Linux Audio

Check our new training course

Loading...
v3.15
   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
  21#include <linux/module.h>
  22#include <linux/uaccess.h>
  23#include <linux/uprobes.h>
  24#include <linux/namei.h>
  25#include <linux/string.h>
 
  26
  27#include "trace_probe.h"
  28
  29#define UPROBE_EVENT_SYSTEM	"uprobes"
  30
  31struct uprobe_trace_entry_head {
  32	struct trace_entry	ent;
  33	unsigned long		vaddr[];
  34};
  35
  36#define SIZEOF_TRACE_ENTRY(is_return)			\
  37	(sizeof(struct uprobe_trace_entry_head) +	\
  38	 sizeof(unsigned long) * (is_return ? 2 : 1))
  39
  40#define DATAOF_TRACE_ENTRY(entry, is_return)		\
  41	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  42
  43struct trace_uprobe_filter {
  44	rwlock_t		rwlock;
  45	int			nr_systemwide;
  46	struct list_head	perf_events;
  47};
  48
  49/*
  50 * uprobe event core functions
  51 */
  52struct trace_uprobe {
  53	struct list_head		list;
  54	struct trace_uprobe_filter	filter;
  55	struct uprobe_consumer		consumer;
 
  56	struct inode			*inode;
  57	char				*filename;
  58	unsigned long			offset;
  59	unsigned long			nhit;
  60	struct trace_probe		tp;
  61};
  62
  63#define SIZEOF_TRACE_UPROBE(n)				\
  64	(offsetof(struct trace_uprobe, tp.args) +	\
  65	(sizeof(struct probe_arg) * (n)))
  66
  67static int register_uprobe_event(struct trace_uprobe *tu);
  68static int unregister_uprobe_event(struct trace_uprobe *tu);
  69
  70static DEFINE_MUTEX(uprobe_lock);
  71static LIST_HEAD(uprobe_list);
  72
  73struct uprobe_dispatch_data {
  74	struct trace_uprobe	*tu;
  75	unsigned long		bp_addr;
  76};
  77
  78static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  79static int uretprobe_dispatcher(struct uprobe_consumer *con,
  80				unsigned long func, struct pt_regs *regs);
  81
  82#ifdef CONFIG_STACK_GROWSUP
  83static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  84{
  85	return addr - (n * sizeof(long));
  86}
  87#else
  88static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  89{
  90	return addr + (n * sizeof(long));
  91}
  92#endif
  93
  94static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  95{
  96	unsigned long ret;
  97	unsigned long addr = user_stack_pointer(regs);
  98
  99	addr = adjust_stack_addr(addr, n);
 100
 101	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 102		return 0;
 103
 104	return ret;
 105}
 106
 107/*
 108 * Uprobes-specific fetch functions
 109 */
 110#define DEFINE_FETCH_stack(type)					\
 111static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
 112					  void *offset, void *dest)	\
 113{									\
 114	*(type *)dest = (type)get_user_stack_nth(regs,			\
 115					      ((unsigned long)offset)); \
 116}
 117DEFINE_BASIC_FETCH_FUNCS(stack)
 118/* No string on the stack entry */
 119#define fetch_stack_string	NULL
 120#define fetch_stack_string_size	NULL
 121
 122#define DEFINE_FETCH_memory(type)					\
 123static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
 124						void *addr, void *dest) \
 125{									\
 126	type retval;							\
 127	void __user *vaddr = (void __force __user *) addr;		\
 128									\
 129	if (copy_from_user(&retval, vaddr, sizeof(type)))		\
 130		*(type *)dest = 0;					\
 131	else								\
 132		*(type *) dest = retval;				\
 133}
 134DEFINE_BASIC_FETCH_FUNCS(memory)
 135/*
 136 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 137 * length and relative data location.
 138 */
 139static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 140						      void *addr, void *dest)
 141{
 142	long ret;
 143	u32 rloc = *(u32 *)dest;
 144	int maxlen  = get_rloc_len(rloc);
 145	u8 *dst = get_rloc_data(dest);
 146	void __user *src = (void __force __user *) addr;
 147
 148	if (!maxlen)
 149		return;
 150
 151	ret = strncpy_from_user(dst, src, maxlen);
 
 
 152
 153	if (ret < 0) {	/* Failed to fetch string */
 154		((u8 *)get_rloc_data(dest))[0] = '\0';
 155		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 156	} else {
 157		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 158	}
 159}
 160
 161static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 162						      void *addr, void *dest)
 163{
 164	int len;
 165	void __user *vaddr = (void __force __user *) addr;
 166
 167	len = strnlen_user(vaddr, MAX_STRING_SIZE);
 168
 169	if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 170		*(u32 *)dest = 0;
 171	else
 172		*(u32 *)dest = len;
 173}
 174
 175static unsigned long translate_user_vaddr(void *file_offset)
 176{
 177	unsigned long base_addr;
 178	struct uprobe_dispatch_data *udd;
 179
 180	udd = (void *) current->utask->vaddr;
 181
 182	base_addr = udd->bp_addr - udd->tu->offset;
 183	return base_addr + (unsigned long)file_offset;
 184}
 185
 186#define DEFINE_FETCH_file_offset(type)					\
 187static __kprobes void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,\
 188					void *offset, void *dest) 	\
 189{									\
 190	void *vaddr = (void *)translate_user_vaddr(offset);		\
 191									\
 192	FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);		\
 193}
 194DEFINE_BASIC_FETCH_FUNCS(file_offset)
 195DEFINE_FETCH_file_offset(string)
 196DEFINE_FETCH_file_offset(string_size)
 197
 198/* Fetch type information table */
 199const struct fetch_type uprobes_fetch_type_table[] = {
 200	/* Special types */
 201	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 202					sizeof(u32), 1, "__data_loc char[]"),
 203	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 204					string_size, sizeof(u32), 0, "u32"),
 205	/* Basic types */
 206	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 207	ASSIGN_FETCH_TYPE(u16, u16, 0),
 208	ASSIGN_FETCH_TYPE(u32, u32, 0),
 209	ASSIGN_FETCH_TYPE(u64, u64, 0),
 210	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 211	ASSIGN_FETCH_TYPE(s16, u16, 1),
 212	ASSIGN_FETCH_TYPE(s32, u32, 1),
 213	ASSIGN_FETCH_TYPE(s64, u64, 1),
 
 
 
 
 214
 215	ASSIGN_FETCH_TYPE_END
 216};
 217
 218static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 219{
 220	rwlock_init(&filter->rwlock);
 221	filter->nr_systemwide = 0;
 222	INIT_LIST_HEAD(&filter->perf_events);
 223}
 224
 225static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 226{
 227	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 228}
 229
 230static inline bool is_ret_probe(struct trace_uprobe *tu)
 231{
 232	return tu->consumer.ret_handler != NULL;
 233}
 234
 235/*
 236 * Allocate new trace_uprobe and initialize it (including uprobes).
 237 */
 238static struct trace_uprobe *
 239alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 240{
 241	struct trace_uprobe *tu;
 242
 243	if (!event || !is_good_name(event))
 244		return ERR_PTR(-EINVAL);
 245
 246	if (!group || !is_good_name(group))
 247		return ERR_PTR(-EINVAL);
 248
 249	tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 250	if (!tu)
 251		return ERR_PTR(-ENOMEM);
 252
 253	tu->tp.call.class = &tu->tp.class;
 254	tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 255	if (!tu->tp.call.name)
 256		goto error;
 257
 258	tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 259	if (!tu->tp.class.system)
 260		goto error;
 261
 262	INIT_LIST_HEAD(&tu->list);
 263	INIT_LIST_HEAD(&tu->tp.files);
 264	tu->consumer.handler = uprobe_dispatcher;
 265	if (is_ret)
 266		tu->consumer.ret_handler = uretprobe_dispatcher;
 267	init_trace_uprobe_filter(&tu->filter);
 268	tu->tp.call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
 269	return tu;
 270
 271error:
 272	kfree(tu->tp.call.name);
 273	kfree(tu);
 274
 275	return ERR_PTR(-ENOMEM);
 276}
 277
 278static void free_trace_uprobe(struct trace_uprobe *tu)
 279{
 280	int i;
 281
 282	for (i = 0; i < tu->tp.nr_args; i++)
 283		traceprobe_free_probe_arg(&tu->tp.args[i]);
 284
 285	iput(tu->inode);
 286	kfree(tu->tp.call.class->system);
 287	kfree(tu->tp.call.name);
 288	kfree(tu->filename);
 289	kfree(tu);
 290}
 291
 292static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 293{
 294	struct trace_uprobe *tu;
 295
 296	list_for_each_entry(tu, &uprobe_list, list)
 297		if (strcmp(ftrace_event_name(&tu->tp.call), event) == 0 &&
 298		    strcmp(tu->tp.call.class->system, group) == 0)
 299			return tu;
 300
 301	return NULL;
 302}
 303
 304/* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 305static int unregister_trace_uprobe(struct trace_uprobe *tu)
 306{
 307	int ret;
 308
 309	ret = unregister_uprobe_event(tu);
 310	if (ret)
 311		return ret;
 312
 313	list_del(&tu->list);
 314	free_trace_uprobe(tu);
 315	return 0;
 316}
 317
 318/* Register a trace_uprobe and probe_event */
 319static int register_trace_uprobe(struct trace_uprobe *tu)
 320{
 321	struct trace_uprobe *old_tu;
 322	int ret;
 323
 324	mutex_lock(&uprobe_lock);
 325
 326	/* register as an event */
 327	old_tu = find_probe_event(ftrace_event_name(&tu->tp.call),
 328			tu->tp.call.class->system);
 329	if (old_tu) {
 330		/* delete old event */
 331		ret = unregister_trace_uprobe(old_tu);
 332		if (ret)
 333			goto end;
 334	}
 335
 336	ret = register_uprobe_event(tu);
 337	if (ret) {
 338		pr_warning("Failed to register probe event(%d)\n", ret);
 339		goto end;
 340	}
 341
 342	list_add_tail(&tu->list, &uprobe_list);
 343
 344end:
 345	mutex_unlock(&uprobe_lock);
 346
 347	return ret;
 348}
 349
 350/*
 351 * Argument syntax:
 352 *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 353 *
 354 *  - Remove uprobe: -:[GRP/]EVENT
 355 */
 356static int create_trace_uprobe(int argc, char **argv)
 357{
 358	struct trace_uprobe *tu;
 359	struct inode *inode;
 360	char *arg, *event, *group, *filename;
 361	char buf[MAX_EVENT_NAME_LEN];
 362	struct path path;
 363	unsigned long offset;
 364	bool is_delete, is_return;
 365	int i, ret;
 366
 367	inode = NULL;
 368	ret = 0;
 369	is_delete = false;
 370	is_return = false;
 371	event = NULL;
 372	group = NULL;
 373
 374	/* argc must be >= 1 */
 375	if (argv[0][0] == '-')
 376		is_delete = true;
 377	else if (argv[0][0] == 'r')
 378		is_return = true;
 379	else if (argv[0][0] != 'p') {
 380		pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 381		return -EINVAL;
 382	}
 383
 384	if (argv[0][1] == ':') {
 385		event = &argv[0][2];
 386		arg = strchr(event, '/');
 387
 388		if (arg) {
 389			group = event;
 390			event = arg + 1;
 391			event[-1] = '\0';
 392
 393			if (strlen(group) == 0) {
 394				pr_info("Group name is not specified\n");
 395				return -EINVAL;
 396			}
 397		}
 398		if (strlen(event) == 0) {
 399			pr_info("Event name is not specified\n");
 400			return -EINVAL;
 401		}
 402	}
 403	if (!group)
 404		group = UPROBE_EVENT_SYSTEM;
 405
 406	if (is_delete) {
 407		int ret;
 408
 409		if (!event) {
 410			pr_info("Delete command needs an event name.\n");
 411			return -EINVAL;
 412		}
 413		mutex_lock(&uprobe_lock);
 414		tu = find_probe_event(event, group);
 415
 416		if (!tu) {
 417			mutex_unlock(&uprobe_lock);
 418			pr_info("Event %s/%s doesn't exist.\n", group, event);
 419			return -ENOENT;
 420		}
 421		/* delete an event */
 422		ret = unregister_trace_uprobe(tu);
 423		mutex_unlock(&uprobe_lock);
 424		return ret;
 425	}
 426
 427	if (argc < 2) {
 428		pr_info("Probe point is not specified.\n");
 429		return -EINVAL;
 430	}
 431	if (isdigit(argv[1][0])) {
 432		pr_info("probe point must be have a filename.\n");
 
 433		return -EINVAL;
 434	}
 435	arg = strchr(argv[1], ':');
 436	if (!arg) {
 437		ret = -EINVAL;
 438		goto fail_address_parse;
 439	}
 440
 441	*arg++ = '\0';
 442	filename = argv[1];
 443	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 444	if (ret)
 445		goto fail_address_parse;
 446
 447	inode = igrab(path.dentry->d_inode);
 448	path_put(&path);
 449
 450	if (!inode || !S_ISREG(inode->i_mode)) {
 451		ret = -EINVAL;
 452		goto fail_address_parse;
 453	}
 454
 455	ret = kstrtoul(arg, 0, &offset);
 456	if (ret)
 457		goto fail_address_parse;
 458
 459	argc -= 2;
 460	argv += 2;
 461
 462	/* setup a probe */
 463	if (!event) {
 464		char *tail;
 465		char *ptr;
 466
 467		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 468		if (!tail) {
 469			ret = -ENOMEM;
 470			goto fail_address_parse;
 471		}
 472
 473		ptr = strpbrk(tail, ".-_");
 474		if (ptr)
 475			*ptr = '\0';
 476
 477		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 478		event = buf;
 479		kfree(tail);
 480	}
 481
 482	tu = alloc_trace_uprobe(group, event, argc, is_return);
 483	if (IS_ERR(tu)) {
 484		pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 485		ret = PTR_ERR(tu);
 486		goto fail_address_parse;
 487	}
 488	tu->offset = offset;
 489	tu->inode = inode;
 490	tu->filename = kstrdup(filename, GFP_KERNEL);
 491
 492	if (!tu->filename) {
 493		pr_info("Failed to allocate filename.\n");
 494		ret = -ENOMEM;
 495		goto error;
 496	}
 497
 498	/* parse arguments */
 499	ret = 0;
 500	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 501		struct probe_arg *parg = &tu->tp.args[i];
 502
 503		/* Increment count for freeing args in error case */
 504		tu->tp.nr_args++;
 505
 506		/* Parse argument name */
 507		arg = strchr(argv[i], '=');
 508		if (arg) {
 509			*arg++ = '\0';
 510			parg->name = kstrdup(argv[i], GFP_KERNEL);
 511		} else {
 512			arg = argv[i];
 513			/* If argument name is omitted, set "argN" */
 514			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 515			parg->name = kstrdup(buf, GFP_KERNEL);
 516		}
 517
 518		if (!parg->name) {
 519			pr_info("Failed to allocate argument[%d] name.\n", i);
 520			ret = -ENOMEM;
 521			goto error;
 522		}
 523
 524		if (!is_good_name(parg->name)) {
 525			pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 526			ret = -EINVAL;
 527			goto error;
 528		}
 529
 530		if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 531			pr_info("Argument[%d] name '%s' conflicts with "
 532				"another field.\n", i, argv[i]);
 533			ret = -EINVAL;
 534			goto error;
 535		}
 536
 537		/* Parse fetch argument */
 538		ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 539						 is_return, false);
 
 540		if (ret) {
 541			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 542			goto error;
 543		}
 544	}
 545
 546	ret = register_trace_uprobe(tu);
 547	if (ret)
 548		goto error;
 549	return 0;
 550
 551error:
 552	free_trace_uprobe(tu);
 553	return ret;
 554
 555fail_address_parse:
 556	if (inode)
 557		iput(inode);
 558
 559	pr_info("Failed to parse address or file.\n");
 560
 561	return ret;
 562}
 563
 564static int cleanup_all_probes(void)
 565{
 566	struct trace_uprobe *tu;
 567	int ret = 0;
 568
 569	mutex_lock(&uprobe_lock);
 570	while (!list_empty(&uprobe_list)) {
 571		tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 572		ret = unregister_trace_uprobe(tu);
 573		if (ret)
 574			break;
 575	}
 576	mutex_unlock(&uprobe_lock);
 577	return ret;
 578}
 579
 580/* Probes listing interfaces */
 581static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 582{
 583	mutex_lock(&uprobe_lock);
 584	return seq_list_start(&uprobe_list, *pos);
 585}
 586
 587static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 588{
 589	return seq_list_next(v, &uprobe_list, pos);
 590}
 591
 592static void probes_seq_stop(struct seq_file *m, void *v)
 593{
 594	mutex_unlock(&uprobe_lock);
 595}
 596
 597static int probes_seq_show(struct seq_file *m, void *v)
 598{
 599	struct trace_uprobe *tu = v;
 600	char c = is_ret_probe(tu) ? 'r' : 'p';
 601	int i;
 602
 603	seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 604			ftrace_event_name(&tu->tp.call));
 605	seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 606
 607	for (i = 0; i < tu->tp.nr_args; i++)
 608		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 609
 610	seq_printf(m, "\n");
 611	return 0;
 612}
 613
 614static const struct seq_operations probes_seq_op = {
 615	.start	= probes_seq_start,
 616	.next	= probes_seq_next,
 617	.stop	= probes_seq_stop,
 618	.show	= probes_seq_show
 619};
 620
 621static int probes_open(struct inode *inode, struct file *file)
 622{
 623	int ret;
 624
 625	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 626		ret = cleanup_all_probes();
 627		if (ret)
 628			return ret;
 629	}
 630
 631	return seq_open(file, &probes_seq_op);
 632}
 633
 634static ssize_t probes_write(struct file *file, const char __user *buffer,
 635			    size_t count, loff_t *ppos)
 636{
 637	return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 638}
 639
 640static const struct file_operations uprobe_events_ops = {
 641	.owner		= THIS_MODULE,
 642	.open		= probes_open,
 643	.read		= seq_read,
 644	.llseek		= seq_lseek,
 645	.release	= seq_release,
 646	.write		= probes_write,
 647};
 648
 649/* Probes profiling interfaces */
 650static int probes_profile_seq_show(struct seq_file *m, void *v)
 651{
 652	struct trace_uprobe *tu = v;
 653
 654	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 655			ftrace_event_name(&tu->tp.call), tu->nhit);
 656	return 0;
 657}
 658
 659static const struct seq_operations profile_seq_op = {
 660	.start	= probes_seq_start,
 661	.next	= probes_seq_next,
 662	.stop	= probes_seq_stop,
 663	.show	= probes_profile_seq_show
 664};
 665
 666static int profile_open(struct inode *inode, struct file *file)
 667{
 668	return seq_open(file, &profile_seq_op);
 669}
 670
 671static const struct file_operations uprobe_profile_ops = {
 672	.owner		= THIS_MODULE,
 673	.open		= profile_open,
 674	.read		= seq_read,
 675	.llseek		= seq_lseek,
 676	.release	= seq_release,
 677};
 678
 679struct uprobe_cpu_buffer {
 680	struct mutex mutex;
 681	void *buf;
 682};
 683static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 684static int uprobe_buffer_refcnt;
 685
 686static int uprobe_buffer_init(void)
 687{
 688	int cpu, err_cpu;
 689
 690	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 691	if (uprobe_cpu_buffer == NULL)
 692		return -ENOMEM;
 693
 694	for_each_possible_cpu(cpu) {
 695		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 696						  GFP_KERNEL, 0);
 697		if (p == NULL) {
 698			err_cpu = cpu;
 699			goto err;
 700		}
 701		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 702		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 703	}
 704
 705	return 0;
 706
 707err:
 708	for_each_possible_cpu(cpu) {
 709		if (cpu == err_cpu)
 710			break;
 711		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 712	}
 713
 714	free_percpu(uprobe_cpu_buffer);
 715	return -ENOMEM;
 716}
 717
 718static int uprobe_buffer_enable(void)
 719{
 720	int ret = 0;
 721
 722	BUG_ON(!mutex_is_locked(&event_mutex));
 723
 724	if (uprobe_buffer_refcnt++ == 0) {
 725		ret = uprobe_buffer_init();
 726		if (ret < 0)
 727			uprobe_buffer_refcnt--;
 728	}
 729
 730	return ret;
 731}
 732
 733static void uprobe_buffer_disable(void)
 734{
 735	int cpu;
 736
 737	BUG_ON(!mutex_is_locked(&event_mutex));
 738
 739	if (--uprobe_buffer_refcnt == 0) {
 740		for_each_possible_cpu(cpu)
 741			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 742							     cpu)->buf);
 743
 744		free_percpu(uprobe_cpu_buffer);
 745		uprobe_cpu_buffer = NULL;
 746	}
 747}
 748
 749static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 750{
 751	struct uprobe_cpu_buffer *ucb;
 752	int cpu;
 753
 754	cpu = raw_smp_processor_id();
 755	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 756
 757	/*
 758	 * Use per-cpu buffers for fastest access, but we might migrate
 759	 * so the mutex makes sure we have sole access to it.
 760	 */
 761	mutex_lock(&ucb->mutex);
 762
 763	return ucb;
 764}
 765
 766static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 767{
 768	mutex_unlock(&ucb->mutex);
 769}
 770
 771static void __uprobe_trace_func(struct trace_uprobe *tu,
 772				unsigned long func, struct pt_regs *regs,
 773				struct uprobe_cpu_buffer *ucb, int dsize,
 774				struct ftrace_event_file *ftrace_file)
 775{
 776	struct uprobe_trace_entry_head *entry;
 777	struct ring_buffer_event *event;
 778	struct ring_buffer *buffer;
 779	void *data;
 780	int size, esize;
 781	struct ftrace_event_call *call = &tu->tp.call;
 782
 783	WARN_ON(call != ftrace_file->event_call);
 784
 785	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 786		return;
 787
 788	if (ftrace_trigger_soft_disabled(ftrace_file))
 789		return;
 790
 791	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 792	size = esize + tu->tp.size + dsize;
 793	event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 794						call->event.type, size, 0, 0);
 795	if (!event)
 796		return;
 797
 798	entry = ring_buffer_event_data(event);
 799	if (is_ret_probe(tu)) {
 800		entry->vaddr[0] = func;
 801		entry->vaddr[1] = instruction_pointer(regs);
 802		data = DATAOF_TRACE_ENTRY(entry, true);
 803	} else {
 804		entry->vaddr[0] = instruction_pointer(regs);
 805		data = DATAOF_TRACE_ENTRY(entry, false);
 806	}
 807
 808	memcpy(data, ucb->buf, tu->tp.size + dsize);
 809
 810	event_trigger_unlock_commit(ftrace_file, buffer, event, entry, 0, 0);
 811}
 812
 813/* uprobe handler */
 814static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 815			     struct uprobe_cpu_buffer *ucb, int dsize)
 816{
 817	struct event_file_link *link;
 818
 819	if (is_ret_probe(tu))
 820		return 0;
 821
 822	rcu_read_lock();
 823	list_for_each_entry_rcu(link, &tu->tp.files, list)
 824		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 825	rcu_read_unlock();
 826
 827	return 0;
 828}
 829
 830static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 831				 struct pt_regs *regs,
 832				 struct uprobe_cpu_buffer *ucb, int dsize)
 833{
 834	struct event_file_link *link;
 835
 836	rcu_read_lock();
 837	list_for_each_entry_rcu(link, &tu->tp.files, list)
 838		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 839	rcu_read_unlock();
 840}
 841
 842/* Event entry printers */
 843static enum print_line_t
 844print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 845{
 846	struct uprobe_trace_entry_head *entry;
 847	struct trace_seq *s = &iter->seq;
 848	struct trace_uprobe *tu;
 849	u8 *data;
 850	int i;
 851
 852	entry = (struct uprobe_trace_entry_head *)iter->ent;
 853	tu = container_of(event, struct trace_uprobe, tp.call.event);
 854
 855	if (is_ret_probe(tu)) {
 856		if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 857					ftrace_event_name(&tu->tp.call),
 858					entry->vaddr[1], entry->vaddr[0]))
 859			goto partial;
 860		data = DATAOF_TRACE_ENTRY(entry, true);
 861	} else {
 862		if (!trace_seq_printf(s, "%s: (0x%lx)",
 863					ftrace_event_name(&tu->tp.call),
 864					entry->vaddr[0]))
 865			goto partial;
 866		data = DATAOF_TRACE_ENTRY(entry, false);
 867	}
 868
 869	for (i = 0; i < tu->tp.nr_args; i++) {
 870		struct probe_arg *parg = &tu->tp.args[i];
 871
 872		if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 873			goto partial;
 874	}
 875
 876	if (trace_seq_puts(s, "\n"))
 877		return TRACE_TYPE_HANDLED;
 878
 879partial:
 880	return TRACE_TYPE_PARTIAL_LINE;
 881}
 882
 883typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 884				enum uprobe_filter_ctx ctx,
 885				struct mm_struct *mm);
 886
 887static int
 888probe_event_enable(struct trace_uprobe *tu, struct ftrace_event_file *file,
 889		   filter_func_t filter)
 890{
 891	bool enabled = trace_probe_is_enabled(&tu->tp);
 892	struct event_file_link *link = NULL;
 893	int ret;
 894
 895	if (file) {
 
 
 
 896		link = kmalloc(sizeof(*link), GFP_KERNEL);
 897		if (!link)
 898			return -ENOMEM;
 899
 900		link->file = file;
 901		list_add_tail_rcu(&link->list, &tu->tp.files);
 902
 903		tu->tp.flags |= TP_FLAG_TRACE;
 904	} else
 905		tu->tp.flags |= TP_FLAG_PROFILE;
 
 906
 907	ret = uprobe_buffer_enable();
 908	if (ret < 0)
 909		return ret;
 910
 911	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 912
 913	if (enabled)
 914		return 0;
 915
 
 
 
 
 916	tu->consumer.filter = filter;
 
 917	ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 918	if (ret) {
 919		if (file) {
 920			list_del(&link->list);
 921			kfree(link);
 922			tu->tp.flags &= ~TP_FLAG_TRACE;
 923		} else
 924			tu->tp.flags &= ~TP_FLAG_PROFILE;
 925	}
 926
 
 
 
 
 
 
 
 
 
 
 
 
 
 927	return ret;
 928}
 929
 930static void
 931probe_event_disable(struct trace_uprobe *tu, struct ftrace_event_file *file)
 932{
 933	if (!trace_probe_is_enabled(&tu->tp))
 934		return;
 935
 936	if (file) {
 937		struct event_file_link *link;
 938
 939		link = find_event_file_link(&tu->tp, file);
 940		if (!link)
 941			return;
 942
 943		list_del_rcu(&link->list);
 944		/* synchronize with u{,ret}probe_trace_func */
 945		synchronize_sched();
 946		kfree(link);
 947
 948		if (!list_empty(&tu->tp.files))
 949			return;
 950	}
 951
 952	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 953
 954	uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 
 955	tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 956
 957	uprobe_buffer_disable();
 958}
 959
 960static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 961{
 962	int ret, i, size;
 963	struct uprobe_trace_entry_head field;
 964	struct trace_uprobe *tu = event_call->data;
 965
 966	if (is_ret_probe(tu)) {
 967		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 968		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 969		size = SIZEOF_TRACE_ENTRY(true);
 970	} else {
 971		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 972		size = SIZEOF_TRACE_ENTRY(false);
 973	}
 974	/* Set argument names as fields */
 975	for (i = 0; i < tu->tp.nr_args; i++) {
 976		struct probe_arg *parg = &tu->tp.args[i];
 977
 978		ret = trace_define_field(event_call, parg->type->fmttype,
 979					 parg->name, size + parg->offset,
 980					 parg->type->size, parg->type->is_signed,
 981					 FILTER_OTHER);
 982
 983		if (ret)
 984			return ret;
 985	}
 986	return 0;
 987}
 988
 989#ifdef CONFIG_PERF_EVENTS
 990static bool
 991__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 992{
 993	struct perf_event *event;
 994
 995	if (filter->nr_systemwide)
 996		return true;
 997
 998	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
 999		if (event->hw.tp_target->mm == mm)
1000			return true;
1001	}
1002
1003	return false;
1004}
1005
1006static inline bool
1007uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1008{
1009	return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
1010}
1011
1012static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1013{
1014	bool done;
1015
1016	write_lock(&tu->filter.rwlock);
1017	if (event->hw.tp_target) {
1018		/*
1019		 * event->parent != NULL means copy_process(), we can avoid
1020		 * uprobe_apply(). current->mm must be probed and we can rely
1021		 * on dup_mmap() which preserves the already installed bp's.
1022		 *
1023		 * attr.enable_on_exec means that exec/mmap will install the
1024		 * breakpoints we need.
1025		 */
1026		done = tu->filter.nr_systemwide ||
1027			event->parent || event->attr.enable_on_exec ||
1028			uprobe_filter_event(tu, event);
1029		list_add(&event->hw.tp_list, &tu->filter.perf_events);
1030	} else {
 
1031		done = tu->filter.nr_systemwide;
1032		tu->filter.nr_systemwide++;
1033	}
1034	write_unlock(&tu->filter.rwlock);
1035
1036	if (!done)
1037		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1038
1039	return 0;
1040}
1041
1042static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1043{
1044	bool done;
 
1045
1046	write_lock(&tu->filter.rwlock);
1047	if (event->hw.tp_target) {
1048		list_del(&event->hw.tp_list);
 
 
 
 
 
 
 
1049		done = tu->filter.nr_systemwide ||
1050			(event->hw.tp_target->flags & PF_EXITING) ||
1051			uprobe_filter_event(tu, event);
 
1052	} else {
1053		tu->filter.nr_systemwide--;
1054		done = tu->filter.nr_systemwide;
 
1055	}
1056	write_unlock(&tu->filter.rwlock);
1057
1058	if (!done)
1059		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1060
1061	return 0;
 
 
 
1062}
1063
1064static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1065				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1066{
1067	struct trace_uprobe *tu;
1068	int ret;
1069
1070	tu = container_of(uc, struct trace_uprobe, consumer);
1071	read_lock(&tu->filter.rwlock);
1072	ret = __uprobe_perf_filter(&tu->filter, mm);
1073	read_unlock(&tu->filter.rwlock);
1074
1075	return ret;
1076}
1077
1078static void __uprobe_perf_func(struct trace_uprobe *tu,
1079			       unsigned long func, struct pt_regs *regs,
1080			       struct uprobe_cpu_buffer *ucb, int dsize)
1081{
1082	struct ftrace_event_call *call = &tu->tp.call;
1083	struct uprobe_trace_entry_head *entry;
1084	struct hlist_head *head;
1085	void *data;
1086	int size, esize;
1087	int rctx;
1088
 
 
 
1089	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1090
1091	size = esize + tu->tp.size + dsize;
1092	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1093	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1094		return;
1095
1096	preempt_disable();
1097	head = this_cpu_ptr(call->perf_events);
1098	if (hlist_empty(head))
1099		goto out;
1100
1101	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1102	if (!entry)
1103		goto out;
1104
1105	if (is_ret_probe(tu)) {
1106		entry->vaddr[0] = func;
1107		entry->vaddr[1] = instruction_pointer(regs);
1108		data = DATAOF_TRACE_ENTRY(entry, true);
1109	} else {
1110		entry->vaddr[0] = instruction_pointer(regs);
1111		data = DATAOF_TRACE_ENTRY(entry, false);
1112	}
1113
1114	memcpy(data, ucb->buf, tu->tp.size + dsize);
1115
1116	if (size - esize > tu->tp.size + dsize) {
1117		int len = tu->tp.size + dsize;
1118
1119		memset(data + len, 0, size - esize - len);
1120	}
1121
1122	perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 
1123 out:
1124	preempt_enable();
1125}
1126
1127/* uprobe profile handler */
1128static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1129			    struct uprobe_cpu_buffer *ucb, int dsize)
1130{
1131	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1132		return UPROBE_HANDLER_REMOVE;
1133
1134	if (!is_ret_probe(tu))
1135		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1136	return 0;
1137}
1138
1139static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1140				struct pt_regs *regs,
1141				struct uprobe_cpu_buffer *ucb, int dsize)
1142{
1143	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1144}
1145#endif	/* CONFIG_PERF_EVENTS */
1146
1147static int
1148trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type,
1149		      void *data)
1150{
1151	struct trace_uprobe *tu = event->data;
1152	struct ftrace_event_file *file = data;
1153
1154	switch (type) {
1155	case TRACE_REG_REGISTER:
1156		return probe_event_enable(tu, file, NULL);
1157
1158	case TRACE_REG_UNREGISTER:
1159		probe_event_disable(tu, file);
1160		return 0;
1161
1162#ifdef CONFIG_PERF_EVENTS
1163	case TRACE_REG_PERF_REGISTER:
1164		return probe_event_enable(tu, NULL, uprobe_perf_filter);
1165
1166	case TRACE_REG_PERF_UNREGISTER:
1167		probe_event_disable(tu, NULL);
1168		return 0;
1169
1170	case TRACE_REG_PERF_OPEN:
1171		return uprobe_perf_open(tu, data);
1172
1173	case TRACE_REG_PERF_CLOSE:
1174		return uprobe_perf_close(tu, data);
1175
1176#endif
1177	default:
1178		return 0;
1179	}
1180	return 0;
1181}
1182
1183static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1184{
1185	struct trace_uprobe *tu;
1186	struct uprobe_dispatch_data udd;
1187	struct uprobe_cpu_buffer *ucb;
1188	int dsize, esize;
1189	int ret = 0;
1190
1191
1192	tu = container_of(con, struct trace_uprobe, consumer);
1193	tu->nhit++;
1194
1195	udd.tu = tu;
1196	udd.bp_addr = instruction_pointer(regs);
1197
1198	current->utask->vaddr = (unsigned long) &udd;
1199
1200#ifdef CONFIG_PERF_EVENTS
1201	if ((tu->tp.flags & TP_FLAG_TRACE) == 0 &&
1202	    !uprobe_perf_filter(&tu->consumer, 0, current->mm))
1203		return UPROBE_HANDLER_REMOVE;
1204#endif
1205
1206	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1207		return 0;
1208
1209	dsize = __get_data_size(&tu->tp, regs);
1210	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1211
1212	ucb = uprobe_buffer_get();
1213	store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1214
1215	if (tu->tp.flags & TP_FLAG_TRACE)
1216		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1217
1218#ifdef CONFIG_PERF_EVENTS
1219	if (tu->tp.flags & TP_FLAG_PROFILE)
1220		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1221#endif
1222	uprobe_buffer_put(ucb);
1223	return ret;
1224}
1225
1226static int uretprobe_dispatcher(struct uprobe_consumer *con,
1227				unsigned long func, struct pt_regs *regs)
1228{
1229	struct trace_uprobe *tu;
1230	struct uprobe_dispatch_data udd;
1231	struct uprobe_cpu_buffer *ucb;
1232	int dsize, esize;
1233
1234	tu = container_of(con, struct trace_uprobe, consumer);
1235
1236	udd.tu = tu;
1237	udd.bp_addr = func;
1238
1239	current->utask->vaddr = (unsigned long) &udd;
1240
1241	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1242		return 0;
1243
1244	dsize = __get_data_size(&tu->tp, regs);
1245	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1246
1247	ucb = uprobe_buffer_get();
1248	store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1249
1250	if (tu->tp.flags & TP_FLAG_TRACE)
1251		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1252
1253#ifdef CONFIG_PERF_EVENTS
1254	if (tu->tp.flags & TP_FLAG_PROFILE)
1255		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1256#endif
1257	uprobe_buffer_put(ucb);
1258	return 0;
1259}
1260
1261static struct trace_event_functions uprobe_funcs = {
1262	.trace		= print_uprobe_event
1263};
1264
1265static int register_uprobe_event(struct trace_uprobe *tu)
 
1266{
1267	struct ftrace_event_call *call = &tu->tp.call;
1268	int ret;
1269
1270	/* Initialize ftrace_event_call */
1271	INIT_LIST_HEAD(&call->class->fields);
1272	call->event.funcs = &uprobe_funcs;
1273	call->class->define_fields = uprobe_event_define_fields;
1274
 
 
 
 
 
 
 
 
 
 
 
 
1275	if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1276		return -ENOMEM;
1277
1278	ret = register_ftrace_event(&call->event);
1279	if (!ret) {
1280		kfree(call->print_fmt);
1281		return -ENODEV;
1282	}
1283	call->flags = 0;
1284	call->class->reg = trace_uprobe_register;
1285	call->data = tu;
1286	ret = trace_add_event_call(call);
1287
1288	if (ret) {
1289		pr_info("Failed to register uprobe event: %s\n",
1290			ftrace_event_name(call));
1291		kfree(call->print_fmt);
1292		unregister_ftrace_event(&call->event);
1293	}
1294
1295	return ret;
1296}
1297
1298static int unregister_uprobe_event(struct trace_uprobe *tu)
1299{
1300	int ret;
1301
1302	/* tu->event is unregistered in trace_remove_event_call() */
1303	ret = trace_remove_event_call(&tu->tp.call);
1304	if (ret)
1305		return ret;
1306	kfree(tu->tp.call.print_fmt);
1307	tu->tp.call.print_fmt = NULL;
1308	return 0;
1309}
1310
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1311/* Make a trace interface for controling probe points */
1312static __init int init_uprobe_trace(void)
1313{
1314	struct dentry *d_tracer;
1315
1316	d_tracer = tracing_init_dentry();
1317	if (!d_tracer)
1318		return 0;
1319
1320	trace_create_file("uprobe_events", 0644, d_tracer,
1321				    NULL, &uprobe_events_ops);
1322	/* Profile interface */
1323	trace_create_file("uprobe_profile", 0444, d_tracer,
1324				    NULL, &uprobe_profile_ops);
1325	return 0;
1326}
1327
1328fs_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);