0*lx",
 298				(int)(sizeof(void *) * 2), tu->offset);
 299	else
 300		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 301				(int)(sizeof(void *) * 2), tu->offset,
 302				tu->ref_ctr_offset);
 303	if (strcmp(buf, &argv[0][len + 1]))
 304		return false;
 305
 306	argc--; argv++;
 307
 308	return trace_probe_match_command_args(&tu->tp, argc, argv);
 309}
 310
 311static bool trace_uprobe_match(const char *system, const char *event,
 312			int argc, const char **argv, struct dyn_event *ev)
 313{
 314	struct trace_uprobe *tu = to_trace_uprobe(ev);
 315
 316	return (event[0] == '\0' ||
 317		strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 318	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 319	   trace_uprobe_match_command_head(tu, argc, argv);
 320}
 321
 322static nokprobe_inline struct trace_uprobe *
 323trace_uprobe_primary_from_call(struct trace_event_call *call)
 324{
 325	struct trace_probe *tp;
 326
 327	tp = trace_probe_primary_from_call(call);
 328	if (WARN_ON_ONCE(!tp))
 329		return NULL;
 330
 331	return container_of(tp, struct trace_uprobe, tp);
 332}
 333
 334/*
 335 * Allocate new trace_uprobe and initialize it (including uprobes).
 336 */
 337static struct trace_uprobe *
 338alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 339{
 340	struct trace_uprobe *tu;
 341	int ret;
 342
 343	tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 344	if (!tu)
 345		return ERR_PTR(-ENOMEM);
 346
 347	ret = trace_probe_init(&tu->tp, event, group, true);
 348	if (ret < 0)
 349		goto error;
 350
 351	dyn_event_init(&tu->devent, &trace_uprobe_ops);
 352	tu->consumer.handler = uprobe_dispatcher;
 353	if (is_ret)
 354		tu->consumer.ret_handler = uretprobe_dispatcher;
 355	init_trace_uprobe_filter(tu->tp.event->filter);
 356	return tu;
 357
 358error:
 359	kfree(tu);
 360
 361	return ERR_PTR(ret);
 362}
 363
 364static void free_trace_uprobe(struct trace_uprobe *tu)
 365{
 366	if (!tu)
 367		return;
 368
 369	path_put(&tu->path);
 370	trace_probe_cleanup(&tu->tp);
 371	kfree(tu->filename);
 372	kfree(tu);
 373}
 374
 375static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 376{
 377	struct dyn_event *pos;
 378	struct trace_uprobe *tu;
 379
 380	for_each_trace_uprobe(tu, pos)
 381		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 382		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 383			return tu;
 384
 385	return NULL;
 386}
 387
 388/* Unregister a trace_uprobe and probe_event */
 389static int unregister_trace_uprobe(struct trace_uprobe *tu)
 390{
 391	int ret;
 392
 393	if (trace_probe_has_sibling(&tu->tp))
 394		goto unreg;
 395
 396	/* If there's a reference to the dynamic event */
 397	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 398		return -EBUSY;
 399
 400	ret = unregister_uprobe_event(tu);
 401	if (ret)
 402		return ret;
 403
 404unreg:
 405	dyn_event_remove(&tu->devent);
 406	trace_probe_unlink(&tu->tp);
 407	free_trace_uprobe(tu);
 408	return 0;
 409}
 410
 411static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 412					 struct trace_uprobe *comp)
 413{
 414	struct trace_probe_event *tpe = orig->tp.event;
 
 415	struct inode *comp_inode = d_real_inode(comp->path.dentry);
 416	int i;
 417
 418	list_for_each_entry(orig, &tpe->probes, tp.list) {
 
 419		if (comp_inode != d_real_inode(orig->path.dentry) ||
 420		    comp->offset != orig->offset)
 421			continue;
 422
 423		/*
 424		 * trace_probe_compare_arg_type() ensured that nr_args and
 425		 * each argument name and type are same. Let's compare comm.
 426		 */
 427		for (i = 0; i < orig->tp.nr_args; i++) {
 428			if (strcmp(orig->tp.args[i].comm,
 429				   comp->tp.args[i].comm))
 430				break;
 431		}
 432
 433		if (i == orig->tp.nr_args)
 434			return true;
 435	}
 436
 437	return false;
 438}
 439
 440static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 441{
 442	int ret;
 443
 444	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 445	if (ret) {
 446		/* Note that argument starts index = 2 */
 447		trace_probe_log_set_index(ret + 1);
 448		trace_probe_log_err(0, DIFF_ARG_TYPE);
 449		return -EEXIST;
 450	}
 451	if (trace_uprobe_has_same_uprobe(to, tu)) {
 452		trace_probe_log_set_index(0);
 453		trace_probe_log_err(0, SAME_PROBE);
 454		return -EEXIST;
 455	}
 456
 457	/* Append to existing event */
 458	ret = trace_probe_append(&tu->tp, &to->tp);
 459	if (!ret)
 460		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 461
 462	return ret;
 463}
 464
 465/*
 466 * Uprobe with multiple reference counter is not allowed. i.e.
 467 * If inode and offset matches, reference counter offset *must*
 468 * match as well. Though, there is one exception: If user is
 469 * replacing old trace_uprobe with new one(same group/event),
 470 * then we allow same uprobe with new reference counter as far
 471 * as the new one does not conflict with any other existing
 472 * ones.
 473 */
 474static int validate_ref_ctr_offset(struct trace_uprobe *new)
 475{
 476	struct dyn_event *pos;
 477	struct trace_uprobe *tmp;
 478	struct inode *new_inode = d_real_inode(new->path.dentry);
 479
 480	for_each_trace_uprobe(tmp, pos) {
 481		if (new_inode == d_real_inode(tmp->path.dentry) &&
 482		    new->offset == tmp->offset &&
 483		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
 484			pr_warn("Reference counter offset mismatch.");
 485			return -EINVAL;
 486		}
 487	}
 488	return 0;
 489}
 490
 491/* Register a trace_uprobe and probe_event */
 492static int register_trace_uprobe(struct trace_uprobe *tu)
 493{
 494	struct trace_uprobe *old_tu;
 495	int ret;
 496
 497	mutex_lock(&event_mutex);
 498
 499	ret = validate_ref_ctr_offset(tu);
 500	if (ret)
 501		goto end;
 502
 503	/* register as an event */
 504	old_tu = find_probe_event(trace_probe_name(&tu->tp),
 505				  trace_probe_group_name(&tu->tp));
 506	if (old_tu) {
 507		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 508			trace_probe_log_set_index(0);
 509			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 510			ret = -EEXIST;
 511		} else {
 512			ret = append_trace_uprobe(tu, old_tu);
 513		}
 514		goto end;
 515	}
 516
 517	ret = register_uprobe_event(tu);
 518	if (ret) {
 519		if (ret == -EEXIST) {
 520			trace_probe_log_set_index(0);
 521			trace_probe_log_err(0, EVENT_EXIST);
 522		} else
 523			pr_warn("Failed to register probe event(%d)\n", ret);
 524		goto end;
 525	}
 526
 527	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 528
 529end:
 530	mutex_unlock(&event_mutex);
 531
 532	return ret;
 533}
 534
 535/*
 536 * Argument syntax:
 537 *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 538 */
 539static int __trace_uprobe_create(int argc, const char **argv)
 540{
 541	struct trace_uprobe *tu;
 542	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 543	char *arg, *filename, *rctr, *rctr_end, *tmp;
 544	char buf[MAX_EVENT_NAME_LEN];
 545	char gbuf[MAX_EVENT_NAME_LEN];
 546	enum probe_print_type ptype;
 547	struct path path;
 548	unsigned long offset, ref_ctr_offset;
 549	bool is_return = false;
 550	int i, ret;
 551
 
 552	ref_ctr_offset = 0;
 553
 554	switch (argv[0][0]) {
 555	case 'r':
 556		is_return = true;
 557		break;
 558	case 'p':
 559		break;
 560	default:
 561		return -ECANCELED;
 562	}
 563
 564	if (argc < 2)
 565		return -ECANCELED;
 566
 567	if (argv[0][1] == ':')
 568		event = &argv[0][2];
 569
 570	if (!strchr(argv[1], '/'))
 571		return -ECANCELED;
 572
 573	filename = kstrdup(argv[1], GFP_KERNEL);
 574	if (!filename)
 575		return -ENOMEM;
 576
 577	/* Find the last occurrence, in case the path contains ':' too. */
 578	arg = strrchr(filename, ':');
 579	if (!arg || !isdigit(arg[1])) {
 580		kfree(filename);
 581		return -ECANCELED;
 582	}
 583
 584	trace_probe_log_init("trace_uprobe", argc, argv);
 585	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
 586
 587	*arg++ = '\0';
 588	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 589	if (ret) {
 590		trace_probe_log_err(0, FILE_NOT_FOUND);
 591		kfree(filename);
 592		trace_probe_log_clear();
 593		return ret;
 594	}
 595	if (!d_is_reg(path.dentry)) {
 596		trace_probe_log_err(0, NO_REGULAR_FILE);
 597		ret = -EINVAL;
 598		goto fail_address_parse;
 599	}
 600
 601	/* Parse reference counter offset if specified. */
 602	rctr = strchr(arg, '(');
 603	if (rctr) {
 604		rctr_end = strchr(rctr, ')');
 605		if (!rctr_end) {
 606			ret = -EINVAL;
 607			rctr_end = rctr + strlen(rctr);
 608			trace_probe_log_err(rctr_end - filename,
 609					    REFCNT_OPEN_BRACE);
 610			goto fail_address_parse;
 611		} else if (rctr_end[1] != '\0') {
 612			ret = -EINVAL;
 613			trace_probe_log_err(rctr_end + 1 - filename,
 614					    BAD_REFCNT_SUFFIX);
 615			goto fail_address_parse;
 616		}
 617
 618		*rctr++ = '\0';
 619		*rctr_end = '\0';
 620		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 621		if (ret) {
 622			trace_probe_log_err(rctr - filename, BAD_REFCNT);
 623			goto fail_address_parse;
 624		}
 625	}
 626
 627	/* Check if there is %return suffix */
 628	tmp = strchr(arg, '%');
 629	if (tmp) {
 630		if (!strcmp(tmp, "%return")) {
 631			*tmp = '\0';
 632			is_return = true;
 633		} else {
 634			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 635			ret = -EINVAL;
 636			goto fail_address_parse;
 637		}
 638	}
 639
 640	/* Parse uprobe offset. */
 641	ret = kstrtoul(arg, 0, &offset);
 642	if (ret) {
 643		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 644		goto fail_address_parse;
 645	}
 646
 647	/* setup a probe */
 648	trace_probe_log_set_index(0);
 649	if (event) {
 650		ret = traceprobe_parse_event_name(&event, &group, gbuf,
 651						  event - argv[0]);
 652		if (ret)
 653			goto fail_address_parse;
 654	}
 655
 656	if (!event) {
 657		char *tail;
 658		char *ptr;
 659
 660		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 661		if (!tail) {
 662			ret = -ENOMEM;
 663			goto fail_address_parse;
 664		}
 665
 666		ptr = strpbrk(tail, ".-_");
 667		if (ptr)
 668			*ptr = '\0';
 669
 670		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 671		event = buf;
 672		kfree(tail);
 673	}
 674
 675	argc -= 2;
 676	argv += 2;
 677
 678	tu = alloc_trace_uprobe(group, event, argc, is_return);
 679	if (IS_ERR(tu)) {
 680		ret = PTR_ERR(tu);
 681		/* This must return -ENOMEM otherwise there is a bug */
 682		WARN_ON_ONCE(ret != -ENOMEM);
 683		goto fail_address_parse;
 684	}
 685	tu->offset = offset;
 686	tu->ref_ctr_offset = ref_ctr_offset;
 687	tu->path = path;
 688	tu->filename = filename;
 689
 690	/* parse arguments */
 691	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 
 
 
 
 
 
 692		trace_probe_log_set_index(i + 2);
 693		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
 694					(is_return ? TPARG_FL_RETURN : 0) |
 695					TPARG_FL_USER);
 696		if (ret)
 697			goto error;
 698	}
 699
 700	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 701	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 702	if (ret < 0)
 703		goto error;
 704
 705	ret = register_trace_uprobe(tu);
 706	if (!ret)
 707		goto out;
 708
 709error:
 710	free_trace_uprobe(tu);
 711out:
 712	trace_probe_log_clear();
 713	return ret;
 714
 715fail_address_parse:
 716	trace_probe_log_clear();
 717	path_put(&path);
 718	kfree(filename);
 719
 720	return ret;
 721}
 722
 723int trace_uprobe_create(const char *raw_command)
 724{
 725	return trace_probe_create(raw_command, __trace_uprobe_create);
 726}
 727
 728static int create_or_delete_trace_uprobe(const char *raw_command)
 729{
 730	int ret;
 731
 732	if (raw_command[0] == '-')
 733		return dyn_event_release(raw_command, &trace_uprobe_ops);
 734
 735	ret = trace_uprobe_create(raw_command);
 736	return ret == -ECANCELED ? -EINVAL : ret;
 737}
 738
 739static int trace_uprobe_release(struct dyn_event *ev)
 740{
 741	struct trace_uprobe *tu = to_trace_uprobe(ev);
 742
 743	return unregister_trace_uprobe(tu);
 744}
 745
 746/* Probes listing interfaces */
 747static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 748{
 749	struct trace_uprobe *tu = to_trace_uprobe(ev);
 750	char c = is_ret_probe(tu) ? 'r' : 'p';
 751	int i;
 752
 753	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 754			trace_probe_name(&tu->tp), tu->filename,
 755			(int)(sizeof(void *) * 2), tu->offset);
 756
 757	if (tu->ref_ctr_offset)
 758		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 759
 760	for (i = 0; i < tu->tp.nr_args; i++)
 761		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 762
 763	seq_putc(m, '\n');
 764	return 0;
 765}
 766
 767static int probes_seq_show(struct seq_file *m, void *v)
 768{
 769	struct dyn_event *ev = v;
 770
 771	if (!is_trace_uprobe(ev))
 772		return 0;
 773
 774	return trace_uprobe_show(m, ev);
 775}
 776
 777static const struct seq_operations probes_seq_op = {
 778	.start  = dyn_event_seq_start,
 779	.next   = dyn_event_seq_next,
 780	.stop   = dyn_event_seq_stop,
 781	.show   = probes_seq_show
 782};
 783
 784static int probes_open(struct inode *inode, struct file *file)
 785{
 786	int ret;
 787
 788	ret = security_locked_down(LOCKDOWN_TRACEFS);
 789	if (ret)
 790		return ret;
 791
 792	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 793		ret = dyn_events_release_all(&trace_uprobe_ops);
 794		if (ret)
 795			return ret;
 796	}
 797
 798	return seq_open(file, &probes_seq_op);
 799}
 800
 801static ssize_t probes_write(struct file *file, const char __user *buffer,
 802			    size_t count, loff_t *ppos)
 803{
 804	return trace_parse_run_command(file, buffer, count, ppos,
 805					create_or_delete_trace_uprobe);
 806}
 807
 808static const struct file_operations uprobe_events_ops = {
 809	.owner		= THIS_MODULE,
 810	.open		= probes_open,
 811	.read		= seq_read,
 812	.llseek		= seq_lseek,
 813	.release	= seq_release,
 814	.write		= probes_write,
 815};
 816
 817/* Probes profiling interfaces */
 818static int probes_profile_seq_show(struct seq_file *m, void *v)
 819{
 820	struct dyn_event *ev = v;
 821	struct trace_uprobe *tu;
 822
 823	if (!is_trace_uprobe(ev))
 824		return 0;
 825
 826	tu = to_trace_uprobe(ev);
 827	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 828			trace_probe_name(&tu->tp), tu->nhit);
 829	return 0;
 830}
 831
 832static const struct seq_operations profile_seq_op = {
 833	.start  = dyn_event_seq_start,
 834	.next   = dyn_event_seq_next,
 835	.stop   = dyn_event_seq_stop,
 836	.show	= probes_profile_seq_show
 837};
 838
 839static int profile_open(struct inode *inode, struct file *file)
 840{
 841	int ret;
 842
 843	ret = security_locked_down(LOCKDOWN_TRACEFS);
 844	if (ret)
 845		return ret;
 846
 847	return seq_open(file, &profile_seq_op);
 848}
 849
 850static const struct file_operations uprobe_profile_ops = {
 851	.owner		= THIS_MODULE,
 852	.open		= profile_open,
 853	.read		= seq_read,
 854	.llseek		= seq_lseek,
 855	.release	= seq_release,
 856};
 857
 858struct uprobe_cpu_buffer {
 859	struct mutex mutex;
 860	void *buf;
 861};
 862static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 863static int uprobe_buffer_refcnt;
 864
 865static int uprobe_buffer_init(void)
 866{
 867	int cpu, err_cpu;
 868
 869	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 870	if (uprobe_cpu_buffer == NULL)
 871		return -ENOMEM;
 872
 873	for_each_possible_cpu(cpu) {
 874		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 875						  GFP_KERNEL, 0);
 876		if (p == NULL) {
 877			err_cpu = cpu;
 878			goto err;
 879		}
 880		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 881		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 882	}
 883
 884	return 0;
 885
 886err:
 887	for_each_possible_cpu(cpu) {
 888		if (cpu == err_cpu)
 889			break;
 890		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 891	}
 892
 893	free_percpu(uprobe_cpu_buffer);
 894	return -ENOMEM;
 895}
 896
 897static int uprobe_buffer_enable(void)
 898{
 899	int ret = 0;
 900
 901	BUG_ON(!mutex_is_locked(&event_mutex));
 902
 903	if (uprobe_buffer_refcnt++ == 0) {
 904		ret = uprobe_buffer_init();
 905		if (ret < 0)
 906			uprobe_buffer_refcnt--;
 907	}
 908
 909	return ret;
 910}
 911
 912static void uprobe_buffer_disable(void)
 913{
 914	int cpu;
 915
 916	BUG_ON(!mutex_is_locked(&event_mutex));
 917
 918	if (--uprobe_buffer_refcnt == 0) {
 919		for_each_possible_cpu(cpu)
 920			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 921							     cpu)->buf);
 922
 923		free_percpu(uprobe_cpu_buffer);
 924		uprobe_cpu_buffer = NULL;
 925	}
 926}
 927
 928static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 929{
 930	struct uprobe_cpu_buffer *ucb;
 931	int cpu;
 932
 933	cpu = raw_smp_processor_id();
 934	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 935
 936	/*
 937	 * Use per-cpu buffers for fastest access, but we might migrate
 938	 * so the mutex makes sure we have sole access to it.
 939	 */
 940	mutex_lock(&ucb->mutex);
 941
 942	return ucb;
 943}
 944
 945static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 946{
 947	mutex_unlock(&ucb->mutex);
 948}
 949
 950static void __uprobe_trace_func(struct trace_uprobe *tu,
 951				unsigned long func, struct pt_regs *regs,
 952				struct uprobe_cpu_buffer *ucb, int dsize,
 953				struct trace_event_file *trace_file)
 954{
 955	struct uprobe_trace_entry_head *entry;
 956	struct trace_event_buffer fbuffer;
 
 957	void *data;
 958	int size, esize;
 959	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 960
 961	WARN_ON(call != trace_file->event_call);
 962
 963	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 964		return;
 965
 966	if (trace_trigger_soft_disabled(trace_file))
 967		return;
 968
 969	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 970	size = esize + tu->tp.size + dsize;
 971	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 972	if (!entry)
 
 973		return;
 974
 
 975	if (is_ret_probe(tu)) {
 976		entry->vaddr[0] = func;
 977		entry->vaddr[1] = instruction_pointer(regs);
 978		data = DATAOF_TRACE_ENTRY(entry, true);
 979	} else {
 980		entry->vaddr[0] = instruction_pointer(regs);
 981		data = DATAOF_TRACE_ENTRY(entry, false);
 982	}
 983
 984	memcpy(data, ucb->buf, tu->tp.size + dsize);
 985
 986	trace_event_buffer_commit(&fbuffer);
 987}
 988
 989/* uprobe handler */
 990static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 991			     struct uprobe_cpu_buffer *ucb, int dsize)
 992{
 993	struct event_file_link *link;
 994
 995	if (is_ret_probe(tu))
 996		return 0;
 997
 998	rcu_read_lock();
 999	trace_probe_for_each_link_rcu(link, &tu->tp)
1000		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1001	rcu_read_unlock();
1002
1003	return 0;
1004}
1005
1006static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1007				 struct pt_regs *regs,
1008				 struct uprobe_cpu_buffer *ucb, int dsize)
1009{
1010	struct event_file_link *link;
1011
1012	rcu_read_lock();
1013	trace_probe_for_each_link_rcu(link, &tu->tp)
1014		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1015	rcu_read_unlock();
1016}
1017
1018/* Event entry printers */
1019static enum print_line_t
1020print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1021{
1022	struct uprobe_trace_entry_head *entry;
1023	struct trace_seq *s = &iter->seq;
1024	struct trace_uprobe *tu;
1025	u8 *data;
1026
1027	entry = (struct uprobe_trace_entry_head *)iter->ent;
1028	tu = trace_uprobe_primary_from_call(
1029		container_of(event, struct trace_event_call, event));
1030	if (unlikely(!tu))
1031		goto out;
1032
1033	if (is_ret_probe(tu)) {
1034		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1035				 trace_probe_name(&tu->tp),
1036				 entry->vaddr[1], entry->vaddr[0]);
1037		data = DATAOF_TRACE_ENTRY(entry, true);
1038	} else {
1039		trace_seq_printf(s, "%s: (0x%lx)",
1040				 trace_probe_name(&tu->tp),
1041				 entry->vaddr[0]);
1042		data = DATAOF_TRACE_ENTRY(entry, false);
1043	}
1044
1045	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1046		goto out;
1047
1048	trace_seq_putc(s, '\n');
1049
1050 out:
1051	return trace_handle_return(s);
1052}
1053
1054typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1055				enum uprobe_filter_ctx ctx,
1056				struct mm_struct *mm);
1057
1058static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1059{
1060	int ret;
1061
1062	tu->consumer.filter = filter;
1063	tu->inode = d_real_inode(tu->path.dentry);
1064
1065	if (tu->ref_ctr_offset)
1066		ret = uprobe_register_refctr(tu->inode, tu->offset,
1067				tu->ref_ctr_offset, &tu->consumer);
1068	else
1069		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1070
1071	if (ret)
1072		tu->inode = NULL;
1073
1074	return ret;
1075}
1076
1077static void __probe_event_disable(struct trace_probe *tp)
1078{
 
1079	struct trace_uprobe *tu;
1080
1081	tu = container_of(tp, struct trace_uprobe, tp);
1082	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1083
1084	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
 
1085		if (!tu->inode)
1086			continue;
1087
1088		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1089		tu->inode = NULL;
1090	}
1091}
1092
1093static int probe_event_enable(struct trace_event_call *call,
1094			struct trace_event_file *file, filter_func_t filter)
1095{
1096	struct trace_probe *tp;
1097	struct trace_uprobe *tu;
1098	bool enabled;
1099	int ret;
1100
1101	tp = trace_probe_primary_from_call(call);
1102	if (WARN_ON_ONCE(!tp))
1103		return -ENODEV;
1104	enabled = trace_probe_is_enabled(tp);
1105
1106	/* This may also change "enabled" state */
1107	if (file) {
1108		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1109			return -EINTR;
1110
1111		ret = trace_probe_add_file(tp, file);
1112		if (ret < 0)
1113			return ret;
1114	} else {
1115		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1116			return -EINTR;
1117
1118		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1119	}
1120
1121	tu = container_of(tp, struct trace_uprobe, tp);
1122	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1123
1124	if (enabled)
1125		return 0;
1126
1127	ret = uprobe_buffer_enable();
1128	if (ret)
1129		goto err_flags;
1130
1131	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
 
1132		ret = trace_uprobe_enable(tu, filter);
1133		if (ret) {
1134			__probe_event_disable(tp);
1135			goto err_buffer;
1136		}
1137	}
1138
1139	return 0;
1140
1141 err_buffer:
1142	uprobe_buffer_disable();
1143
1144 err_flags:
1145	if (file)
1146		trace_probe_remove_file(tp, file);
1147	else
1148		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1149
1150	return ret;
1151}
1152
1153static void probe_event_disable(struct trace_event_call *call,
1154				struct trace_event_file *file)
1155{
1156	struct trace_probe *tp;
1157
1158	tp = trace_probe_primary_from_call(call);
1159	if (WARN_ON_ONCE(!tp))
1160		return;
1161
1162	if (!trace_probe_is_enabled(tp))
1163		return;
1164
1165	if (file) {
1166		if (trace_probe_remove_file(tp, file) < 0)
1167			return;
1168
1169		if (trace_probe_is_enabled(tp))
1170			return;
1171	} else
1172		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1173
1174	__probe_event_disable(tp);
1175	uprobe_buffer_disable();
1176}
1177
1178static int uprobe_event_define_fields(struct trace_event_call *event_call)
1179{
1180	int ret, size;
1181	struct uprobe_trace_entry_head field;
1182	struct trace_uprobe *tu;
1183
1184	tu = trace_uprobe_primary_from_call(event_call);
1185	if (unlikely(!tu))
1186		return -ENODEV;
1187
1188	if (is_ret_probe(tu)) {
1189		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1190		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1191		size = SIZEOF_TRACE_ENTRY(true);
1192	} else {
1193		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1194		size = SIZEOF_TRACE_ENTRY(false);
1195	}
1196
1197	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1198}
1199
1200#ifdef CONFIG_PERF_EVENTS
1201static bool
1202__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1203{
1204	struct perf_event *event;
1205
1206	if (filter->nr_systemwide)
1207		return true;
1208
1209	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1210		if (event->hw.target->mm == mm)
1211			return true;
1212	}
1213
1214	return false;
1215}
1216
1217static inline bool
1218trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1219			  struct perf_event *event)
1220{
1221	return __uprobe_perf_filter(filter, event->hw.target->mm);
1222}
1223
1224static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1225				       struct perf_event *event)
1226{
1227	bool done;
1228
1229	write_lock(&filter->rwlock);
1230	if (event->hw.target) {
1231		list_del(&event->hw.tp_list);
1232		done = filter->nr_systemwide ||
1233			(event->hw.target->flags & PF_EXITING) ||
1234			trace_uprobe_filter_event(filter, event);
1235	} else {
1236		filter->nr_systemwide--;
1237		done = filter->nr_systemwide;
1238	}
1239	write_unlock(&filter->rwlock);
1240
1241	return done;
1242}
1243
1244/* This returns true if the filter always covers target mm */
1245static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1246				    struct perf_event *event)
1247{
1248	bool done;
1249
1250	write_lock(&filter->rwlock);
1251	if (event->hw.target) {
1252		/*
1253		 * event->parent != NULL means copy_process(), we can avoid
1254		 * uprobe_apply(). current->mm must be probed and we can rely
1255		 * on dup_mmap() which preserves the already installed bp's.
1256		 *
1257		 * attr.enable_on_exec means that exec/mmap will install the
1258		 * breakpoints we need.
1259		 */
1260		done = filter->nr_systemwide ||
1261			event->parent || event->attr.enable_on_exec ||
1262			trace_uprobe_filter_event(filter, event);
1263		list_add(&event->hw.tp_list, &filter->perf_events);
1264	} else {
1265		done = filter->nr_systemwide;
1266		filter->nr_systemwide++;
1267	}
1268	write_unlock(&filter->rwlock);
1269
1270	return done;
1271}
1272
1273static int uprobe_perf_close(struct trace_event_call *call,
1274			     struct perf_event *event)
1275{
1276	struct trace_probe *tp;
1277	struct trace_uprobe *tu;
1278	int ret = 0;
1279
1280	tp = trace_probe_primary_from_call(call);
1281	if (WARN_ON_ONCE(!tp))
1282		return -ENODEV;
1283
1284	tu = container_of(tp, struct trace_uprobe, tp);
1285	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1286		return 0;
1287
1288	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
 
1289		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1290		if (ret)
1291			break;
1292	}
1293
1294	return ret;
1295}
1296
1297static int uprobe_perf_open(struct trace_event_call *call,
1298			    struct perf_event *event)
1299{
1300	struct trace_probe *tp;
1301	struct trace_uprobe *tu;
1302	int err = 0;
1303
1304	tp = trace_probe_primary_from_call(call);
1305	if (WARN_ON_ONCE(!tp))
1306		return -ENODEV;
1307
1308	tu = container_of(tp, struct trace_uprobe, tp);
1309	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1310		return 0;
1311
1312	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1313		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1314		if (err) {
1315			uprobe_perf_close(call, event);
1316			break;
1317		}
1318	}
1319
1320	return err;
1321}
1322
1323static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1324				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1325{
1326	struct trace_uprobe_filter *filter;
1327	struct trace_uprobe *tu;
1328	int ret;
1329
1330	tu = container_of(uc, struct trace_uprobe, consumer);
1331	filter = tu->tp.event->filter;
1332
1333	read_lock(&filter->rwlock);
1334	ret = __uprobe_perf_filter(filter, mm);
1335	read_unlock(&filter->rwlock);
1336
1337	return ret;
1338}
1339
1340static void __uprobe_perf_func(struct trace_uprobe *tu,
1341			       unsigned long func, struct pt_regs *regs,
1342			       struct uprobe_cpu_buffer *ucb, int dsize)
1343{
1344	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1345	struct uprobe_trace_entry_head *entry;
1346	struct hlist_head *head;
1347	void *data;
1348	int size, esize;
1349	int rctx;
1350
1351#ifdef CONFIG_BPF_EVENTS
1352	if (bpf_prog_array_valid(call)) {
1353		u32 ret;
1354
1355		ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
 
 
1356		if (!ret)
1357			return;
1358	}
1359#endif /* CONFIG_BPF_EVENTS */
1360
1361	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1362
1363	size = esize + tu->tp.size + dsize;
1364	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1365	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1366		return;
1367
1368	preempt_disable();
1369	head = this_cpu_ptr(call->perf_events);
1370	if (hlist_empty(head))
1371		goto out;
1372
1373	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1374	if (!entry)
1375		goto out;
1376
1377	if (is_ret_probe(tu)) {
1378		entry->vaddr[0] = func;
1379		entry->vaddr[1] = instruction_pointer(regs);
1380		data = DATAOF_TRACE_ENTRY(entry, true);
1381	} else {
1382		entry->vaddr[0] = instruction_pointer(regs);
1383		data = DATAOF_TRACE_ENTRY(entry, false);
1384	}
1385
1386	memcpy(data, ucb->buf, tu->tp.size + dsize);
1387
1388	if (size - esize > tu->tp.size + dsize) {
1389		int len = tu->tp.size + dsize;
1390
1391		memset(data + len, 0, size - esize - len);
1392	}
1393
1394	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1395			      head, NULL);
1396 out:
1397	preempt_enable();
1398}
1399
1400/* uprobe profile handler */
1401static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1402			    struct uprobe_cpu_buffer *ucb, int dsize)
1403{
1404	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1405		return UPROBE_HANDLER_REMOVE;
1406
1407	if (!is_ret_probe(tu))
1408		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1409	return 0;
1410}
1411
1412static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1413				struct pt_regs *regs,
1414				struct uprobe_cpu_buffer *ucb, int dsize)
1415{
1416	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1417}
1418
1419int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1420			const char **filename, u64 *probe_offset,
1421			bool perf_type_tracepoint)
1422{
1423	const char *pevent = trace_event_name(event->tp_event);
1424	const char *group = event->tp_event->class->system;
1425	struct trace_uprobe *tu;
1426
1427	if (perf_type_tracepoint)
1428		tu = find_probe_event(pevent, group);
1429	else
1430		tu = trace_uprobe_primary_from_call(event->tp_event);
1431	if (!tu)
1432		return -EINVAL;
1433
1434	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1435				    : BPF_FD_TYPE_UPROBE;
1436	*filename = tu->filename;
1437	*probe_offset = tu->offset;
1438	return 0;
1439}
1440#endif	/* CONFIG_PERF_EVENTS */
1441
1442static int
1443trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1444		      void *data)
1445{
1446	struct trace_event_file *file = data;
1447
1448	switch (type) {
1449	case TRACE_REG_REGISTER:
1450		return probe_event_enable(event, file, NULL);
1451
1452	case TRACE_REG_UNREGISTER:
1453		probe_event_disable(event, file);
1454		return 0;
1455
1456#ifdef CONFIG_PERF_EVENTS
1457	case TRACE_REG_PERF_REGISTER:
1458		return probe_event_enable(event, NULL, uprobe_perf_filter);
1459
1460	case TRACE_REG_PERF_UNREGISTER:
1461		probe_event_disable(event, NULL);
1462		return 0;
1463
1464	case TRACE_REG_PERF_OPEN:
1465		return uprobe_perf_open(event, data);
1466
1467	case TRACE_REG_PERF_CLOSE:
1468		return uprobe_perf_close(event, data);
1469
1470#endif
1471	default:
1472		return 0;
1473	}
1474}
1475
1476static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1477{
1478	struct trace_uprobe *tu;
1479	struct uprobe_dispatch_data udd;
1480	struct uprobe_cpu_buffer *ucb;
1481	int dsize, esize;
1482	int ret = 0;
1483
1484
1485	tu = container_of(con, struct trace_uprobe, consumer);
1486	tu->nhit++;
1487
1488	udd.tu = tu;
1489	udd.bp_addr = instruction_pointer(regs);
1490
1491	current->utask->vaddr = (unsigned long) &udd;
1492
1493	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1494		return 0;
1495
1496	dsize = __get_data_size(&tu->tp, regs);
1497	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1498
1499	ucb = uprobe_buffer_get();
1500	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1501
1502	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1503		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1504
1505#ifdef CONFIG_PERF_EVENTS
1506	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1507		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1508#endif
1509	uprobe_buffer_put(ucb);
1510	return ret;
1511}
1512
1513static int uretprobe_dispatcher(struct uprobe_consumer *con,
1514				unsigned long func, struct pt_regs *regs)
1515{
1516	struct trace_uprobe *tu;
1517	struct uprobe_dispatch_data udd;
1518	struct uprobe_cpu_buffer *ucb;
1519	int dsize, esize;
1520
1521	tu = container_of(con, struct trace_uprobe, consumer);
1522
1523	udd.tu = tu;
1524	udd.bp_addr = func;
1525
1526	current->utask->vaddr = (unsigned long) &udd;
1527
1528	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1529		return 0;
1530
1531	dsize = __get_data_size(&tu->tp, regs);
1532	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1533
1534	ucb = uprobe_buffer_get();
1535	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1536
1537	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1538		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1539
1540#ifdef CONFIG_PERF_EVENTS
1541	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1542		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1543#endif
1544	uprobe_buffer_put(ucb);
1545	return 0;
1546}
1547
1548static struct trace_event_functions uprobe_funcs = {
1549	.trace		= print_uprobe_event
1550};
1551
1552static struct trace_event_fields uprobe_fields_array[] = {
1553	{ .type = TRACE_FUNCTION_TYPE,
1554	  .define_fields = uprobe_event_define_fields },
1555	{}
1556};
1557
1558static inline void init_trace_event_call(struct trace_uprobe *tu)
1559{
1560	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1561	call->event.funcs = &uprobe_funcs;
1562	call->class->fields_array = uprobe_fields_array;
1563
1564	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1565	call->class->reg = trace_uprobe_register;
1566}
1567
1568static int register_uprobe_event(struct trace_uprobe *tu)
1569{
1570	init_trace_event_call(tu);
1571
1572	return trace_probe_register_event_call(&tu->tp);
1573}
1574
1575static int unregister_uprobe_event(struct trace_uprobe *tu)
1576{
1577	return trace_probe_unregister_event_call(&tu->tp);
1578}
1579
1580#ifdef CONFIG_PERF_EVENTS
1581struct trace_event_call *
1582create_local_trace_uprobe(char *name, unsigned long offs,
1583			  unsigned long ref_ctr_offset, bool is_return)
1584{
1585	enum probe_print_type ptype;
1586	struct trace_uprobe *tu;
1587	struct path path;
1588	int ret;
1589
1590	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1591	if (ret)
1592		return ERR_PTR(ret);
1593
1594	if (!d_is_reg(path.dentry)) {
1595		path_put(&path);
1596		return ERR_PTR(-EINVAL);
1597	}
1598
1599	/*
1600	 * local trace_kprobes are not added to dyn_event, so they are never
1601	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1602	 * duplicated name "DUMMY_EVENT" here.
1603	 */
1604	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1605				is_return);
1606
1607	if (IS_ERR(tu)) {
1608		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1609			(int)PTR_ERR(tu));
1610		path_put(&path);
1611		return ERR_CAST(tu);
1612	}
1613
1614	tu->offset = offs;
1615	tu->path = path;
1616	tu->ref_ctr_offset = ref_ctr_offset;
1617	tu->filename = kstrdup(name, GFP_KERNEL);
1618	if (!tu->filename) {
1619		ret = -ENOMEM;
1620		goto error;
1621	}
1622
1623	init_trace_event_call(tu);
1624
1625	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1626	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1627		ret = -ENOMEM;
1628		goto error;
1629	}
1630
1631	return trace_probe_event_call(&tu->tp);
1632error:
1633	free_trace_uprobe(tu);
1634	return ERR_PTR(ret);
1635}
1636
1637void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1638{
1639	struct trace_uprobe *tu;
1640
1641	tu = trace_uprobe_primary_from_call(event_call);
1642
1643	free_trace_uprobe(tu);
1644}
1645#endif /* CONFIG_PERF_EVENTS */
1646
1647/* Make a trace interface for controlling probe points */
1648static __init int init_uprobe_trace(void)
1649{
 
1650	int ret;
1651
1652	ret = dyn_event_register(&trace_uprobe_ops);
1653	if (ret)
1654		return ret;
1655
1656	ret = tracing_init_dentry();
1657	if (ret)
1658		return 0;
1659
1660	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1661				    NULL, &uprobe_events_ops);
1662	/* Profile interface */
1663	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1664				    NULL, &uprobe_profile_ops);
1665	return 0;
1666}
1667
1668fs_initcall(init_uprobe_trace);
v5.9
   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/security.h>
  11#include <linux/ctype.h>
  12#include <linux/module.h>
  13#include <linux/uaccess.h>
  14#include <linux/uprobes.h>
  15#include <linux/namei.h>
  16#include <linux/string.h>
  17#include <linux/rculist.h>
 
  18
  19#include "trace_dynevent.h"
  20#include "trace_probe.h"
  21#include "trace_probe_tmpl.h"
  22
  23#define UPROBE_EVENT_SYSTEM	"uprobes"
  24
  25struct uprobe_trace_entry_head {
  26	struct trace_entry	ent;
  27	unsigned long		vaddr[];
  28};
  29
  30#define SIZEOF_TRACE_ENTRY(is_return)			\
  31	(sizeof(struct uprobe_trace_entry_head) +	\
  32	 sizeof(unsigned long) * (is_return ? 2 : 1))
  33
  34#define DATAOF_TRACE_ENTRY(entry, is_return)		\
  35	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  36
  37static int trace_uprobe_create(int argc, const char **argv);
  38static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  39static int trace_uprobe_release(struct dyn_event *ev);
  40static bool trace_uprobe_is_busy(struct dyn_event *ev);
  41static bool trace_uprobe_match(const char *system, const char *event,
  42			int argc, const char **argv, struct dyn_event *ev);
  43
  44static struct dyn_event_operations trace_uprobe_ops = {
  45	.create = trace_uprobe_create,
  46	.show = trace_uprobe_show,
  47	.is_busy = trace_uprobe_is_busy,
  48	.free = trace_uprobe_release,
  49	.match = trace_uprobe_match,
  50};
  51
  52/*
  53 * uprobe event core functions
  54 */
  55struct trace_uprobe {
  56	struct dyn_event		devent;
  57	struct uprobe_consumer		consumer;
  58	struct path			path;
  59	struct inode			*inode;
  60	char				*filename;
  61	unsigned long			offset;
  62	unsigned long			ref_ctr_offset;
  63	unsigned long			nhit;
  64	struct trace_probe		tp;
  65};
  66
  67static bool is_trace_uprobe(struct dyn_event *ev)
  68{
  69	return ev->ops == &trace_uprobe_ops;
  70}
  71
  72static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  73{
  74	return container_of(ev, struct trace_uprobe, devent);
  75}
  76
  77/**
  78 * for_each_trace_uprobe - iterate over the trace_uprobe list
  79 * @pos:	the struct trace_uprobe * for each entry
  80 * @dpos:	the struct dyn_event * to use as a loop cursor
  81 */
  82#define for_each_trace_uprobe(pos, dpos)	\
  83	for_each_dyn_event(dpos)		\
  84		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  85
  86#define SIZEOF_TRACE_UPROBE(n)				\
  87	(offsetof(struct trace_uprobe, tp.args) +	\
  88	(sizeof(struct probe_arg) * (n)))
  89
  90static int register_uprobe_event(struct trace_uprobe *tu);
  91static int unregister_uprobe_event(struct trace_uprobe *tu);
  92
  93struct uprobe_dispatch_data {
  94	struct trace_uprobe	*tu;
  95	unsigned long		bp_addr;
  96};
  97
  98static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  99static int uretprobe_dispatcher(struct uprobe_consumer *con,
 100				unsigned long func, struct pt_regs *regs);
 101
 102#ifdef CONFIG_STACK_GROWSUP
 103static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 104{
 105	return addr - (n * sizeof(long));
 106}
 107#else
 108static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 109{
 110	return addr + (n * sizeof(long));
 111}
 112#endif
 113
 114static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 115{
 116	unsigned long ret;
 117	unsigned long addr = user_stack_pointer(regs);
 118
 119	addr = adjust_stack_addr(addr, n);
 120
 121	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 122		return 0;
 123
 124	return ret;
 125}
 126
 127/*
 128 * Uprobes-specific fetch functions
 129 */
 130static nokprobe_inline int
 131probe_mem_read(void *dest, void *src, size_t size)
 132{
 133	void __user *vaddr = (void __force __user *)src;
 134
 135	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 136}
 137
 138static nokprobe_inline int
 139probe_mem_read_user(void *dest, void *src, size_t size)
 140{
 141	return probe_mem_read(dest, src, size);
 142}
 143
 144/*
 145 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 146 * length and relative data location.
 147 */
 148static nokprobe_inline int
 149fetch_store_string(unsigned long addr, void *dest, void *base)
 150{
 151	long ret;
 152	u32 loc = *(u32 *)dest;
 153	int maxlen  = get_loc_len(loc);
 154	u8 *dst = get_loc_data(dest, base);
 155	void __user *src = (void __force __user *) addr;
 156
 157	if (unlikely(!maxlen))
 158		return -ENOMEM;
 159
 160	if (addr == FETCH_TOKEN_COMM)
 161		ret = strlcpy(dst, current->comm, maxlen);
 162	else
 163		ret = strncpy_from_user(dst, src, maxlen);
 164	if (ret >= 0) {
 165		if (ret == maxlen)
 166			dst[ret - 1] = '\0';
 167		else
 168			/*
 169			 * Include the terminating null byte. In this case it
 170			 * was copied by strncpy_from_user but not accounted
 171			 * for in ret.
 172			 */
 173			ret++;
 174		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 175	}
 176
 177	return ret;
 178}
 179
 180static nokprobe_inline int
 181fetch_store_string_user(unsigned long addr, void *dest, void *base)
 182{
 183	return fetch_store_string(addr, dest, base);
 184}
 185
 186/* Return the length of string -- including null terminal byte */
 187static nokprobe_inline int
 188fetch_store_strlen(unsigned long addr)
 189{
 190	int len;
 191	void __user *vaddr = (void __force __user *) addr;
 192
 193	if (addr == FETCH_TOKEN_COMM)
 194		len = strlen(current->comm) + 1;
 195	else
 196		len = strnlen_user(vaddr, MAX_STRING_SIZE);
 197
 198	return (len > MAX_STRING_SIZE) ? 0 : len;
 199}
 200
 201static nokprobe_inline int
 202fetch_store_strlen_user(unsigned long addr)
 203{
 204	return fetch_store_strlen(addr);
 205}
 206
 207static unsigned long translate_user_vaddr(unsigned long file_offset)
 208{
 209	unsigned long base_addr;
 210	struct uprobe_dispatch_data *udd;
 211
 212	udd = (void *) current->utask->vaddr;
 213
 214	base_addr = udd->bp_addr - udd->tu->offset;
 215	return base_addr + file_offset;
 216}
 217
 218/* Note that we don't verify it, since the code does not come from user space */
 219static int
 220process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 221		   void *base)
 222{
 
 223	unsigned long val;
 224
 225	/* 1st stage: get value from context */
 226	switch (code->op) {
 227	case FETCH_OP_REG:
 228		val = regs_get_register(regs, code->param);
 229		break;
 230	case FETCH_OP_STACK:
 231		val = get_user_stack_nth(regs, code->param);
 232		break;
 233	case FETCH_OP_STACKP:
 234		val = user_stack_pointer(regs);
 235		break;
 236	case FETCH_OP_RETVAL:
 237		val = regs_return_value(regs);
 238		break;
 239	case FETCH_OP_IMM:
 240		val = code->immediate;
 241		break;
 242	case FETCH_OP_COMM:
 243		val = FETCH_TOKEN_COMM;
 244		break;
 245	case FETCH_OP_DATA:
 246		val = (unsigned long)code->data;
 247		break;
 248	case FETCH_OP_FOFFS:
 249		val = translate_user_vaddr(code->immediate);
 250		break;
 251	default:
 252		return -EILSEQ;
 253	}
 254	code++;
 255
 256	return process_fetch_insn_bottom(code, val, dest, base);
 257}
 258NOKPROBE_SYMBOL(process_fetch_insn)
 259
 260static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 261{
 262	rwlock_init(&filter->rwlock);
 263	filter->nr_systemwide = 0;
 264	INIT_LIST_HEAD(&filter->perf_events);
 265}
 266
 267static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 268{
 269	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 270}
 271
 272static inline bool is_ret_probe(struct trace_uprobe *tu)
 273{
 274	return tu->consumer.ret_handler != NULL;
 275}
 276
 277static bool trace_uprobe_is_busy(struct dyn_event *ev)
 278{
 279	struct trace_uprobe *tu = to_trace_uprobe(ev);
 280
 281	return trace_probe_is_enabled(&tu->tp);
 282}
 283
 284static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 285					    int argc, const char **argv)
 286{
 287	char buf[MAX_ARGSTR_LEN + 1];
 288	int len;
 289
 290	if (!argc)
 291		return true;
 292
 293	len = strlen(tu->filename);
 294	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 295		return false;
 296
 297	if (tu->ref_ctr_offset == 0)
 298		snprintf(buf, sizeof(buf), "0x%0*lx",
 299				(int)(sizeof(void *) * 2), tu->offset);
 300	else
 301		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 302				(int)(sizeof(void *) * 2), tu->offset,
 303				tu->ref_ctr_offset);
 304	if (strcmp(buf, &argv[0][len + 1]))
 305		return false;
 306
 307	argc--; argv++;
 308
 309	return trace_probe_match_command_args(&tu->tp, argc, argv);
 310}
 311
 312static bool trace_uprobe_match(const char *system, const char *event,
 313			int argc, const char **argv, struct dyn_event *ev)
 314{
 315	struct trace_uprobe *tu = to_trace_uprobe(ev);
 316
 317	return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 
 318	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 319	   trace_uprobe_match_command_head(tu, argc, argv);
 320}
 321
 322static nokprobe_inline struct trace_uprobe *
 323trace_uprobe_primary_from_call(struct trace_event_call *call)
 324{
 325	struct trace_probe *tp;
 326
 327	tp = trace_probe_primary_from_call(call);
 328	if (WARN_ON_ONCE(!tp))
 329		return NULL;
 330
 331	return container_of(tp, struct trace_uprobe, tp);
 332}
 333
 334/*
 335 * Allocate new trace_uprobe and initialize it (including uprobes).
 336 */
 337static struct trace_uprobe *
 338alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 339{
 340	struct trace_uprobe *tu;
 341	int ret;
 342
 343	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 344	if (!tu)
 345		return ERR_PTR(-ENOMEM);
 346
 347	ret = trace_probe_init(&tu->tp, event, group, true);
 348	if (ret < 0)
 349		goto error;
 350
 351	dyn_event_init(&tu->devent, &trace_uprobe_ops);
 352	tu->consumer.handler = uprobe_dispatcher;
 353	if (is_ret)
 354		tu->consumer.ret_handler = uretprobe_dispatcher;
 355	init_trace_uprobe_filter(tu->tp.event->filter);
 356	return tu;
 357
 358error:
 359	kfree(tu);
 360
 361	return ERR_PTR(ret);
 362}
 363
 364static void free_trace_uprobe(struct trace_uprobe *tu)
 365{
 366	if (!tu)
 367		return;
 368
 369	path_put(&tu->path);
 370	trace_probe_cleanup(&tu->tp);
 371	kfree(tu->filename);
 372	kfree(tu);
 373}
 374
 375static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 376{
 377	struct dyn_event *pos;
 378	struct trace_uprobe *tu;
 379
 380	for_each_trace_uprobe(tu, pos)
 381		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 382		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 383			return tu;
 384
 385	return NULL;
 386}
 387
 388/* Unregister a trace_uprobe and probe_event */
 389static int unregister_trace_uprobe(struct trace_uprobe *tu)
 390{
 391	int ret;
 392
 393	if (trace_probe_has_sibling(&tu->tp))
 394		goto unreg;
 395
 
 
 
 
 396	ret = unregister_uprobe_event(tu);
 397	if (ret)
 398		return ret;
 399
 400unreg:
 401	dyn_event_remove(&tu->devent);
 402	trace_probe_unlink(&tu->tp);
 403	free_trace_uprobe(tu);
 404	return 0;
 405}
 406
 407static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 408					 struct trace_uprobe *comp)
 409{
 410	struct trace_probe_event *tpe = orig->tp.event;
 411	struct trace_probe *pos;
 412	struct inode *comp_inode = d_real_inode(comp->path.dentry);
 413	int i;
 414
 415	list_for_each_entry(pos, &tpe->probes, list) {
 416		orig = container_of(pos, struct trace_uprobe, tp);
 417		if (comp_inode != d_real_inode(orig->path.dentry) ||
 418		    comp->offset != orig->offset)
 419			continue;
 420
 421		/*
 422		 * trace_probe_compare_arg_type() ensured that nr_args and
 423		 * each argument name and type are same. Let's compare comm.
 424		 */
 425		for (i = 0; i < orig->tp.nr_args; i++) {
 426			if (strcmp(orig->tp.args[i].comm,
 427				   comp->tp.args[i].comm))
 428				break;
 429		}
 430
 431		if (i == orig->tp.nr_args)
 432			return true;
 433	}
 434
 435	return false;
 436}
 437
 438static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 439{
 440	int ret;
 441
 442	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 443	if (ret) {
 444		/* Note that argument starts index = 2 */
 445		trace_probe_log_set_index(ret + 1);
 446		trace_probe_log_err(0, DIFF_ARG_TYPE);
 447		return -EEXIST;
 448	}
 449	if (trace_uprobe_has_same_uprobe(to, tu)) {
 450		trace_probe_log_set_index(0);
 451		trace_probe_log_err(0, SAME_PROBE);
 452		return -EEXIST;
 453	}
 454
 455	/* Append to existing event */
 456	ret = trace_probe_append(&tu->tp, &to->tp);
 457	if (!ret)
 458		dyn_event_add(&tu->devent);
 459
 460	return ret;
 461}
 462
 463/*
 464 * Uprobe with multiple reference counter is not allowed. i.e.
 465 * If inode and offset matches, reference counter offset *must*
 466 * match as well. Though, there is one exception: If user is
 467 * replacing old trace_uprobe with new one(same group/event),
 468 * then we allow same uprobe with new reference counter as far
 469 * as the new one does not conflict with any other existing
 470 * ones.
 471 */
 472static int validate_ref_ctr_offset(struct trace_uprobe *new)
 473{
 474	struct dyn_event *pos;
 475	struct trace_uprobe *tmp;
 476	struct inode *new_inode = d_real_inode(new->path.dentry);
 477
 478	for_each_trace_uprobe(tmp, pos) {
 479		if (new_inode == d_real_inode(tmp->path.dentry) &&
 480		    new->offset == tmp->offset &&
 481		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
 482			pr_warn("Reference counter offset mismatch.");
 483			return -EINVAL;
 484		}
 485	}
 486	return 0;
 487}
 488
 489/* Register a trace_uprobe and probe_event */
 490static int register_trace_uprobe(struct trace_uprobe *tu)
 491{
 492	struct trace_uprobe *old_tu;
 493	int ret;
 494
 495	mutex_lock(&event_mutex);
 496
 497	ret = validate_ref_ctr_offset(tu);
 498	if (ret)
 499		goto end;
 500
 501	/* register as an event */
 502	old_tu = find_probe_event(trace_probe_name(&tu->tp),
 503				  trace_probe_group_name(&tu->tp));
 504	if (old_tu) {
 505		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 506			trace_probe_log_set_index(0);
 507			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 508			ret = -EEXIST;
 509		} else {
 510			ret = append_trace_uprobe(tu, old_tu);
 511		}
 512		goto end;
 513	}
 514
 515	ret = register_uprobe_event(tu);
 516	if (ret) {
 517		pr_warn("Failed to register probe event(%d)\n", ret);
 
 
 
 
 518		goto end;
 519	}
 520
 521	dyn_event_add(&tu->devent);
 522
 523end:
 524	mutex_unlock(&event_mutex);
 525
 526	return ret;
 527}
 528
 529/*
 530 * Argument syntax:
 531 *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 532 */
 533static int trace_uprobe_create(int argc, const char **argv)
 534{
 535	struct trace_uprobe *tu;
 536	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 537	char *arg, *filename, *rctr, *rctr_end, *tmp;
 538	char buf[MAX_EVENT_NAME_LEN];
 
 
 539	struct path path;
 540	unsigned long offset, ref_ctr_offset;
 541	bool is_return = false;
 542	int i, ret;
 543
 544	ret = 0;
 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	/* Parse uprobe offset. */
 621	ret = kstrtoul(arg, 0, &offset);
 622	if (ret) {
 623		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 624		goto fail_address_parse;
 625	}
 626
 627	/* setup a probe */
 628	trace_probe_log_set_index(0);
 629	if (event) {
 630		ret = traceprobe_parse_event_name(&event, &group, buf,
 631						  event - argv[0]);
 632		if (ret)
 633			goto fail_address_parse;
 634	} else {
 
 
 635		char *tail;
 636		char *ptr;
 637
 638		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 639		if (!tail) {
 640			ret = -ENOMEM;
 641			goto fail_address_parse;
 642		}
 643
 644		ptr = strpbrk(tail, ".-_");
 645		if (ptr)
 646			*ptr = '\0';
 647
 648		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 649		event = buf;
 650		kfree(tail);
 651	}
 652
 653	argc -= 2;
 654	argv += 2;
 655
 656	tu = alloc_trace_uprobe(group, event, argc, is_return);
 657	if (IS_ERR(tu)) {
 658		ret = PTR_ERR(tu);
 659		/* This must return -ENOMEM otherwise there is a bug */
 660		WARN_ON_ONCE(ret != -ENOMEM);
 661		goto fail_address_parse;
 662	}
 663	tu->offset = offset;
 664	tu->ref_ctr_offset = ref_ctr_offset;
 665	tu->path = path;
 666	tu->filename = filename;
 667
 668	/* parse arguments */
 669	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 670		tmp = kstrdup(argv[i], GFP_KERNEL);
 671		if (!tmp) {
 672			ret = -ENOMEM;
 673			goto error;
 674		}
 675
 676		trace_probe_log_set_index(i + 2);
 677		ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 678					is_return ? TPARG_FL_RETURN : 0);
 679		kfree(tmp);
 680		if (ret)
 681			goto error;
 682	}
 683
 684	ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 
 685	if (ret < 0)
 686		goto error;
 687
 688	ret = register_trace_uprobe(tu);
 689	if (!ret)
 690		goto out;
 691
 692error:
 693	free_trace_uprobe(tu);
 694out:
 695	trace_probe_log_clear();
 696	return ret;
 697
 698fail_address_parse:
 699	trace_probe_log_clear();
 700	path_put(&path);
 701	kfree(filename);
 702
 703	return ret;
 704}
 705
 706static int create_or_delete_trace_uprobe(int argc, char **argv)
 
 
 
 
 
 707{
 708	int ret;
 709
 710	if (argv[0][0] == '-')
 711		return dyn_event_release(argc, argv, &trace_uprobe_ops);
 712
 713	ret = trace_uprobe_create(argc, (const char **)argv);
 714	return ret == -ECANCELED ? -EINVAL : ret;
 715}
 716
 717static int trace_uprobe_release(struct dyn_event *ev)
 718{
 719	struct trace_uprobe *tu = to_trace_uprobe(ev);
 720
 721	return unregister_trace_uprobe(tu);
 722}
 723
 724/* Probes listing interfaces */
 725static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 726{
 727	struct trace_uprobe *tu = to_trace_uprobe(ev);
 728	char c = is_ret_probe(tu) ? 'r' : 'p';
 729	int i;
 730
 731	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 732			trace_probe_name(&tu->tp), tu->filename,
 733			(int)(sizeof(void *) * 2), tu->offset);
 734
 735	if (tu->ref_ctr_offset)
 736		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 737
 738	for (i = 0; i < tu->tp.nr_args; i++)
 739		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 740
 741	seq_putc(m, '\n');
 742	return 0;
 743}
 744
 745static int probes_seq_show(struct seq_file *m, void *v)
 746{
 747	struct dyn_event *ev = v;
 748
 749	if (!is_trace_uprobe(ev))
 750		return 0;
 751
 752	return trace_uprobe_show(m, ev);
 753}
 754
 755static const struct seq_operations probes_seq_op = {
 756	.start  = dyn_event_seq_start,
 757	.next   = dyn_event_seq_next,
 758	.stop   = dyn_event_seq_stop,
 759	.show   = probes_seq_show
 760};
 761
 762static int probes_open(struct inode *inode, struct file *file)
 763{
 764	int ret;
 765
 766	ret = security_locked_down(LOCKDOWN_TRACEFS);
 767	if (ret)
 768		return ret;
 769
 770	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 771		ret = dyn_events_release_all(&trace_uprobe_ops);
 772		if (ret)
 773			return ret;
 774	}
 775
 776	return seq_open(file, &probes_seq_op);
 777}
 778
 779static ssize_t probes_write(struct file *file, const char __user *buffer,
 780			    size_t count, loff_t *ppos)
 781{
 782	return trace_parse_run_command(file, buffer, count, ppos,
 783					create_or_delete_trace_uprobe);
 784}
 785
 786static const struct file_operations uprobe_events_ops = {
 787	.owner		= THIS_MODULE,
 788	.open		= probes_open,
 789	.read		= seq_read,
 790	.llseek		= seq_lseek,
 791	.release	= seq_release,
 792	.write		= probes_write,
 793};
 794
 795/* Probes profiling interfaces */
 796static int probes_profile_seq_show(struct seq_file *m, void *v)
 797{
 798	struct dyn_event *ev = v;
 799	struct trace_uprobe *tu;
 800
 801	if (!is_trace_uprobe(ev))
 802		return 0;
 803
 804	tu = to_trace_uprobe(ev);
 805	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 806			trace_probe_name(&tu->tp), tu->nhit);
 807	return 0;
 808}
 809
 810static const struct seq_operations profile_seq_op = {
 811	.start  = dyn_event_seq_start,
 812	.next   = dyn_event_seq_next,
 813	.stop   = dyn_event_seq_stop,
 814	.show	= probes_profile_seq_show
 815};
 816
 817static int profile_open(struct inode *inode, struct file *file)
 818{
 819	int ret;
 820
 821	ret = security_locked_down(LOCKDOWN_TRACEFS);
 822	if (ret)
 823		return ret;
 824
 825	return seq_open(file, &profile_seq_op);
 826}
 827
 828static const struct file_operations uprobe_profile_ops = {
 829	.owner		= THIS_MODULE,
 830	.open		= profile_open,
 831	.read		= seq_read,
 832	.llseek		= seq_lseek,
 833	.release	= seq_release,
 834};
 835
 836struct uprobe_cpu_buffer {
 837	struct mutex mutex;
 838	void *buf;
 839};
 840static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 841static int uprobe_buffer_refcnt;
 842
 843static int uprobe_buffer_init(void)
 844{
 845	int cpu, err_cpu;
 846
 847	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 848	if (uprobe_cpu_buffer == NULL)
 849		return -ENOMEM;
 850
 851	for_each_possible_cpu(cpu) {
 852		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 853						  GFP_KERNEL, 0);
 854		if (p == NULL) {
 855			err_cpu = cpu;
 856			goto err;
 857		}
 858		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 859		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 860	}
 861
 862	return 0;
 863
 864err:
 865	for_each_possible_cpu(cpu) {
 866		if (cpu == err_cpu)
 867			break;
 868		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 869	}
 870
 871	free_percpu(uprobe_cpu_buffer);
 872	return -ENOMEM;
 873}
 874
 875static int uprobe_buffer_enable(void)
 876{
 877	int ret = 0;
 878
 879	BUG_ON(!mutex_is_locked(&event_mutex));
 880
 881	if (uprobe_buffer_refcnt++ == 0) {
 882		ret = uprobe_buffer_init();
 883		if (ret < 0)
 884			uprobe_buffer_refcnt--;
 885	}
 886
 887	return ret;
 888}
 889
 890static void uprobe_buffer_disable(void)
 891{
 892	int cpu;
 893
 894	BUG_ON(!mutex_is_locked(&event_mutex));
 895
 896	if (--uprobe_buffer_refcnt == 0) {
 897		for_each_possible_cpu(cpu)
 898			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 899							     cpu)->buf);
 900
 901		free_percpu(uprobe_cpu_buffer);
 902		uprobe_cpu_buffer = NULL;
 903	}
 904}
 905
 906static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 907{
 908	struct uprobe_cpu_buffer *ucb;
 909	int cpu;
 910
 911	cpu = raw_smp_processor_id();
 912	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 913
 914	/*
 915	 * Use per-cpu buffers for fastest access, but we might migrate
 916	 * so the mutex makes sure we have sole access to it.
 917	 */
 918	mutex_lock(&ucb->mutex);
 919
 920	return ucb;
 921}
 922
 923static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 924{
 925	mutex_unlock(&ucb->mutex);
 926}
 927
 928static void __uprobe_trace_func(struct trace_uprobe *tu,
 929				unsigned long func, struct pt_regs *regs,
 930				struct uprobe_cpu_buffer *ucb, int dsize,
 931				struct trace_event_file *trace_file)
 932{
 933	struct uprobe_trace_entry_head *entry;
 934	struct trace_buffer *buffer;
 935	struct ring_buffer_event *event;
 936	void *data;
 937	int size, esize;
 938	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 939
 940	WARN_ON(call != trace_file->event_call);
 941
 942	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 943		return;
 944
 945	if (trace_trigger_soft_disabled(trace_file))
 946		return;
 947
 948	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 949	size = esize + tu->tp.size + dsize;
 950	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 951						call->event.type, size, 0, 0);
 952	if (!event)
 953		return;
 954
 955	entry = ring_buffer_event_data(event);
 956	if (is_ret_probe(tu)) {
 957		entry->vaddr[0] = func;
 958		entry->vaddr[1] = instruction_pointer(regs);
 959		data = DATAOF_TRACE_ENTRY(entry, true);
 960	} else {
 961		entry->vaddr[0] = instruction_pointer(regs);
 962		data = DATAOF_TRACE_ENTRY(entry, false);
 963	}
 964
 965	memcpy(data, ucb->buf, tu->tp.size + dsize);
 966
 967	event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 968}
 969
 970/* uprobe handler */
 971static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 972			     struct uprobe_cpu_buffer *ucb, int dsize)
 973{
 974	struct event_file_link *link;
 975
 976	if (is_ret_probe(tu))
 977		return 0;
 978
 979	rcu_read_lock();
 980	trace_probe_for_each_link_rcu(link, &tu->tp)
 981		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 982	rcu_read_unlock();
 983
 984	return 0;
 985}
 986
 987static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 988				 struct pt_regs *regs,
 989				 struct uprobe_cpu_buffer *ucb, int dsize)
 990{
 991	struct event_file_link *link;
 992
 993	rcu_read_lock();
 994	trace_probe_for_each_link_rcu(link, &tu->tp)
 995		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 996	rcu_read_unlock();
 997}
 998
 999/* Event entry printers */
1000static enum print_line_t
1001print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1002{
1003	struct uprobe_trace_entry_head *entry;
1004	struct trace_seq *s = &iter->seq;
1005	struct trace_uprobe *tu;
1006	u8 *data;
1007
1008	entry = (struct uprobe_trace_entry_head *)iter->ent;
1009	tu = trace_uprobe_primary_from_call(
1010		container_of(event, struct trace_event_call, event));
1011	if (unlikely(!tu))
1012		goto out;
1013
1014	if (is_ret_probe(tu)) {
1015		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1016				 trace_probe_name(&tu->tp),
1017				 entry->vaddr[1], entry->vaddr[0]);
1018		data = DATAOF_TRACE_ENTRY(entry, true);
1019	} else {
1020		trace_seq_printf(s, "%s: (0x%lx)",
1021				 trace_probe_name(&tu->tp),
1022				 entry->vaddr[0]);
1023		data = DATAOF_TRACE_ENTRY(entry, false);
1024	}
1025
1026	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1027		goto out;
1028
1029	trace_seq_putc(s, '\n');
1030
1031 out:
1032	return trace_handle_return(s);
1033}
1034
1035typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1036				enum uprobe_filter_ctx ctx,
1037				struct mm_struct *mm);
1038
1039static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1040{
1041	int ret;
1042
1043	tu->consumer.filter = filter;
1044	tu->inode = d_real_inode(tu->path.dentry);
1045
1046	if (tu->ref_ctr_offset)
1047		ret = uprobe_register_refctr(tu->inode, tu->offset,
1048				tu->ref_ctr_offset, &tu->consumer);
1049	else
1050		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1051
1052	if (ret)
1053		tu->inode = NULL;
1054
1055	return ret;
1056}
1057
1058static void __probe_event_disable(struct trace_probe *tp)
1059{
1060	struct trace_probe *pos;
1061	struct trace_uprobe *tu;
1062
1063	tu = container_of(tp, struct trace_uprobe, tp);
1064	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1065
1066	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1067		tu = container_of(pos, struct trace_uprobe, tp);
1068		if (!tu->inode)
1069			continue;
1070
1071		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1072		tu->inode = NULL;
1073	}
1074}
1075
1076static int probe_event_enable(struct trace_event_call *call,
1077			struct trace_event_file *file, filter_func_t filter)
1078{
1079	struct trace_probe *pos, *tp;
1080	struct trace_uprobe *tu;
1081	bool enabled;
1082	int ret;
1083
1084	tp = trace_probe_primary_from_call(call);
1085	if (WARN_ON_ONCE(!tp))
1086		return -ENODEV;
1087	enabled = trace_probe_is_enabled(tp);
1088
1089	/* This may also change "enabled" state */
1090	if (file) {
1091		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1092			return -EINTR;
1093
1094		ret = trace_probe_add_file(tp, file);
1095		if (ret < 0)
1096			return ret;
1097	} else {
1098		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1099			return -EINTR;
1100
1101		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1102	}
1103
1104	tu = container_of(tp, struct trace_uprobe, tp);
1105	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1106
1107	if (enabled)
1108		return 0;
1109
1110	ret = uprobe_buffer_enable();
1111	if (ret)
1112		goto err_flags;
1113
1114	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1115		tu = container_of(pos, struct trace_uprobe, tp);
1116		ret = trace_uprobe_enable(tu, filter);
1117		if (ret) {
1118			__probe_event_disable(tp);
1119			goto err_buffer;
1120		}
1121	}
1122
1123	return 0;
1124
1125 err_buffer:
1126	uprobe_buffer_disable();
1127
1128 err_flags:
1129	if (file)
1130		trace_probe_remove_file(tp, file);
1131	else
1132		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1133
1134	return ret;
1135}
1136
1137static void probe_event_disable(struct trace_event_call *call,
1138				struct trace_event_file *file)
1139{
1140	struct trace_probe *tp;
1141
1142	tp = trace_probe_primary_from_call(call);
1143	if (WARN_ON_ONCE(!tp))
1144		return;
1145
1146	if (!trace_probe_is_enabled(tp))
1147		return;
1148
1149	if (file) {
1150		if (trace_probe_remove_file(tp, file) < 0)
1151			return;
1152
1153		if (trace_probe_is_enabled(tp))
1154			return;
1155	} else
1156		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1157
1158	__probe_event_disable(tp);
1159	uprobe_buffer_disable();
1160}
1161
1162static int uprobe_event_define_fields(struct trace_event_call *event_call)
1163{
1164	int ret, size;
1165	struct uprobe_trace_entry_head field;
1166	struct trace_uprobe *tu;
1167
1168	tu = trace_uprobe_primary_from_call(event_call);
1169	if (unlikely(!tu))
1170		return -ENODEV;
1171
1172	if (is_ret_probe(tu)) {
1173		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1174		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1175		size = SIZEOF_TRACE_ENTRY(true);
1176	} else {
1177		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1178		size = SIZEOF_TRACE_ENTRY(false);
1179	}
1180
1181	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1182}
1183
1184#ifdef CONFIG_PERF_EVENTS
1185static bool
1186__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1187{
1188	struct perf_event *event;
1189
1190	if (filter->nr_systemwide)
1191		return true;
1192
1193	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1194		if (event->hw.target->mm == mm)
1195			return true;
1196	}
1197
1198	return false;
1199}
1200
1201static inline bool
1202trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1203			  struct perf_event *event)
1204{
1205	return __uprobe_perf_filter(filter, event->hw.target->mm);
1206}
1207
1208static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1209				       struct perf_event *event)
1210{
1211	bool done;
1212
1213	write_lock(&filter->rwlock);
1214	if (event->hw.target) {
1215		list_del(&event->hw.tp_list);
1216		done = filter->nr_systemwide ||
1217			(event->hw.target->flags & PF_EXITING) ||
1218			trace_uprobe_filter_event(filter, event);
1219	} else {
1220		filter->nr_systemwide--;
1221		done = filter->nr_systemwide;
1222	}
1223	write_unlock(&filter->rwlock);
1224
1225	return done;
1226}
1227
1228/* This returns true if the filter always covers target mm */
1229static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1230				    struct perf_event *event)
1231{
1232	bool done;
1233
1234	write_lock(&filter->rwlock);
1235	if (event->hw.target) {
1236		/*
1237		 * event->parent != NULL means copy_process(), we can avoid
1238		 * uprobe_apply(). current->mm must be probed and we can rely
1239		 * on dup_mmap() which preserves the already installed bp's.
1240		 *
1241		 * attr.enable_on_exec means that exec/mmap will install the
1242		 * breakpoints we need.
1243		 */
1244		done = filter->nr_systemwide ||
1245			event->parent || event->attr.enable_on_exec ||
1246			trace_uprobe_filter_event(filter, event);
1247		list_add(&event->hw.tp_list, &filter->perf_events);
1248	} else {
1249		done = filter->nr_systemwide;
1250		filter->nr_systemwide++;
1251	}
1252	write_unlock(&filter->rwlock);
1253
1254	return done;
1255}
1256
1257static int uprobe_perf_close(struct trace_event_call *call,
1258			     struct perf_event *event)
1259{
1260	struct trace_probe *pos, *tp;
1261	struct trace_uprobe *tu;
1262	int ret = 0;
1263
1264	tp = trace_probe_primary_from_call(call);
1265	if (WARN_ON_ONCE(!tp))
1266		return -ENODEV;
1267
1268	tu = container_of(tp, struct trace_uprobe, tp);
1269	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1270		return 0;
1271
1272	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1273		tu = container_of(pos, struct trace_uprobe, tp);
1274		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1275		if (ret)
1276			break;
1277	}
1278
1279	return ret;
1280}
1281
1282static int uprobe_perf_open(struct trace_event_call *call,
1283			    struct perf_event *event)
1284{
1285	struct trace_probe *pos, *tp;
1286	struct trace_uprobe *tu;
1287	int err = 0;
1288
1289	tp = trace_probe_primary_from_call(call);
1290	if (WARN_ON_ONCE(!tp))
1291		return -ENODEV;
1292
1293	tu = container_of(tp, struct trace_uprobe, tp);
1294	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1295		return 0;
1296
1297	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1298		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1299		if (err) {
1300			uprobe_perf_close(call, event);
1301			break;
1302		}
1303	}
1304
1305	return err;
1306}
1307
1308static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1309				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1310{
1311	struct trace_uprobe_filter *filter;
1312	struct trace_uprobe *tu;
1313	int ret;
1314
1315	tu = container_of(uc, struct trace_uprobe, consumer);
1316	filter = tu->tp.event->filter;
1317
1318	read_lock(&filter->rwlock);
1319	ret = __uprobe_perf_filter(filter, mm);
1320	read_unlock(&filter->rwlock);
1321
1322	return ret;
1323}
1324
1325static void __uprobe_perf_func(struct trace_uprobe *tu,
1326			       unsigned long func, struct pt_regs *regs,
1327			       struct uprobe_cpu_buffer *ucb, int dsize)
1328{
1329	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1330	struct uprobe_trace_entry_head *entry;
1331	struct hlist_head *head;
1332	void *data;
1333	int size, esize;
1334	int rctx;
1335
 
1336	if (bpf_prog_array_valid(call)) {
1337		u32 ret;
1338
1339		preempt_disable();
1340		ret = trace_call_bpf(call, regs);
1341		preempt_enable();
1342		if (!ret)
1343			return;
1344	}
 
1345
1346	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1347
1348	size = esize + tu->tp.size + dsize;
1349	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1350	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1351		return;
1352
1353	preempt_disable();
1354	head = this_cpu_ptr(call->perf_events);
1355	if (hlist_empty(head))
1356		goto out;
1357
1358	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1359	if (!entry)
1360		goto out;
1361
1362	if (is_ret_probe(tu)) {
1363		entry->vaddr[0] = func;
1364		entry->vaddr[1] = instruction_pointer(regs);
1365		data = DATAOF_TRACE_ENTRY(entry, true);
1366	} else {
1367		entry->vaddr[0] = instruction_pointer(regs);
1368		data = DATAOF_TRACE_ENTRY(entry, false);
1369	}
1370
1371	memcpy(data, ucb->buf, tu->tp.size + dsize);
1372
1373	if (size - esize > tu->tp.size + dsize) {
1374		int len = tu->tp.size + dsize;
1375
1376		memset(data + len, 0, size - esize - len);
1377	}
1378
1379	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1380			      head, NULL);
1381 out:
1382	preempt_enable();
1383}
1384
1385/* uprobe profile handler */
1386static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1387			    struct uprobe_cpu_buffer *ucb, int dsize)
1388{
1389	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1390		return UPROBE_HANDLER_REMOVE;
1391
1392	if (!is_ret_probe(tu))
1393		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1394	return 0;
1395}
1396
1397static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1398				struct pt_regs *regs,
1399				struct uprobe_cpu_buffer *ucb, int dsize)
1400{
1401	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1402}
1403
1404int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1405			const char **filename, u64 *probe_offset,
1406			bool perf_type_tracepoint)
1407{
1408	const char *pevent = trace_event_name(event->tp_event);
1409	const char *group = event->tp_event->class->system;
1410	struct trace_uprobe *tu;
1411
1412	if (perf_type_tracepoint)
1413		tu = find_probe_event(pevent, group);
1414	else
1415		tu = trace_uprobe_primary_from_call(event->tp_event);
1416	if (!tu)
1417		return -EINVAL;
1418
1419	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1420				    : BPF_FD_TYPE_UPROBE;
1421	*filename = tu->filename;
1422	*probe_offset = tu->offset;
1423	return 0;
1424}
1425#endif	/* CONFIG_PERF_EVENTS */
1426
1427static int
1428trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1429		      void *data)
1430{
1431	struct trace_event_file *file = data;
1432
1433	switch (type) {
1434	case TRACE_REG_REGISTER:
1435		return probe_event_enable(event, file, NULL);
1436
1437	case TRACE_REG_UNREGISTER:
1438		probe_event_disable(event, file);
1439		return 0;
1440
1441#ifdef CONFIG_PERF_EVENTS
1442	case TRACE_REG_PERF_REGISTER:
1443		return probe_event_enable(event, NULL, uprobe_perf_filter);
1444
1445	case TRACE_REG_PERF_UNREGISTER:
1446		probe_event_disable(event, NULL);
1447		return 0;
1448
1449	case TRACE_REG_PERF_OPEN:
1450		return uprobe_perf_open(event, data);
1451
1452	case TRACE_REG_PERF_CLOSE:
1453		return uprobe_perf_close(event, data);
1454
1455#endif
1456	default:
1457		return 0;
1458	}
1459}
1460
1461static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1462{
1463	struct trace_uprobe *tu;
1464	struct uprobe_dispatch_data udd;
1465	struct uprobe_cpu_buffer *ucb;
1466	int dsize, esize;
1467	int ret = 0;
1468
1469
1470	tu = container_of(con, struct trace_uprobe, consumer);
1471	tu->nhit++;
1472
1473	udd.tu = tu;
1474	udd.bp_addr = instruction_pointer(regs);
1475
1476	current->utask->vaddr = (unsigned long) &udd;
1477
1478	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1479		return 0;
1480
1481	dsize = __get_data_size(&tu->tp, regs);
1482	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1483
1484	ucb = uprobe_buffer_get();
1485	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1486
1487	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1488		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1489
1490#ifdef CONFIG_PERF_EVENTS
1491	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1492		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1493#endif
1494	uprobe_buffer_put(ucb);
1495	return ret;
1496}
1497
1498static int uretprobe_dispatcher(struct uprobe_consumer *con,
1499				unsigned long func, struct pt_regs *regs)
1500{
1501	struct trace_uprobe *tu;
1502	struct uprobe_dispatch_data udd;
1503	struct uprobe_cpu_buffer *ucb;
1504	int dsize, esize;
1505
1506	tu = container_of(con, struct trace_uprobe, consumer);
1507
1508	udd.tu = tu;
1509	udd.bp_addr = func;
1510
1511	current->utask->vaddr = (unsigned long) &udd;
1512
1513	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1514		return 0;
1515
1516	dsize = __get_data_size(&tu->tp, regs);
1517	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1518
1519	ucb = uprobe_buffer_get();
1520	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1521
1522	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1523		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1524
1525#ifdef CONFIG_PERF_EVENTS
1526	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1527		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1528#endif
1529	uprobe_buffer_put(ucb);
1530	return 0;
1531}
1532
1533static struct trace_event_functions uprobe_funcs = {
1534	.trace		= print_uprobe_event
1535};
1536
1537static struct trace_event_fields uprobe_fields_array[] = {
1538	{ .type = TRACE_FUNCTION_TYPE,
1539	  .define_fields = uprobe_event_define_fields },
1540	{}
1541};
1542
1543static inline void init_trace_event_call(struct trace_uprobe *tu)
1544{
1545	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1546	call->event.funcs = &uprobe_funcs;
1547	call->class->fields_array = uprobe_fields_array;
1548
1549	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1550	call->class->reg = trace_uprobe_register;
1551}
1552
1553static int register_uprobe_event(struct trace_uprobe *tu)
1554{
1555	init_trace_event_call(tu);
1556
1557	return trace_probe_register_event_call(&tu->tp);
1558}
1559
1560static int unregister_uprobe_event(struct trace_uprobe *tu)
1561{
1562	return trace_probe_unregister_event_call(&tu->tp);
1563}
1564
1565#ifdef CONFIG_PERF_EVENTS
1566struct trace_event_call *
1567create_local_trace_uprobe(char *name, unsigned long offs,
1568			  unsigned long ref_ctr_offset, bool is_return)
1569{
 
1570	struct trace_uprobe *tu;
1571	struct path path;
1572	int ret;
1573
1574	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1575	if (ret)
1576		return ERR_PTR(ret);
1577
1578	if (!d_is_reg(path.dentry)) {
1579		path_put(&path);
1580		return ERR_PTR(-EINVAL);
1581	}
1582
1583	/*
1584	 * local trace_kprobes are not added to dyn_event, so they are never
1585	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1586	 * duplicated name "DUMMY_EVENT" here.
1587	 */
1588	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1589				is_return);
1590
1591	if (IS_ERR(tu)) {
1592		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1593			(int)PTR_ERR(tu));
1594		path_put(&path);
1595		return ERR_CAST(tu);
1596	}
1597
1598	tu->offset = offs;
1599	tu->path = path;
1600	tu->ref_ctr_offset = ref_ctr_offset;
1601	tu->filename = kstrdup(name, GFP_KERNEL);
 
 
 
 
 
1602	init_trace_event_call(tu);
1603
1604	if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
 
1605		ret = -ENOMEM;
1606		goto error;
1607	}
1608
1609	return trace_probe_event_call(&tu->tp);
1610error:
1611	free_trace_uprobe(tu);
1612	return ERR_PTR(ret);
1613}
1614
1615void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1616{
1617	struct trace_uprobe *tu;
1618
1619	tu = trace_uprobe_primary_from_call(event_call);
1620
1621	free_trace_uprobe(tu);
1622}
1623#endif /* CONFIG_PERF_EVENTS */
1624
1625/* Make a trace interface for controling probe points */
1626static __init int init_uprobe_trace(void)
1627{
1628	struct dentry *d_tracer;
1629	int ret;
1630
1631	ret = dyn_event_register(&trace_uprobe_ops);
1632	if (ret)
1633		return ret;
1634
1635	d_tracer = tracing_init_dentry();
1636	if (IS_ERR(d_tracer))
1637		return 0;
1638
1639	trace_create_file("uprobe_events", 0644, d_tracer,
1640				    NULL, &uprobe_events_ops);
1641	/* Profile interface */
1642	trace_create_file("uprobe_profile", 0444, d_tracer,
1643				    NULL, &uprobe_profile_ops);
1644	return 0;
1645}
1646
1647fs_initcall(init_uprobe_trace);