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);
v4.17
 
   1/*
   2 * uprobes-based tracing events
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License version 2 as
   6 * published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful,
   9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11 * GNU General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public License
  14 * along with this program; if not, write to the Free Software
  15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  16 *
  17 * Copyright (C) IBM Corporation, 2010-2012
  18 * Author:	Srikar Dronamraju <srikar@linux.vnet.ibm.com>
  19 */
  20#define pr_fmt(fmt)	"trace_kprobe: " fmt
  21
 
 
 
  22#include <linux/module.h>
  23#include <linux/uaccess.h>
  24#include <linux/uprobes.h>
  25#include <linux/namei.h>
  26#include <linux/string.h>
  27#include <linux/rculist.h>
 
  28
 
  29#include "trace_probe.h"
 
  30
  31#define UPROBE_EVENT_SYSTEM	"uprobes"
  32
  33struct uprobe_trace_entry_head {
  34	struct trace_entry	ent;
  35	unsigned long		vaddr[];
  36};
  37
  38#define SIZEOF_TRACE_ENTRY(is_return)			\
  39	(sizeof(struct uprobe_trace_entry_head) +	\
  40	 sizeof(unsigned long) * (is_return ? 2 : 1))
  41
  42#define DATAOF_TRACE_ENTRY(entry, is_return)		\
  43	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  44
  45struct trace_uprobe_filter {
  46	rwlock_t		rwlock;
  47	int			nr_systemwide;
  48	struct list_head	perf_events;
 
 
 
 
 
 
 
 
 
  49};
  50
  51/*
  52 * uprobe event core functions
  53 */
  54struct trace_uprobe {
  55	struct list_head		list;
  56	struct trace_uprobe_filter	filter;
  57	struct uprobe_consumer		consumer;
  58	struct path			path;
  59	struct inode			*inode;
  60	char				*filename;
  61	unsigned long			offset;
 
  62	unsigned long			nhit;
  63	struct trace_probe		tp;
  64};
  65
  66#define SIZEOF_TRACE_UPROBE(n)				\
  67	(offsetof(struct trace_uprobe, tp.args) +	\
  68	(sizeof(struct probe_arg) * (n)))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  69
  70static int register_uprobe_event(struct trace_uprobe *tu);
  71static int unregister_uprobe_event(struct trace_uprobe *tu);
  72
  73static DEFINE_MUTEX(uprobe_lock);
  74static LIST_HEAD(uprobe_list);
  75
  76struct uprobe_dispatch_data {
  77	struct trace_uprobe	*tu;
  78	unsigned long		bp_addr;
  79};
  80
  81static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  82static int uretprobe_dispatcher(struct uprobe_consumer *con,
  83				unsigned long func, struct pt_regs *regs);
  84
  85#ifdef CONFIG_STACK_GROWSUP
  86static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  87{
  88	return addr - (n * sizeof(long));
  89}
  90#else
  91static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  92{
  93	return addr + (n * sizeof(long));
  94}
  95#endif
  96
  97static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  98{
  99	unsigned long ret;
 100	unsigned long addr = user_stack_pointer(regs);
 101
 102	addr = adjust_stack_addr(addr, n);
 103
 104	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 105		return 0;
 106
 107	return ret;
 108}
 109
 110/*
 111 * Uprobes-specific fetch functions
 112 */
 113#define DEFINE_FETCH_stack(type)					\
 114static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,		\
 115					 void *offset, void *dest)	\
 116{									\
 117	*(type *)dest = (type)get_user_stack_nth(regs,			\
 118					      ((unsigned long)offset)); \
 119}
 120DEFINE_BASIC_FETCH_FUNCS(stack)
 121/* No string on the stack entry */
 122#define fetch_stack_string	NULL
 123#define fetch_stack_string_size	NULL
 124
 125#define DEFINE_FETCH_memory(type)					\
 126static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
 127					  void *addr, void *dest)	\
 128{									\
 129	type retval;							\
 130	void __user *vaddr = (void __force __user *) addr;		\
 131									\
 132	if (copy_from_user(&retval, vaddr, sizeof(type)))		\
 133		*(type *)dest = 0;					\
 134	else								\
 135		*(type *) dest = retval;				\
 136}
 137DEFINE_BASIC_FETCH_FUNCS(memory)
 138/*
 139 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 140 * length and relative data location.
 141 */
 142static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 143					    void *addr, void *dest)
 144{
 145	long ret;
 146	u32 rloc = *(u32 *)dest;
 147	int maxlen  = get_rloc_len(rloc);
 148	u8 *dst = get_rloc_data(dest);
 149	void __user *src = (void __force __user *) addr;
 150
 151	if (!maxlen)
 152		return;
 153
 154	ret = strncpy_from_user(dst, src, maxlen);
 155	if (ret == maxlen)
 156		dst[--ret] = '\0';
 157
 158	if (ret < 0) {	/* Failed to fetch string */
 159		((u8 *)get_rloc_data(dest))[0] = '\0';
 160		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 161	} else {
 162		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 
 
 
 
 
 
 163	}
 
 
 164}
 165
 166static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 167						 void *addr, void *dest)
 
 
 
 
 
 
 
 168{
 169	int len;
 170	void __user *vaddr = (void __force __user *) addr;
 171
 172	len = strnlen_user(vaddr, MAX_STRING_SIZE);
 
 
 
 
 
 
 173
 174	if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 175		*(u32 *)dest = 0;
 176	else
 177		*(u32 *)dest = len;
 178}
 179
 180static unsigned long translate_user_vaddr(void *file_offset)
 181{
 182	unsigned long base_addr;
 183	struct uprobe_dispatch_data *udd;
 184
 185	udd = (void *) current->utask->vaddr;
 186
 187	base_addr = udd->bp_addr - udd->tu->offset;
 188	return base_addr + (unsigned long)file_offset;
 189}
 190
 191#define DEFINE_FETCH_file_offset(type)					\
 192static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,	\
 193					       void *offset, void *dest)\
 194{									\
 195	void *vaddr = (void *)translate_user_vaddr(offset);		\
 196									\
 197	FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);		\
 198}
 199DEFINE_BASIC_FETCH_FUNCS(file_offset)
 200DEFINE_FETCH_file_offset(string)
 201DEFINE_FETCH_file_offset(string_size)
 202
 203/* Fetch type information table */
 204static const struct fetch_type uprobes_fetch_type_table[] = {
 205	/* Special types */
 206	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 207					sizeof(u32), 1, "__data_loc char[]"),
 208	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 209					string_size, sizeof(u32), 0, "u32"),
 210	/* Basic types */
 211	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 212	ASSIGN_FETCH_TYPE(u16, u16, 0),
 213	ASSIGN_FETCH_TYPE(u32, u32, 0),
 214	ASSIGN_FETCH_TYPE(u64, u64, 0),
 215	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 216	ASSIGN_FETCH_TYPE(s16, u16, 1),
 217	ASSIGN_FETCH_TYPE(s32, u32, 1),
 218	ASSIGN_FETCH_TYPE(s64, u64, 1),
 219	ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 220	ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 221	ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 222	ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 
 
 
 
 
 
 223
 224	ASSIGN_FETCH_TYPE_END
 225};
 
 226
 227static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 228{
 229	rwlock_init(&filter->rwlock);
 230	filter->nr_systemwide = 0;
 231	INIT_LIST_HEAD(&filter->perf_events);
 232}
 233
 234static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 235{
 236	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 237}
 238
 239static inline bool is_ret_probe(struct trace_uprobe *tu)
 240{
 241	return tu->consumer.ret_handler != NULL;
 242}
 243
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 244/*
 245 * Allocate new trace_uprobe and initialize it (including uprobes).
 246 */
 247static struct trace_uprobe *
 248alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 249{
 250	struct trace_uprobe *tu;
 
 251
 252	if (!event || !is_good_name(event))
 253		return ERR_PTR(-EINVAL);
 254
 255	if (!group || !is_good_name(group))
 256		return ERR_PTR(-EINVAL);
 257
 258	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 259	if (!tu)
 260		return ERR_PTR(-ENOMEM);
 261
 262	tu->tp.call.class = &tu->tp.class;
 263	tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 264	if (!tu->tp.call.name)
 265		goto error;
 266
 267	tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 268	if (!tu->tp.class.system)
 269		goto error;
 270
 271	INIT_LIST_HEAD(&tu->list);
 272	INIT_LIST_HEAD(&tu->tp.files);
 273	tu->consumer.handler = uprobe_dispatcher;
 274	if (is_ret)
 275		tu->consumer.ret_handler = uretprobe_dispatcher;
 276	init_trace_uprobe_filter(&tu->filter);
 277	return tu;
 278
 279error:
 280	kfree(tu->tp.call.name);
 281	kfree(tu);
 282
 283	return ERR_PTR(-ENOMEM);
 284}
 285
 286static void free_trace_uprobe(struct trace_uprobe *tu)
 287{
 288	int i;
 289
 290	for (i = 0; i < tu->tp.nr_args; i++)
 291		traceprobe_free_probe_arg(&tu->tp.args[i]);
 292
 293	path_put(&tu->path);
 294	kfree(tu->tp.call.class->system);
 295	kfree(tu->tp.call.name);
 296	kfree(tu->filename);
 297	kfree(tu);
 298}
 299
 300static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 301{
 
 302	struct trace_uprobe *tu;
 303
 304	list_for_each_entry(tu, &uprobe_list, list)
 305		if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 306		    strcmp(tu->tp.call.class->system, group) == 0)
 307			return tu;
 308
 309	return NULL;
 310}
 311
 312/* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 313static int unregister_trace_uprobe(struct trace_uprobe *tu)
 314{
 315	int ret;
 316
 
 
 
 
 
 
 
 317	ret = unregister_uprobe_event(tu);
 318	if (ret)
 319		return ret;
 320
 321	list_del(&tu->list);
 
 
 322	free_trace_uprobe(tu);
 323	return 0;
 324}
 325
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 326/* Register a trace_uprobe and probe_event */
 327static int register_trace_uprobe(struct trace_uprobe *tu)
 328{
 329	struct trace_uprobe *old_tu;
 330	int ret;
 331
 332	mutex_lock(&uprobe_lock);
 
 
 
 
 333
 334	/* register as an event */
 335	old_tu = find_probe_event(trace_event_name(&tu->tp.call),
 336			tu->tp.call.class->system);
 337	if (old_tu) {
 338		/* delete old event */
 339		ret = unregister_trace_uprobe(old_tu);
 340		if (ret)
 341			goto end;
 
 
 
 
 342	}
 343
 344	ret = register_uprobe_event(tu);
 345	if (ret) {
 346		pr_warn("Failed to register probe event(%d)\n", ret);
 
 
 
 
 347		goto end;
 348	}
 349
 350	list_add_tail(&tu->list, &uprobe_list);
 351
 352end:
 353	mutex_unlock(&uprobe_lock);
 354
 355	return ret;
 356}
 357
 358/*
 359 * Argument syntax:
 360 *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 361 *
 362 *  - Remove uprobe: -:[GRP/]EVENT
 363 */
 364static int create_trace_uprobe(int argc, char **argv)
 365{
 366	struct trace_uprobe *tu;
 367	char *arg, *event, *group, *filename;
 
 368	char buf[MAX_EVENT_NAME_LEN];
 
 
 369	struct path path;
 370	unsigned long offset;
 371	bool is_delete, is_return;
 372	int i, ret;
 373
 374	ret = 0;
 375	is_delete = false;
 376	is_return = false;
 377	event = NULL;
 378	group = NULL;
 379
 380	/* argc must be >= 1 */
 381	if (argv[0][0] == '-')
 382		is_delete = true;
 383	else if (argv[0][0] == 'r')
 384		is_return = true;
 385	else if (argv[0][0] != 'p') {
 386		pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 387		return -EINVAL;
 
 
 388	}
 389
 390	if (argv[0][1] == ':') {
 
 
 
 391		event = &argv[0][2];
 392		arg = strchr(event, '/');
 393
 394		if (arg) {
 395			group = event;
 396			event = arg + 1;
 397			event[-1] = '\0';
 398
 399			if (strlen(group) == 0) {
 400				pr_info("Group name is not specified\n");
 401				return -EINVAL;
 402			}
 403		}
 404		if (strlen(event) == 0) {
 405			pr_info("Event name is not specified\n");
 406			return -EINVAL;
 407		}
 408	}
 409	if (!group)
 410		group = UPROBE_EVENT_SYSTEM;
 411
 412	if (is_delete) {
 413		int ret;
 
 414
 415		if (!event) {
 416			pr_info("Delete command needs an event name.\n");
 417			return -EINVAL;
 418		}
 419		mutex_lock(&uprobe_lock);
 420		tu = find_probe_event(event, group);
 421
 422		if (!tu) {
 423			mutex_unlock(&uprobe_lock);
 424			pr_info("Event %s/%s doesn't exist.\n", group, event);
 425			return -ENOENT;
 426		}
 427		/* delete an event */
 428		ret = unregister_trace_uprobe(tu);
 429		mutex_unlock(&uprobe_lock);
 430		return ret;
 431	}
 432
 433	if (argc < 2) {
 434		pr_info("Probe point is not specified.\n");
 435		return -EINVAL;
 436	}
 437	/* Find the last occurrence, in case the path contains ':' too. */
 438	arg = strrchr(argv[1], ':');
 439	if (!arg)
 440		return -EINVAL;
 441
 442	*arg++ = '\0';
 443	filename = argv[1];
 444	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 445	if (ret)
 
 
 
 446		return ret;
 447
 448	if (!d_is_reg(path.dentry)) {
 
 449		ret = -EINVAL;
 450		goto fail_address_parse;
 451	}
 452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 453	ret = kstrtoul(arg, 0, &offset);
 454	if (ret)
 
 455		goto fail_address_parse;
 
 456
 457	argc -= 2;
 458	argv += 2;
 
 
 
 
 
 
 459
 460	/* setup a probe */
 461	if (!event) {
 462		char *tail;
 463		char *ptr;
 464
 465		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 466		if (!tail) {
 467			ret = -ENOMEM;
 468			goto fail_address_parse;
 469		}
 470
 471		ptr = strpbrk(tail, ".-_");
 472		if (ptr)
 473			*ptr = '\0';
 474
 475		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 476		event = buf;
 477		kfree(tail);
 478	}
 479
 
 
 
 480	tu = alloc_trace_uprobe(group, event, argc, is_return);
 481	if (IS_ERR(tu)) {
 482		pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 483		ret = PTR_ERR(tu);
 
 
 484		goto fail_address_parse;
 485	}
 486	tu->offset = offset;
 
 487	tu->path = path;
 488	tu->filename = kstrdup(filename, GFP_KERNEL);
 489
 490	if (!tu->filename) {
 491		pr_info("Failed to allocate filename.\n");
 492		ret = -ENOMEM;
 493		goto error;
 494	}
 495
 496	/* parse arguments */
 497	ret = 0;
 498	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 499		struct probe_arg *parg = &tu->tp.args[i];
 500
 501		/* Increment count for freeing args in error case */
 502		tu->tp.nr_args++;
 503
 504		/* Parse argument name */
 505		arg = strchr(argv[i], '=');
 506		if (arg) {
 507			*arg++ = '\0';
 508			parg->name = kstrdup(argv[i], GFP_KERNEL);
 509		} else {
 510			arg = argv[i];
 511			/* If argument name is omitted, set "argN" */
 512			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 513			parg->name = kstrdup(buf, GFP_KERNEL);
 514		}
 515
 516		if (!parg->name) {
 517			pr_info("Failed to allocate argument[%d] name.\n", i);
 518			ret = -ENOMEM;
 519			goto error;
 520		}
 521
 522		if (!is_good_name(parg->name)) {
 523			pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 524			ret = -EINVAL;
 525			goto error;
 526		}
 527
 528		if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 529			pr_info("Argument[%d] name '%s' conflicts with "
 530				"another field.\n", i, argv[i]);
 531			ret = -EINVAL;
 532			goto error;
 533		}
 534
 535		/* Parse fetch argument */
 536		ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 537						 is_return, false,
 538						 uprobes_fetch_type_table);
 539		if (ret) {
 540			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 541			goto error;
 542		}
 543	}
 544
 545	ret = register_trace_uprobe(tu);
 546	if (ret)
 547		goto error;
 548	return 0;
 549
 550error:
 551	free_trace_uprobe(tu);
 
 
 552	return ret;
 553
 554fail_address_parse:
 
 555	path_put(&path);
 556
 557	pr_info("Failed to parse address or file.\n");
 558
 559	return ret;
 560}
 561
 562static int cleanup_all_probes(void)
 563{
 564	struct trace_uprobe *tu;
 565	int ret = 0;
 566
 567	mutex_lock(&uprobe_lock);
 568	while (!list_empty(&uprobe_list)) {
 569		tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 570		ret = unregister_trace_uprobe(tu);
 571		if (ret)
 572			break;
 573	}
 574	mutex_unlock(&uprobe_lock);
 575	return ret;
 576}
 577
 578/* Probes listing interfaces */
 579static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 580{
 581	mutex_lock(&uprobe_lock);
 582	return seq_list_start(&uprobe_list, *pos);
 
 
 
 
 
 583}
 584
 585static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 586{
 587	return seq_list_next(v, &uprobe_list, pos);
 588}
 589
 590static void probes_seq_stop(struct seq_file *m, void *v)
 591{
 592	mutex_unlock(&uprobe_lock);
 593}
 594
 595static int probes_seq_show(struct seq_file *m, void *v)
 
 596{
 597	struct trace_uprobe *tu = v;
 598	char c = is_ret_probe(tu) ? 'r' : 'p';
 599	int i;
 600
 601	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, tu->tp.call.class->system,
 602			trace_event_name(&tu->tp.call), tu->filename,
 603			(int)(sizeof(void *) * 2), tu->offset);
 604
 
 
 
 605	for (i = 0; i < tu->tp.nr_args; i++)
 606		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 607
 608	seq_putc(m, '\n');
 609	return 0;
 610}
 611
 
 
 
 
 
 
 
 
 
 
 612static const struct seq_operations probes_seq_op = {
 613	.start	= probes_seq_start,
 614	.next	= probes_seq_next,
 615	.stop	= probes_seq_stop,
 616	.show	= probes_seq_show
 617};
 618
 619static int probes_open(struct inode *inode, struct file *file)
 620{
 621	int ret;
 622
 
 
 
 
 623	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 624		ret = cleanup_all_probes();
 625		if (ret)
 626			return ret;
 627	}
 628
 629	return seq_open(file, &probes_seq_op);
 630}
 631
 632static ssize_t probes_write(struct file *file, const char __user *buffer,
 633			    size_t count, loff_t *ppos)
 634{
 635	return trace_parse_run_command(file, buffer, count, ppos, create_trace_uprobe);
 
 636}
 637
 638static 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 */
 648static int probes_profile_seq_show(struct seq_file *m, void *v)
 649{
 650	struct trace_uprobe *tu = v;
 
 
 
 
 651
 
 652	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 653			trace_event_name(&tu->tp.call), tu->nhit);
 654	return 0;
 655}
 656
 657static const struct seq_operations profile_seq_op = {
 658	.start	= probes_seq_start,
 659	.next	= probes_seq_next,
 660	.stop	= probes_seq_stop,
 661	.show	= probes_profile_seq_show
 662};
 663
 664static int profile_open(struct inode *inode, struct file *file)
 665{
 
 
 
 
 
 
 666	return seq_open(file, &profile_seq_op);
 667}
 668
 669static const struct file_operations uprobe_profile_ops = {
 670	.owner		= THIS_MODULE,
 671	.open		= profile_open,
 672	.read		= seq_read,
 673	.llseek		= seq_lseek,
 674	.release	= seq_release,
 675};
 676
 677struct uprobe_cpu_buffer {
 678	struct mutex mutex;
 679	void *buf;
 680};
 681static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 682static int uprobe_buffer_refcnt;
 683
 684static int uprobe_buffer_init(void)
 685{
 686	int cpu, err_cpu;
 687
 688	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 689	if (uprobe_cpu_buffer == NULL)
 690		return -ENOMEM;
 691
 692	for_each_possible_cpu(cpu) {
 693		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 694						  GFP_KERNEL, 0);
 695		if (p == NULL) {
 696			err_cpu = cpu;
 697			goto err;
 698		}
 699		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 700		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 701	}
 702
 703	return 0;
 704
 705err:
 706	for_each_possible_cpu(cpu) {
 707		if (cpu == err_cpu)
 708			break;
 709		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 710	}
 711
 712	free_percpu(uprobe_cpu_buffer);
 713	return -ENOMEM;
 714}
 715
 716static int uprobe_buffer_enable(void)
 717{
 718	int ret = 0;
 719
 720	BUG_ON(!mutex_is_locked(&event_mutex));
 721
 722	if (uprobe_buffer_refcnt++ == 0) {
 723		ret = uprobe_buffer_init();
 724		if (ret < 0)
 725			uprobe_buffer_refcnt--;
 726	}
 727
 728	return ret;
 729}
 730
 731static void uprobe_buffer_disable(void)
 732{
 733	int cpu;
 734
 735	BUG_ON(!mutex_is_locked(&event_mutex));
 736
 737	if (--uprobe_buffer_refcnt == 0) {
 738		for_each_possible_cpu(cpu)
 739			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 740							     cpu)->buf);
 741
 742		free_percpu(uprobe_cpu_buffer);
 743		uprobe_cpu_buffer = NULL;
 744	}
 745}
 746
 747static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 748{
 749	struct uprobe_cpu_buffer *ucb;
 750	int cpu;
 751
 752	cpu = raw_smp_processor_id();
 753	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 754
 755	/*
 756	 * Use per-cpu buffers for fastest access, but we might migrate
 757	 * so the mutex makes sure we have sole access to it.
 758	 */
 759	mutex_lock(&ucb->mutex);
 760
 761	return ucb;
 762}
 763
 764static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 765{
 766	mutex_unlock(&ucb->mutex);
 767}
 768
 769static void __uprobe_trace_func(struct trace_uprobe *tu,
 770				unsigned long func, struct pt_regs *regs,
 771				struct uprobe_cpu_buffer *ucb, int dsize,
 772				struct trace_event_file *trace_file)
 773{
 774	struct uprobe_trace_entry_head *entry;
 775	struct ring_buffer_event *event;
 776	struct ring_buffer *buffer;
 777	void *data;
 778	int size, esize;
 779	struct trace_event_call *call = &tu->tp.call;
 780
 781	WARN_ON(call != trace_file->event_call);
 782
 783	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 784		return;
 785
 786	if (trace_trigger_soft_disabled(trace_file))
 787		return;
 788
 789	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 790	size = esize + tu->tp.size + dsize;
 791	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 792						call->event.type, size, 0, 0);
 793	if (!event)
 794		return;
 795
 796	entry = ring_buffer_event_data(event);
 797	if (is_ret_probe(tu)) {
 798		entry->vaddr[0] = func;
 799		entry->vaddr[1] = instruction_pointer(regs);
 800		data = DATAOF_TRACE_ENTRY(entry, true);
 801	} else {
 802		entry->vaddr[0] = instruction_pointer(regs);
 803		data = DATAOF_TRACE_ENTRY(entry, false);
 804	}
 805
 806	memcpy(data, ucb->buf, tu->tp.size + dsize);
 807
 808	event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 809}
 810
 811/* uprobe handler */
 812static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 813			     struct uprobe_cpu_buffer *ucb, int dsize)
 814{
 815	struct event_file_link *link;
 816
 817	if (is_ret_probe(tu))
 818		return 0;
 819
 820	rcu_read_lock();
 821	list_for_each_entry_rcu(link, &tu->tp.files, list)
 822		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 823	rcu_read_unlock();
 824
 825	return 0;
 826}
 827
 828static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 829				 struct pt_regs *regs,
 830				 struct uprobe_cpu_buffer *ucb, int dsize)
 831{
 832	struct event_file_link *link;
 833
 834	rcu_read_lock();
 835	list_for_each_entry_rcu(link, &tu->tp.files, list)
 836		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 837	rcu_read_unlock();
 838}
 839
 840/* Event entry printers */
 841static enum print_line_t
 842print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 843{
 844	struct uprobe_trace_entry_head *entry;
 845	struct trace_seq *s = &iter->seq;
 846	struct trace_uprobe *tu;
 847	u8 *data;
 848	int i;
 849
 850	entry = (struct uprobe_trace_entry_head *)iter->ent;
 851	tu = container_of(event, struct trace_uprobe, tp.call.event);
 
 
 
 852
 853	if (is_ret_probe(tu)) {
 854		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 855				 trace_event_name(&tu->tp.call),
 856				 entry->vaddr[1], entry->vaddr[0]);
 857		data = DATAOF_TRACE_ENTRY(entry, true);
 858	} else {
 859		trace_seq_printf(s, "%s: (0x%lx)",
 860				 trace_event_name(&tu->tp.call),
 861				 entry->vaddr[0]);
 862		data = DATAOF_TRACE_ENTRY(entry, false);
 863	}
 864
 865	for (i = 0; i < tu->tp.nr_args; i++) {
 866		struct probe_arg *parg = &tu->tp.args[i];
 867
 868		if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 869			goto out;
 870	}
 871
 872	trace_seq_putc(s, '\n');
 873
 874 out:
 875	return trace_handle_return(s);
 876}
 877
 878typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 879				enum uprobe_filter_ctx ctx,
 880				struct mm_struct *mm);
 881
 882static int
 883probe_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	ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 921	if (ret)
 922		goto err_buffer;
 
 
 923
 924	return 0;
 925
 926 err_buffer:
 927	uprobe_buffer_disable();
 928
 929 err_flags:
 930	if (file) {
 931		list_del(&link->list);
 932		kfree(link);
 933		tu->tp.flags &= ~TP_FLAG_TRACE;
 934	} else {
 935		tu->tp.flags &= ~TP_FLAG_PROFILE;
 936	}
 937	return ret;
 938}
 939
 940static void
 941probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 942{
 943	if (!trace_probe_is_enabled(&tu->tp))
 
 
 
 
 
 
 944		return;
 945
 946	if (file) {
 947		struct event_file_link *link;
 948
 949		link = find_event_file_link(&tu->tp, file);
 950		if (!link)
 951			return;
 952
 953		list_del_rcu(&link->list);
 954		/* synchronize with u{,ret}probe_trace_func */
 955		synchronize_sched();
 956		kfree(link);
 957
 958		if (!list_empty(&tu->tp.files))
 959			return;
 960	}
 961
 962	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 963
 964	uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 965	tu->inode = NULL;
 966	tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 967
 
 968	uprobe_buffer_disable();
 969}
 970
 971static int uprobe_event_define_fields(struct trace_event_call *event_call)
 972{
 973	int ret, i, size;
 974	struct uprobe_trace_entry_head field;
 975	struct trace_uprobe *tu = event_call->data;
 
 
 
 
 976
 977	if (is_ret_probe(tu)) {
 978		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 979		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 980		size = SIZEOF_TRACE_ENTRY(true);
 981	} else {
 982		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 983		size = SIZEOF_TRACE_ENTRY(false);
 984	}
 985	/* Set argument names as fields */
 986	for (i = 0; i < tu->tp.nr_args; i++) {
 987		struct probe_arg *parg = &tu->tp.args[i];
 988
 989		ret = trace_define_field(event_call, parg->type->fmttype,
 990					 parg->name, size + parg->offset,
 991					 parg->type->size, parg->type->is_signed,
 992					 FILTER_OTHER);
 993
 994		if (ret)
 995			return ret;
 996	}
 997	return 0;
 998}
 999
1000#ifdef CONFIG_PERF_EVENTS
1001static bool
1002__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1003{
1004	struct perf_event *event;
1005
1006	if (filter->nr_systemwide)
1007		return true;
1008
1009	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1010		if (event->hw.target->mm == mm)
1011			return true;
1012	}
1013
1014	return false;
1015}
1016
1017static inline bool
1018uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 
1019{
1020	return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1021}
1022
1023static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 
1024{
1025	bool done;
1026
1027	write_lock(&tu->filter.rwlock);
1028	if (event->hw.target) {
1029		list_del(&event->hw.tp_list);
1030		done = tu->filter.nr_systemwide ||
1031			(event->hw.target->flags & PF_EXITING) ||
1032			uprobe_filter_event(tu, event);
1033	} else {
1034		tu->filter.nr_systemwide--;
1035		done = tu->filter.nr_systemwide;
1036	}
1037	write_unlock(&tu->filter.rwlock);
1038
1039	if (!done)
1040		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1041
1042	return 0;
1043}
1044
1045static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 
 
1046{
1047	bool done;
1048	int err;
1049
1050	write_lock(&tu->filter.rwlock);
1051	if (event->hw.target) {
1052		/*
1053		 * event->parent != NULL means copy_process(), we can avoid
1054		 * uprobe_apply(). current->mm must be probed and we can rely
1055		 * on dup_mmap() which preserves the already installed bp's.
1056		 *
1057		 * attr.enable_on_exec means that exec/mmap will install the
1058		 * breakpoints we need.
1059		 */
1060		done = tu->filter.nr_systemwide ||
1061			event->parent || event->attr.enable_on_exec ||
1062			uprobe_filter_event(tu, event);
1063		list_add(&event->hw.tp_list, &tu->filter.perf_events);
1064	} else {
1065		done = tu->filter.nr_systemwide;
1066		tu->filter.nr_systemwide++;
1067	}
1068	write_unlock(&tu->filter.rwlock);
 
 
 
1069
1070	err = 0;
1071	if (!done) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1072		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1073		if (err)
1074			uprobe_perf_close(tu, event);
 
 
1075	}
 
1076	return err;
1077}
1078
1079static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1080				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1081{
 
1082	struct trace_uprobe *tu;
1083	int ret;
1084
1085	tu = container_of(uc, struct trace_uprobe, consumer);
1086	read_lock(&tu->filter.rwlock);
1087	ret = __uprobe_perf_filter(&tu->filter, mm);
1088	read_unlock(&tu->filter.rwlock);
 
 
1089
1090	return ret;
1091}
1092
1093static void __uprobe_perf_func(struct trace_uprobe *tu,
1094			       unsigned long func, struct pt_regs *regs,
1095			       struct uprobe_cpu_buffer *ucb, int dsize)
1096{
1097	struct trace_event_call *call = &tu->tp.call;
1098	struct uprobe_trace_entry_head *entry;
1099	struct hlist_head *head;
1100	void *data;
1101	int size, esize;
1102	int rctx;
1103
1104	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1105		return;
 
 
 
 
 
 
 
1106
1107	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1108
1109	size = esize + tu->tp.size + dsize;
1110	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1111	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1112		return;
1113
1114	preempt_disable();
1115	head = this_cpu_ptr(call->perf_events);
1116	if (hlist_empty(head))
1117		goto out;
1118
1119	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1120	if (!entry)
1121		goto out;
1122
1123	if (is_ret_probe(tu)) {
1124		entry->vaddr[0] = func;
1125		entry->vaddr[1] = instruction_pointer(regs);
1126		data = DATAOF_TRACE_ENTRY(entry, true);
1127	} else {
1128		entry->vaddr[0] = instruction_pointer(regs);
1129		data = DATAOF_TRACE_ENTRY(entry, false);
1130	}
1131
1132	memcpy(data, ucb->buf, tu->tp.size + dsize);
1133
1134	if (size - esize > tu->tp.size + dsize) {
1135		int len = tu->tp.size + dsize;
1136
1137		memset(data + len, 0, size - esize - len);
1138	}
1139
1140	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1141			      head, NULL);
1142 out:
1143	preempt_enable();
1144}
1145
1146/* uprobe profile handler */
1147static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1148			    struct uprobe_cpu_buffer *ucb, int dsize)
1149{
1150	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1151		return UPROBE_HANDLER_REMOVE;
1152
1153	if (!is_ret_probe(tu))
1154		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1155	return 0;
1156}
1157
1158static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1159				struct pt_regs *regs,
1160				struct uprobe_cpu_buffer *ucb, int dsize)
1161{
1162	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1163}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1164#endif	/* CONFIG_PERF_EVENTS */
1165
1166static int
1167trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1168		      void *data)
1169{
1170	struct trace_uprobe *tu = event->data;
1171	struct trace_event_file *file = data;
1172
1173	switch (type) {
1174	case TRACE_REG_REGISTER:
1175		return probe_event_enable(tu, file, NULL);
1176
1177	case TRACE_REG_UNREGISTER:
1178		probe_event_disable(tu, file);
1179		return 0;
1180
1181#ifdef CONFIG_PERF_EVENTS
1182	case TRACE_REG_PERF_REGISTER:
1183		return probe_event_enable(tu, NULL, uprobe_perf_filter);
1184
1185	case TRACE_REG_PERF_UNREGISTER:
1186		probe_event_disable(tu, NULL);
1187		return 0;
1188
1189	case TRACE_REG_PERF_OPEN:
1190		return uprobe_perf_open(tu, data);
1191
1192	case TRACE_REG_PERF_CLOSE:
1193		return uprobe_perf_close(tu, data);
1194
1195#endif
1196	default:
1197		return 0;
1198	}
1199	return 0;
1200}
1201
1202static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1203{
1204	struct trace_uprobe *tu;
1205	struct uprobe_dispatch_data udd;
1206	struct uprobe_cpu_buffer *ucb;
1207	int dsize, esize;
1208	int ret = 0;
1209
1210
1211	tu = container_of(con, struct trace_uprobe, consumer);
1212	tu->nhit++;
1213
1214	udd.tu = tu;
1215	udd.bp_addr = instruction_pointer(regs);
1216
1217	current->utask->vaddr = (unsigned long) &udd;
1218
1219	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1220		return 0;
1221
1222	dsize = __get_data_size(&tu->tp, regs);
1223	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1224
1225	ucb = uprobe_buffer_get();
1226	store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1227
1228	if (tu->tp.flags & TP_FLAG_TRACE)
1229		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1230
1231#ifdef CONFIG_PERF_EVENTS
1232	if (tu->tp.flags & TP_FLAG_PROFILE)
1233		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1234#endif
1235	uprobe_buffer_put(ucb);
1236	return ret;
1237}
1238
1239static int uretprobe_dispatcher(struct uprobe_consumer *con,
1240				unsigned long func, struct pt_regs *regs)
1241{
1242	struct trace_uprobe *tu;
1243	struct uprobe_dispatch_data udd;
1244	struct uprobe_cpu_buffer *ucb;
1245	int dsize, esize;
1246
1247	tu = container_of(con, struct trace_uprobe, consumer);
1248
1249	udd.tu = tu;
1250	udd.bp_addr = func;
1251
1252	current->utask->vaddr = (unsigned long) &udd;
1253
1254	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1255		return 0;
1256
1257	dsize = __get_data_size(&tu->tp, regs);
1258	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1259
1260	ucb = uprobe_buffer_get();
1261	store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1262
1263	if (tu->tp.flags & TP_FLAG_TRACE)
1264		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1265
1266#ifdef CONFIG_PERF_EVENTS
1267	if (tu->tp.flags & TP_FLAG_PROFILE)
1268		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1269#endif
1270	uprobe_buffer_put(ucb);
1271	return 0;
1272}
1273
1274static struct trace_event_functions uprobe_funcs = {
1275	.trace		= print_uprobe_event
1276};
1277
1278static inline void init_trace_event_call(struct trace_uprobe *tu,
1279					 struct trace_event_call *call)
 
 
 
 
 
1280{
1281	INIT_LIST_HEAD(&call->class->fields);
1282	call->event.funcs = &uprobe_funcs;
1283	call->class->define_fields = uprobe_event_define_fields;
1284
1285	call->flags = TRACE_EVENT_FL_UPROBE;
1286	call->class->reg = trace_uprobe_register;
1287	call->data = tu;
1288}
1289
1290static int register_uprobe_event(struct trace_uprobe *tu)
1291{
1292	struct trace_event_call *call = &tu->tp.call;
1293	int ret = 0;
1294
1295	init_trace_event_call(tu, call);
1296
1297	if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1298		return -ENOMEM;
1299
1300	ret = register_trace_event(&call->event);
1301	if (!ret) {
1302		kfree(call->print_fmt);
1303		return -ENODEV;
1304	}
1305
1306	ret = trace_add_event_call(call);
1307
1308	if (ret) {
1309		pr_info("Failed to register uprobe event: %s\n",
1310			trace_event_name(call));
1311		kfree(call->print_fmt);
1312		unregister_trace_event(&call->event);
1313	}
1314
1315	return ret;
1316}
1317
1318static int unregister_uprobe_event(struct trace_uprobe *tu)
1319{
1320	int ret;
1321
1322	/* tu->event is unregistered in trace_remove_event_call() */
1323	ret = trace_remove_event_call(&tu->tp.call);
1324	if (ret)
1325		return ret;
1326	kfree(tu->tp.call.print_fmt);
1327	tu->tp.call.print_fmt = NULL;
1328	return 0;
1329}
1330
1331#ifdef CONFIG_PERF_EVENTS
1332struct trace_event_call *
1333create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
 
1334{
 
1335	struct trace_uprobe *tu;
1336	struct path path;
1337	int ret;
1338
1339	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1340	if (ret)
1341		return ERR_PTR(ret);
1342
1343	if (!d_is_reg(path.dentry)) {
1344		path_put(&path);
1345		return ERR_PTR(-EINVAL);
1346	}
1347
1348	/*
1349	 * local trace_kprobes are not added to probe_list, so they are never
1350	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1351	 * duplicated name "DUMMY_EVENT" here.
1352	 */
1353	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1354				is_return);
1355
1356	if (IS_ERR(tu)) {
1357		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1358			(int)PTR_ERR(tu));
1359		path_put(&path);
1360		return ERR_CAST(tu);
1361	}
1362
1363	tu->offset = offs;
1364	tu->path = path;
 
1365	tu->filename = kstrdup(name, GFP_KERNEL);
1366	init_trace_event_call(tu, &tu->tp.call);
 
 
 
 
 
1367
1368	if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
 
1369		ret = -ENOMEM;
1370		goto error;
1371	}
1372
1373	return &tu->tp.call;
1374error:
1375	free_trace_uprobe(tu);
1376	return ERR_PTR(ret);
1377}
1378
1379void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1380{
1381	struct trace_uprobe *tu;
1382
1383	tu = container_of(event_call, struct trace_uprobe, tp.call);
1384
1385	kfree(tu->tp.call.print_fmt);
1386	tu->tp.call.print_fmt = NULL;
1387
1388	free_trace_uprobe(tu);
1389}
1390#endif /* CONFIG_PERF_EVENTS */
1391
1392/* Make a trace interface for controling probe points */
1393static __init int init_uprobe_trace(void)
1394{
1395	struct dentry *d_tracer;
1396
1397	d_tracer = tracing_init_dentry();
1398	if (IS_ERR(d_tracer))
 
 
 
 
1399		return 0;
1400
1401	trace_create_file("uprobe_events", 0644, d_tracer,
1402				    NULL, &uprobe_events_ops);
1403	/* Profile interface */
1404	trace_create_file("uprobe_profile", 0444, d_tracer,
1405				    NULL, &uprobe_profile_ops);
1406	return 0;
1407}
1408
1409fs_initcall(init_uprobe_trace);