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);
v6.8
   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
  91static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  92static int uretprobe_dispatcher(struct uprobe_consumer *con,
  93				unsigned long func, struct pt_regs *regs);
  94
  95#ifdef CONFIG_STACK_GROWSUP
  96static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  97{
  98	return addr - (n * sizeof(long));
  99}
 100#else
 101static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 102{
 103	return addr + (n * sizeof(long));
 104}
 105#endif
 106
 107static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 108{
 109	unsigned long ret;
 110	unsigned long addr = user_stack_pointer(regs);
 111
 112	addr = adjust_stack_addr(addr, n);
 113
 114	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 115		return 0;
 116
 117	return ret;
 118}
 119
 120/*
 121 * Uprobes-specific fetch functions
 122 */
 123static nokprobe_inline int
 124probe_mem_read(void *dest, void *src, size_t size)
 125{
 126	void __user *vaddr = (void __force __user *)src;
 127
 128	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 129}
 130
 131static nokprobe_inline int
 132probe_mem_read_user(void *dest, void *src, size_t size)
 133{
 134	return probe_mem_read(dest, src, size);
 
 
 
 
 
 
 
 
 
 
 
 135}
 136
 137/*
 138 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 139 * length and relative data location.
 140 */
 141static nokprobe_inline int
 142fetch_store_string(unsigned long addr, void *dest, void *base)
 143{
 144	long ret;
 145	u32 loc = *(u32 *)dest;
 146	int maxlen  = get_loc_len(loc);
 147	u8 *dst = get_loc_data(dest, base);
 148	void __user *src = (void __force __user *) addr;
 149
 150	if (unlikely(!maxlen))
 151		return -ENOMEM;
 152
 153	if (addr == FETCH_TOKEN_COMM)
 154		ret = strscpy(dst, current->comm, maxlen);
 155	else
 156		ret = strncpy_from_user(dst, src, maxlen);
 157	if (ret >= 0) {
 158		if (ret == maxlen)
 159			dst[ret - 1] = '\0';
 160		else
 161			/*
 162			 * Include the terminating null byte. In this case it
 163			 * was copied by strncpy_from_user but not accounted
 164			 * for in ret.
 165			 */
 166			ret++;
 167		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 168	} else
 169		*(u32 *)dest = make_data_loc(0, (void *)dst - base);
 170
 171	return ret;
 172}
 173
 174static nokprobe_inline int
 175fetch_store_string_user(unsigned long addr, void *dest, void *base)
 176{
 177	return fetch_store_string(addr, dest, base);
 178}
 179
 180/* Return the length of string -- including null terminal byte */
 181static nokprobe_inline int
 182fetch_store_strlen(unsigned long addr)
 183{
 184	int len;
 185	void __user *vaddr = (void __force __user *) addr;
 186
 187	if (addr == FETCH_TOKEN_COMM)
 188		len = strlen(current->comm) + 1;
 
 
 189	else
 190		len = strnlen_user(vaddr, MAX_STRING_SIZE);
 191
 192	return (len > MAX_STRING_SIZE) ? 0 : len;
 193}
 194
 195static nokprobe_inline int
 196fetch_store_strlen_user(unsigned long addr)
 197{
 198	return fetch_store_strlen(addr);
 199}
 200
 201static unsigned long translate_user_vaddr(unsigned long file_offset)
 202{
 203	unsigned long base_addr;
 204	struct uprobe_dispatch_data *udd;
 205
 206	udd = (void *) current->utask->vaddr;
 207
 208	base_addr = udd->bp_addr - udd->tu->offset;
 209	return base_addr + file_offset;
 210}
 211
 212/* Note that we don't verify it, since the code does not come from user space */
 213static int
 214process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 215		   void *base)
 216{
 217	struct pt_regs *regs = rec;
 218	unsigned long val;
 219	int ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 220
 221	/* 1st stage: get value from context */
 222	switch (code->op) {
 223	case FETCH_OP_REG:
 224		val = regs_get_register(regs, code->param);
 225		break;
 226	case FETCH_OP_STACK:
 227		val = get_user_stack_nth(regs, code->param);
 228		break;
 229	case FETCH_OP_STACKP:
 230		val = user_stack_pointer(regs);
 231		break;
 232	case FETCH_OP_RETVAL:
 233		val = regs_return_value(regs);
 234		break;
 235	case FETCH_OP_COMM:
 236		val = FETCH_TOKEN_COMM;
 237		break;
 238	case FETCH_OP_FOFFS:
 239		val = translate_user_vaddr(code->immediate);
 240		break;
 241	default:
 242		ret = process_common_fetch_insn(code, &val);
 243		if (ret < 0)
 244			return ret;
 245	}
 246	code++;
 247
 248	return process_fetch_insn_bottom(code, val, dest, base);
 249}
 250NOKPROBE_SYMBOL(process_fetch_insn)
 251
 252static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 253{
 254	rwlock_init(&filter->rwlock);
 255	filter->nr_systemwide = 0;
 256	INIT_LIST_HEAD(&filter->perf_events);
 257}
 258
 259static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 260{
 261	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 262}
 263
 264static inline bool is_ret_probe(struct trace_uprobe *tu)
 265{
 266	return tu->consumer.ret_handler != NULL;
 267}
 268
 269static bool trace_uprobe_is_busy(struct dyn_event *ev)
 270{
 271	struct trace_uprobe *tu = to_trace_uprobe(ev);
 272
 273	return trace_probe_is_enabled(&tu->tp);
 274}
 275
 276static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 277					    int argc, const char **argv)
 278{
 279	char buf[MAX_ARGSTR_LEN + 1];
 280	int len;
 281
 282	if (!argc)
 283		return true;
 284
 285	len = strlen(tu->filename);
 286	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 287		return false;
 288
 289	if (tu->ref_ctr_offset == 0)
 290		snprintf(buf, sizeof(buf), "0x%0*lx",
 291				(int)(sizeof(void *) * 2), tu->offset);
 292	else
 293		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 294				(int)(sizeof(void *) * 2), tu->offset,
 295				tu->ref_ctr_offset);
 296	if (strcmp(buf, &argv[0][len + 1]))
 297		return false;
 298
 299	argc--; argv++;
 300
 301	return trace_probe_match_command_args(&tu->tp, argc, argv);
 302}
 303
 304static bool trace_uprobe_match(const char *system, const char *event,
 305			int argc, const char **argv, struct dyn_event *ev)
 306{
 307	struct trace_uprobe *tu = to_trace_uprobe(ev);
 308
 309	return (event[0] == '\0' ||
 310		strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 311	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 312	   trace_uprobe_match_command_head(tu, argc, argv);
 313}
 314
 315static nokprobe_inline struct trace_uprobe *
 316trace_uprobe_primary_from_call(struct trace_event_call *call)
 317{
 318	struct trace_probe *tp;
 319
 320	tp = trace_probe_primary_from_call(call);
 321	if (WARN_ON_ONCE(!tp))
 322		return NULL;
 323
 324	return container_of(tp, struct trace_uprobe, tp);
 325}
 326
 327/*
 328 * Allocate new trace_uprobe and initialize it (including uprobes).
 329 */
 330static struct trace_uprobe *
 331alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 332{
 333	struct trace_uprobe *tu;
 334	int ret;
 335
 336	tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 
 
 
 
 
 
 337	if (!tu)
 338		return ERR_PTR(-ENOMEM);
 339
 340	ret = trace_probe_init(&tu->tp, event, group, true);
 341	if (ret < 0)
 
 
 
 
 
 342		goto error;
 343
 344	dyn_event_init(&tu->devent, &trace_uprobe_ops);
 
 345	tu->consumer.handler = uprobe_dispatcher;
 346	if (is_ret)
 347		tu->consumer.ret_handler = uretprobe_dispatcher;
 348	init_trace_uprobe_filter(tu->tp.event->filter);
 
 349	return tu;
 350
 351error:
 
 352	kfree(tu);
 353
 354	return ERR_PTR(ret);
 355}
 356
 357static void free_trace_uprobe(struct trace_uprobe *tu)
 358{
 359	if (!tu)
 360		return;
 
 
 361
 362	path_put(&tu->path);
 363	trace_probe_cleanup(&tu->tp);
 
 364	kfree(tu->filename);
 365	kfree(tu);
 366}
 367
 368static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 369{
 370	struct dyn_event *pos;
 371	struct trace_uprobe *tu;
 372
 373	for_each_trace_uprobe(tu, pos)
 374		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 375		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 376			return tu;
 377
 378	return NULL;
 379}
 380
 381/* Unregister a trace_uprobe and probe_event */
 382static int unregister_trace_uprobe(struct trace_uprobe *tu)
 383{
 384	int ret;
 385
 386	if (trace_probe_has_sibling(&tu->tp))
 387		goto unreg;
 388
 389	/* If there's a reference to the dynamic event */
 390	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 391		return -EBUSY;
 392
 393	ret = unregister_uprobe_event(tu);
 394	if (ret)
 395		return ret;
 396
 397unreg:
 398	dyn_event_remove(&tu->devent);
 399	trace_probe_unlink(&tu->tp);
 400	free_trace_uprobe(tu);
 401	return 0;
 402}
 403
 404static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 405					 struct trace_uprobe *comp)
 406{
 407	struct trace_probe_event *tpe = orig->tp.event;
 408	struct inode *comp_inode = d_real_inode(comp->path.dentry);
 409	int i;
 410
 411	list_for_each_entry(orig, &tpe->probes, tp.list) {
 412		if (comp_inode != d_real_inode(orig->path.dentry) ||
 413		    comp->offset != orig->offset)
 414			continue;
 415
 416		/*
 417		 * trace_probe_compare_arg_type() ensured that nr_args and
 418		 * each argument name and type are same. Let's compare comm.
 419		 */
 420		for (i = 0; i < orig->tp.nr_args; i++) {
 421			if (strcmp(orig->tp.args[i].comm,
 422				   comp->tp.args[i].comm))
 423				break;
 424		}
 425
 426		if (i == orig->tp.nr_args)
 427			return true;
 428	}
 429
 430	return false;
 431}
 432
 433static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 434{
 435	int ret;
 436
 437	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 438	if (ret) {
 439		/* Note that argument starts index = 2 */
 440		trace_probe_log_set_index(ret + 1);
 441		trace_probe_log_err(0, DIFF_ARG_TYPE);
 442		return -EEXIST;
 443	}
 444	if (trace_uprobe_has_same_uprobe(to, tu)) {
 445		trace_probe_log_set_index(0);
 446		trace_probe_log_err(0, SAME_PROBE);
 447		return -EEXIST;
 448	}
 449
 450	/* Append to existing event */
 451	ret = trace_probe_append(&tu->tp, &to->tp);
 452	if (!ret)
 453		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 454
 455	return ret;
 456}
 457
 458/*
 459 * Uprobe with multiple reference counter is not allowed. i.e.
 460 * If inode and offset matches, reference counter offset *must*
 461 * match as well. Though, there is one exception: If user is
 462 * replacing old trace_uprobe with new one(same group/event),
 463 * then we allow same uprobe with new reference counter as far
 464 * as the new one does not conflict with any other existing
 465 * ones.
 466 */
 467static int validate_ref_ctr_offset(struct trace_uprobe *new)
 468{
 469	struct dyn_event *pos;
 470	struct trace_uprobe *tmp;
 471	struct inode *new_inode = d_real_inode(new->path.dentry);
 472
 473	for_each_trace_uprobe(tmp, pos) {
 474		if (new_inode == d_real_inode(tmp->path.dentry) &&
 475		    new->offset == tmp->offset &&
 476		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
 477			pr_warn("Reference counter offset mismatch.");
 478			return -EINVAL;
 479		}
 480	}
 481	return 0;
 482}
 483
 484/* Register a trace_uprobe and probe_event */
 485static int register_trace_uprobe(struct trace_uprobe *tu)
 486{
 487	struct trace_uprobe *old_tu;
 488	int ret;
 489
 490	mutex_lock(&event_mutex);
 491
 492	ret = validate_ref_ctr_offset(tu);
 493	if (ret)
 494		goto end;
 495
 496	/* register as an event */
 497	old_tu = find_probe_event(trace_probe_name(&tu->tp),
 498				  trace_probe_group_name(&tu->tp));
 499	if (old_tu) {
 500		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 501			trace_probe_log_set_index(0);
 502			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 503			ret = -EEXIST;
 504		} else {
 505			ret = append_trace_uprobe(tu, old_tu);
 506		}
 507		goto end;
 508	}
 509
 510	ret = register_uprobe_event(tu);
 511	if (ret) {
 512		if (ret == -EEXIST) {
 513			trace_probe_log_set_index(0);
 514			trace_probe_log_err(0, EVENT_EXIST);
 515		} else
 516			pr_warn("Failed to register probe event(%d)\n", ret);
 517		goto end;
 518	}
 519
 520	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 521
 522end:
 523	mutex_unlock(&event_mutex);
 524
 525	return ret;
 526}
 527
 528/*
 529 * Argument syntax:
 530 *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 
 
 531 */
 532static int __trace_uprobe_create(int argc, const char **argv)
 533{
 534	struct trace_uprobe *tu;
 535	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 536	char *arg, *filename, *rctr, *rctr_end, *tmp;
 537	char buf[MAX_EVENT_NAME_LEN];
 538	char gbuf[MAX_EVENT_NAME_LEN];
 539	enum probe_print_type ptype;
 540	struct path path;
 541	unsigned long offset, ref_ctr_offset;
 542	bool is_return = false;
 543	int i, ret;
 544
 545	ref_ctr_offset = 0;
 546
 547	switch (argv[0][0]) {
 548	case 'r':
 
 
 
 
 
 
 
 549		is_return = true;
 550		break;
 551	case 'p':
 552		break;
 553	default:
 554		return -ECANCELED;
 555	}
 556
 557	if (argc < 2)
 558		return -ECANCELED;
 
 559
 560	if (argv[0][1] == ':')
 561		event = &argv[0][2];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 562
 563	if (!strchr(argv[1], '/'))
 564		return -ECANCELED;
 565
 566	filename = kstrdup(argv[1], GFP_KERNEL);
 567	if (!filename)
 568		return -ENOMEM;
 
 
 
 569
 570	/* Find the last occurrence, in case the path contains ':' too. */
 571	arg = strrchr(filename, ':');
 572	if (!arg || !isdigit(arg[1])) {
 573		kfree(filename);
 574		return -ECANCELED;
 
 
 
 
 575	}
 576
 577	trace_probe_log_init("trace_uprobe", argc, argv);
 578	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
 579
 580	*arg++ = '\0';
 581	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 582	if (ret) {
 583		trace_probe_log_err(0, FILE_NOT_FOUND);
 584		kfree(filename);
 585		trace_probe_log_clear();
 586		return ret;
 587	}
 588	if (!d_is_reg(path.dentry)) {
 589		trace_probe_log_err(0, NO_REGULAR_FILE);
 590		ret = -EINVAL;
 591		goto fail_address_parse;
 592	}
 593
 594	/* Parse reference counter offset if specified. */
 595	rctr = strchr(arg, '(');
 596	if (rctr) {
 597		rctr_end = strchr(rctr, ')');
 598		if (!rctr_end) {
 599			ret = -EINVAL;
 600			rctr_end = rctr + strlen(rctr);
 601			trace_probe_log_err(rctr_end - filename,
 602					    REFCNT_OPEN_BRACE);
 603			goto fail_address_parse;
 604		} else if (rctr_end[1] != '\0') {
 605			ret = -EINVAL;
 606			trace_probe_log_err(rctr_end + 1 - filename,
 607					    BAD_REFCNT_SUFFIX);
 608			goto fail_address_parse;
 609		}
 610
 611		*rctr++ = '\0';
 612		*rctr_end = '\0';
 613		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 614		if (ret) {
 615			trace_probe_log_err(rctr - filename, BAD_REFCNT);
 616			goto fail_address_parse;
 617		}
 618	}
 619
 620	/* Check if there is %return suffix */
 621	tmp = strchr(arg, '%');
 622	if (tmp) {
 623		if (!strcmp(tmp, "%return")) {
 624			*tmp = '\0';
 625			is_return = true;
 626		} else {
 627			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 628			ret = -EINVAL;
 629			goto fail_address_parse;
 630		}
 631	}
 632
 633	/* Parse uprobe offset. */
 634	ret = kstrtoul(arg, 0, &offset);
 635	if (ret) {
 636		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 637		goto fail_address_parse;
 638	}
 
 
 639
 640	/* setup a probe */
 641	trace_probe_log_set_index(0);
 642	if (event) {
 643		ret = traceprobe_parse_event_name(&event, &group, gbuf,
 644						  event - argv[0]);
 645		if (ret)
 646			goto fail_address_parse;
 647	}
 648
 649	if (!event) {
 650		char *tail;
 651		char *ptr;
 652
 653		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 654		if (!tail) {
 655			ret = -ENOMEM;
 656			goto fail_address_parse;
 657		}
 658
 659		ptr = strpbrk(tail, ".-_");
 660		if (ptr)
 661			*ptr = '\0';
 662
 663		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 664		event = buf;
 665		kfree(tail);
 666	}
 667
 668	argc -= 2;
 669	argv += 2;
 670
 671	tu = alloc_trace_uprobe(group, event, argc, is_return);
 672	if (IS_ERR(tu)) {
 
 673		ret = PTR_ERR(tu);
 674		/* This must return -ENOMEM otherwise there is a bug */
 675		WARN_ON_ONCE(ret != -ENOMEM);
 676		goto fail_address_parse;
 677	}
 678	tu->offset = offset;
 679	tu->ref_ctr_offset = ref_ctr_offset;
 680	tu->path = path;
 681	tu->filename = filename;
 
 
 
 
 
 682
 683	/* parse arguments */
 
 684	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 685		struct traceprobe_parse_context ctx = {
 686			.flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
 687		};
 688
 689		trace_probe_log_set_index(i + 2);
 690		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
 691		traceprobe_finish_parse(&ctx);
 692		if (ret)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 693			goto error;
 
 694	}
 695
 696	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 697	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 698	if (ret < 0)
 699		goto error;
 700
 701	ret = register_trace_uprobe(tu);
 702	if (!ret)
 703		goto out;
 704
 705error:
 706	free_trace_uprobe(tu);
 707out:
 708	trace_probe_log_clear();
 709	return ret;
 710
 711fail_address_parse:
 712	trace_probe_log_clear();
 713	path_put(&path);
 714	kfree(filename);
 
 715
 716	return ret;
 717}
 718
 719int trace_uprobe_create(const char *raw_command)
 720{
 721	return trace_probe_create(raw_command, __trace_uprobe_create);
 
 
 
 
 
 
 
 
 
 
 
 722}
 723
 724static int create_or_delete_trace_uprobe(const char *raw_command)
 
 725{
 726	int ret;
 
 
 727
 728	if (raw_command[0] == '-')
 729		return dyn_event_release(raw_command, &trace_uprobe_ops);
 730
 731	ret = trace_uprobe_create(raw_command);
 732	return ret == -ECANCELED ? -EINVAL : ret;
 733}
 734
 735static int trace_uprobe_release(struct dyn_event *ev)
 736{
 737	struct trace_uprobe *tu = to_trace_uprobe(ev);
 738
 739	return unregister_trace_uprobe(tu);
 740}
 741
 742/* Probes listing interfaces */
 743static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 744{
 745	struct trace_uprobe *tu = to_trace_uprobe(ev);
 746	char c = is_ret_probe(tu) ? 'r' : 'p';
 747	int i;
 748
 749	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 750			trace_probe_name(&tu->tp), tu->filename,
 751			(int)(sizeof(void *) * 2), tu->offset);
 752
 753	if (tu->ref_ctr_offset)
 754		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 755
 756	for (i = 0; i < tu->tp.nr_args; i++)
 757		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 758
 759	seq_putc(m, '\n');
 760	return 0;
 761}
 762
 763static int probes_seq_show(struct seq_file *m, void *v)
 764{
 765	struct dyn_event *ev = v;
 766
 767	if (!is_trace_uprobe(ev))
 768		return 0;
 769
 770	return trace_uprobe_show(m, ev);
 771}
 772
 773static const struct seq_operations probes_seq_op = {
 774	.start  = dyn_event_seq_start,
 775	.next   = dyn_event_seq_next,
 776	.stop   = dyn_event_seq_stop,
 777	.show   = probes_seq_show
 778};
 779
 780static int probes_open(struct inode *inode, struct file *file)
 781{
 782	int ret;
 783
 784	ret = security_locked_down(LOCKDOWN_TRACEFS);
 785	if (ret)
 786		return ret;
 787
 788	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 789		ret = dyn_events_release_all(&trace_uprobe_ops);
 790		if (ret)
 791			return ret;
 792	}
 793
 794	return seq_open(file, &probes_seq_op);
 795}
 796
 797static ssize_t probes_write(struct file *file, const char __user *buffer,
 798			    size_t count, loff_t *ppos)
 799{
 800	return trace_parse_run_command(file, buffer, count, ppos,
 801					create_or_delete_trace_uprobe);
 802}
 803
 804static const struct file_operations uprobe_events_ops = {
 805	.owner		= THIS_MODULE,
 806	.open		= probes_open,
 807	.read		= seq_read,
 808	.llseek		= seq_lseek,
 809	.release	= seq_release,
 810	.write		= probes_write,
 811};
 812
 813/* Probes profiling interfaces */
 814static int probes_profile_seq_show(struct seq_file *m, void *v)
 815{
 816	struct dyn_event *ev = v;
 817	struct trace_uprobe *tu;
 818
 819	if (!is_trace_uprobe(ev))
 820		return 0;
 821
 822	tu = to_trace_uprobe(ev);
 823	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 824			trace_probe_name(&tu->tp), tu->nhit);
 825	return 0;
 826}
 827
 828static const struct seq_operations profile_seq_op = {
 829	.start  = dyn_event_seq_start,
 830	.next   = dyn_event_seq_next,
 831	.stop   = dyn_event_seq_stop,
 832	.show	= probes_profile_seq_show
 833};
 834
 835static int profile_open(struct inode *inode, struct file *file)
 836{
 837	int ret;
 838
 839	ret = security_locked_down(LOCKDOWN_TRACEFS);
 840	if (ret)
 841		return ret;
 842
 843	return seq_open(file, &profile_seq_op);
 844}
 845
 846static const struct file_operations uprobe_profile_ops = {
 847	.owner		= THIS_MODULE,
 848	.open		= profile_open,
 849	.read		= seq_read,
 850	.llseek		= seq_lseek,
 851	.release	= seq_release,
 852};
 853
 854struct uprobe_cpu_buffer {
 855	struct mutex mutex;
 856	void *buf;
 857};
 858static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 859static int uprobe_buffer_refcnt;
 860
 861static int uprobe_buffer_init(void)
 862{
 863	int cpu, err_cpu;
 864
 865	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 866	if (uprobe_cpu_buffer == NULL)
 867		return -ENOMEM;
 868
 869	for_each_possible_cpu(cpu) {
 870		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 871						  GFP_KERNEL, 0);
 872		if (p == NULL) {
 873			err_cpu = cpu;
 874			goto err;
 875		}
 876		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 877		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 878	}
 879
 880	return 0;
 881
 882err:
 883	for_each_possible_cpu(cpu) {
 884		if (cpu == err_cpu)
 885			break;
 886		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 887	}
 888
 889	free_percpu(uprobe_cpu_buffer);
 890	return -ENOMEM;
 891}
 892
 893static int uprobe_buffer_enable(void)
 894{
 895	int ret = 0;
 896
 897	BUG_ON(!mutex_is_locked(&event_mutex));
 898
 899	if (uprobe_buffer_refcnt++ == 0) {
 900		ret = uprobe_buffer_init();
 901		if (ret < 0)
 902			uprobe_buffer_refcnt--;
 903	}
 904
 905	return ret;
 906}
 907
 908static void uprobe_buffer_disable(void)
 909{
 910	int cpu;
 911
 912	BUG_ON(!mutex_is_locked(&event_mutex));
 913
 914	if (--uprobe_buffer_refcnt == 0) {
 915		for_each_possible_cpu(cpu)
 916			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 917							     cpu)->buf);
 918
 919		free_percpu(uprobe_cpu_buffer);
 920		uprobe_cpu_buffer = NULL;
 921	}
 922}
 923
 924static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 925{
 926	struct uprobe_cpu_buffer *ucb;
 927	int cpu;
 928
 929	cpu = raw_smp_processor_id();
 930	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 931
 932	/*
 933	 * Use per-cpu buffers for fastest access, but we might migrate
 934	 * so the mutex makes sure we have sole access to it.
 935	 */
 936	mutex_lock(&ucb->mutex);
 937
 938	return ucb;
 939}
 940
 941static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 942{
 943	mutex_unlock(&ucb->mutex);
 944}
 945
 946static void __uprobe_trace_func(struct trace_uprobe *tu,
 947				unsigned long func, struct pt_regs *regs,
 948				struct uprobe_cpu_buffer *ucb, int dsize,
 949				struct trace_event_file *trace_file)
 950{
 951	struct uprobe_trace_entry_head *entry;
 952	struct trace_event_buffer fbuffer;
 
 953	void *data;
 954	int size, esize;
 955	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 956
 957	WARN_ON(call != trace_file->event_call);
 958
 959	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 960		return;
 961
 962	if (trace_trigger_soft_disabled(trace_file))
 963		return;
 964
 965	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 966	size = esize + tu->tp.size + dsize;
 967	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 968	if (!entry)
 
 969		return;
 970
 
 971	if (is_ret_probe(tu)) {
 972		entry->vaddr[0] = func;
 973		entry->vaddr[1] = instruction_pointer(regs);
 974		data = DATAOF_TRACE_ENTRY(entry, true);
 975	} else {
 976		entry->vaddr[0] = instruction_pointer(regs);
 977		data = DATAOF_TRACE_ENTRY(entry, false);
 978	}
 979
 980	memcpy(data, ucb->buf, tu->tp.size + dsize);
 981
 982	trace_event_buffer_commit(&fbuffer);
 983}
 984
 985/* uprobe handler */
 986static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 987			     struct uprobe_cpu_buffer *ucb, int dsize)
 988{
 989	struct event_file_link *link;
 990
 991	if (is_ret_probe(tu))
 992		return 0;
 993
 994	rcu_read_lock();
 995	trace_probe_for_each_link_rcu(link, &tu->tp)
 996		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 997	rcu_read_unlock();
 998
 999	return 0;
1000}
1001
1002static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1003				 struct pt_regs *regs,
1004				 struct uprobe_cpu_buffer *ucb, int dsize)
1005{
1006	struct event_file_link *link;
1007
1008	rcu_read_lock();
1009	trace_probe_for_each_link_rcu(link, &tu->tp)
1010		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1011	rcu_read_unlock();
1012}
1013
1014/* Event entry printers */
1015static enum print_line_t
1016print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1017{
1018	struct uprobe_trace_entry_head *entry;
1019	struct trace_seq *s = &iter->seq;
1020	struct trace_uprobe *tu;
1021	u8 *data;
 
1022
1023	entry = (struct uprobe_trace_entry_head *)iter->ent;
1024	tu = trace_uprobe_primary_from_call(
1025		container_of(event, struct trace_event_call, event));
1026	if (unlikely(!tu))
1027		goto out;
1028
1029	if (is_ret_probe(tu)) {
1030		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1031				 trace_probe_name(&tu->tp),
1032				 entry->vaddr[1], entry->vaddr[0]);
 
1033		data = DATAOF_TRACE_ENTRY(entry, true);
1034	} else {
1035		trace_seq_printf(s, "%s: (0x%lx)",
1036				 trace_probe_name(&tu->tp),
1037				 entry->vaddr[0]);
 
1038		data = DATAOF_TRACE_ENTRY(entry, false);
1039	}
1040
1041	if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1042		goto out;
 
 
 
 
1043
1044	trace_seq_putc(s, '\n');
 
1045
1046 out:
1047	return trace_handle_return(s);
1048}
1049
1050typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1051				enum uprobe_filter_ctx ctx,
1052				struct mm_struct *mm);
1053
1054static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
 
 
1055{
 
 
1056	int ret;
1057
1058	tu->consumer.filter = filter;
1059	tu->inode = d_real_inode(tu->path.dentry);
 
 
1060
1061	if (tu->ref_ctr_offset)
1062		ret = uprobe_register_refctr(tu->inode, tu->offset,
1063				tu->ref_ctr_offset, &tu->consumer);
1064	else
1065		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1066
1067	if (ret)
1068		tu->inode = NULL;
 
1069
1070	return ret;
1071}
1072
1073static void __probe_event_disable(struct trace_probe *tp)
1074{
1075	struct trace_uprobe *tu;
1076
1077	tu = container_of(tp, struct trace_uprobe, tp);
1078	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1079
1080	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1081		if (!tu->inode)
1082			continue;
1083
1084		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1085		tu->inode = NULL;
1086	}
1087}
1088
1089static int probe_event_enable(struct trace_event_call *call,
1090			struct trace_event_file *file, filter_func_t filter)
1091{
1092	struct trace_probe *tp;
1093	struct trace_uprobe *tu;
1094	bool enabled;
1095	int ret;
1096
1097	tp = trace_probe_primary_from_call(call);
1098	if (WARN_ON_ONCE(!tp))
1099		return -ENODEV;
1100	enabled = trace_probe_is_enabled(tp);
1101
1102	/* This may also change "enabled" state */
1103	if (file) {
1104		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1105			return -EINTR;
1106
1107		ret = trace_probe_add_file(tp, file);
1108		if (ret < 0)
1109			return ret;
1110	} else {
1111		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1112			return -EINTR;
1113
1114		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1115	}
1116
1117	tu = container_of(tp, struct trace_uprobe, tp);
1118	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1119
1120	if (enabled)
1121		return 0;
1122
1123	ret = uprobe_buffer_enable();
1124	if (ret)
1125		goto err_flags;
1126
1127	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1128		ret = trace_uprobe_enable(tu, filter);
1129		if (ret) {
1130			__probe_event_disable(tp);
1131			goto err_buffer;
1132		}
1133	}
1134
1135	return 0;
1136
1137 err_buffer:
1138	uprobe_buffer_disable();
1139
1140 err_flags:
1141	if (file)
1142		trace_probe_remove_file(tp, file);
1143	else
1144		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1145
1146	return ret;
1147}
1148
1149static void probe_event_disable(struct trace_event_call *call,
1150				struct trace_event_file *file)
1151{
1152	struct trace_probe *tp;
1153
1154	tp = trace_probe_primary_from_call(call);
1155	if (WARN_ON_ONCE(!tp))
1156		return;
1157
1158	if (!trace_probe_is_enabled(tp))
1159		return;
1160
1161	if (file) {
1162		if (trace_probe_remove_file(tp, file) < 0)
1163			return;
1164
1165		if (trace_probe_is_enabled(tp))
 
 
 
 
 
1166			return;
1167	} else
1168		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 
 
 
 
1169
1170	__probe_event_disable(tp);
1171	uprobe_buffer_disable();
1172}
1173
1174static int uprobe_event_define_fields(struct trace_event_call *event_call)
1175{
1176	int ret, size;
1177	struct uprobe_trace_entry_head field;
1178	struct trace_uprobe *tu;
1179
1180	tu = trace_uprobe_primary_from_call(event_call);
1181	if (unlikely(!tu))
1182		return -ENODEV;
1183
1184	if (is_ret_probe(tu)) {
1185		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1186		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1187		size = SIZEOF_TRACE_ENTRY(true);
1188	} else {
1189		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1190		size = SIZEOF_TRACE_ENTRY(false);
1191	}
 
 
 
 
 
 
 
 
1192
1193	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
 
 
 
1194}
1195
1196#ifdef CONFIG_PERF_EVENTS
1197static bool
1198__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1199{
1200	struct perf_event *event;
1201
1202	if (filter->nr_systemwide)
1203		return true;
1204
1205	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1206		if (event->hw.target->mm == mm)
1207			return true;
1208	}
1209
1210	return false;
1211}
1212
1213static inline bool
1214trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1215			  struct perf_event *event)
1216{
1217	return __uprobe_perf_filter(filter, event->hw.target->mm);
1218}
1219
1220static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1221				       struct perf_event *event)
1222{
1223	bool done;
1224
1225	write_lock(&filter->rwlock);
1226	if (event->hw.target) {
1227		list_del(&event->hw.tp_list);
1228		done = filter->nr_systemwide ||
1229			(event->hw.target->flags & PF_EXITING) ||
1230			trace_uprobe_filter_event(filter, event);
1231	} else {
1232		filter->nr_systemwide--;
1233		done = filter->nr_systemwide;
1234	}
1235	write_unlock(&filter->rwlock);
1236
1237	return done;
1238}
1239
1240/* This returns true if the filter always covers target mm */
1241static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1242				    struct perf_event *event)
1243{
1244	bool done;
1245
1246	write_lock(&filter->rwlock);
1247	if (event->hw.target) {
1248		/*
1249		 * event->parent != NULL means copy_process(), we can avoid
1250		 * uprobe_apply(). current->mm must be probed and we can rely
1251		 * on dup_mmap() which preserves the already installed bp's.
1252		 *
1253		 * attr.enable_on_exec means that exec/mmap will install the
1254		 * breakpoints we need.
1255		 */
1256		done = filter->nr_systemwide ||
1257			event->parent || event->attr.enable_on_exec ||
1258			trace_uprobe_filter_event(filter, event);
1259		list_add(&event->hw.tp_list, &filter->perf_events);
1260	} else {
1261		done = filter->nr_systemwide;
1262		filter->nr_systemwide++;
1263	}
1264	write_unlock(&filter->rwlock);
 
 
 
1265
1266	return done;
1267}
1268
1269static int uprobe_perf_close(struct trace_event_call *call,
1270			     struct perf_event *event)
1271{
1272	struct trace_probe *tp;
1273	struct trace_uprobe *tu;
1274	int ret = 0;
1275
1276	tp = trace_probe_primary_from_call(call);
1277	if (WARN_ON_ONCE(!tp))
1278		return -ENODEV;
1279
1280	tu = container_of(tp, struct trace_uprobe, tp);
1281	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1282		return 0;
1283
1284	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1285		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1286		if (ret)
1287			break;
1288	}
 
1289
1290	return ret;
1291}
1292
1293static int uprobe_perf_open(struct trace_event_call *call,
1294			    struct perf_event *event)
1295{
1296	struct trace_probe *tp;
1297	struct trace_uprobe *tu;
1298	int err = 0;
1299
1300	tp = trace_probe_primary_from_call(call);
1301	if (WARN_ON_ONCE(!tp))
1302		return -ENODEV;
1303
1304	tu = container_of(tp, struct trace_uprobe, tp);
1305	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1306		return 0;
1307
1308	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1309		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1310		if (err) {
1311			uprobe_perf_close(call, event);
1312			break;
1313		}
1314	}
1315
1316	return err;
1317}
1318
1319static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1320				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1321{
1322	struct trace_uprobe_filter *filter;
1323	struct trace_uprobe *tu;
1324	int ret;
1325
1326	tu = container_of(uc, struct trace_uprobe, consumer);
1327	filter = tu->tp.event->filter;
1328
1329	read_lock(&filter->rwlock);
1330	ret = __uprobe_perf_filter(filter, mm);
1331	read_unlock(&filter->rwlock);
1332
1333	return ret;
1334}
1335
1336static void __uprobe_perf_func(struct trace_uprobe *tu,
1337			       unsigned long func, struct pt_regs *regs,
1338			       struct uprobe_cpu_buffer *ucb, int dsize)
1339{
1340	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1341	struct uprobe_trace_entry_head *entry;
1342	struct hlist_head *head;
1343	void *data;
1344	int size, esize;
1345	int rctx;
1346
1347#ifdef CONFIG_BPF_EVENTS
1348	if (bpf_prog_array_valid(call)) {
1349		u32 ret;
1350
1351		ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run);
1352		if (!ret)
1353			return;
1354	}
1355#endif /* CONFIG_BPF_EVENTS */
1356
1357	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1358
1359	size = esize + tu->tp.size + dsize;
1360	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1361	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1362		return;
1363
1364	preempt_disable();
1365	head = this_cpu_ptr(call->perf_events);
1366	if (hlist_empty(head))
1367		goto out;
1368
1369	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1370	if (!entry)
1371		goto out;
1372
1373	if (is_ret_probe(tu)) {
1374		entry->vaddr[0] = func;
1375		entry->vaddr[1] = instruction_pointer(regs);
1376		data = DATAOF_TRACE_ENTRY(entry, true);
1377	} else {
1378		entry->vaddr[0] = instruction_pointer(regs);
1379		data = DATAOF_TRACE_ENTRY(entry, false);
1380	}
1381
1382	memcpy(data, ucb->buf, tu->tp.size + dsize);
1383
1384	if (size - esize > tu->tp.size + dsize) {
1385		int len = tu->tp.size + dsize;
1386
1387		memset(data + len, 0, size - esize - len);
1388	}
1389
1390	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1391			      head, NULL);
1392 out:
1393	preempt_enable();
1394}
1395
1396/* uprobe profile handler */
1397static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1398			    struct uprobe_cpu_buffer *ucb, int dsize)
1399{
1400	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1401		return UPROBE_HANDLER_REMOVE;
1402
1403	if (!is_ret_probe(tu))
1404		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1405	return 0;
1406}
1407
1408static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1409				struct pt_regs *regs,
1410				struct uprobe_cpu_buffer *ucb, int dsize)
1411{
1412	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1413}
1414
1415int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1416			const char **filename, u64 *probe_offset,
1417			u64 *probe_addr, bool perf_type_tracepoint)
1418{
1419	const char *pevent = trace_event_name(event->tp_event);
1420	const char *group = event->tp_event->class->system;
1421	struct trace_uprobe *tu;
1422
1423	if (perf_type_tracepoint)
1424		tu = find_probe_event(pevent, group);
1425	else
1426		tu = trace_uprobe_primary_from_call(event->tp_event);
1427	if (!tu)
1428		return -EINVAL;
1429
1430	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1431				    : BPF_FD_TYPE_UPROBE;
1432	*filename = tu->filename;
1433	*probe_offset = tu->offset;
1434	*probe_addr = 0;
1435	return 0;
1436}
1437#endif	/* CONFIG_PERF_EVENTS */
1438
1439static int
1440trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1441		      void *data)
1442{
1443	struct trace_event_file *file = data;
 
1444
1445	switch (type) {
1446	case TRACE_REG_REGISTER:
1447		return probe_event_enable(event, file, NULL);
1448
1449	case TRACE_REG_UNREGISTER:
1450		probe_event_disable(event, file);
1451		return 0;
1452
1453#ifdef CONFIG_PERF_EVENTS
1454	case TRACE_REG_PERF_REGISTER:
1455		return probe_event_enable(event, NULL, uprobe_perf_filter);
1456
1457	case TRACE_REG_PERF_UNREGISTER:
1458		probe_event_disable(event, NULL);
1459		return 0;
1460
1461	case TRACE_REG_PERF_OPEN:
1462		return uprobe_perf_open(event, data);
1463
1464	case TRACE_REG_PERF_CLOSE:
1465		return uprobe_perf_close(event, data);
1466
1467#endif
1468	default:
1469		return 0;
1470	}
 
1471}
1472
1473static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1474{
1475	struct trace_uprobe *tu;
1476	struct uprobe_dispatch_data udd;
1477	struct uprobe_cpu_buffer *ucb;
1478	int dsize, esize;
1479	int ret = 0;
1480
1481
1482	tu = container_of(con, struct trace_uprobe, consumer);
1483	tu->nhit++;
1484
1485	udd.tu = tu;
1486	udd.bp_addr = instruction_pointer(regs);
1487
1488	current->utask->vaddr = (unsigned long) &udd;
1489
 
 
 
 
 
 
1490	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1491		return 0;
1492
1493	dsize = __get_data_size(&tu->tp, regs);
1494	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1495
1496	ucb = uprobe_buffer_get();
1497	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1498
1499	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1500		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1501
1502#ifdef CONFIG_PERF_EVENTS
1503	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1504		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1505#endif
1506	uprobe_buffer_put(ucb);
1507	return ret;
1508}
1509
1510static int uretprobe_dispatcher(struct uprobe_consumer *con,
1511				unsigned long func, struct pt_regs *regs)
1512{
1513	struct trace_uprobe *tu;
1514	struct uprobe_dispatch_data udd;
1515	struct uprobe_cpu_buffer *ucb;
1516	int dsize, esize;
1517
1518	tu = container_of(con, struct trace_uprobe, consumer);
1519
1520	udd.tu = tu;
1521	udd.bp_addr = func;
1522
1523	current->utask->vaddr = (unsigned long) &udd;
1524
1525	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1526		return 0;
1527
1528	dsize = __get_data_size(&tu->tp, regs);
1529	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1530
1531	ucb = uprobe_buffer_get();
1532	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1533
1534	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1535		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1536
1537#ifdef CONFIG_PERF_EVENTS
1538	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1539		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1540#endif
1541	uprobe_buffer_put(ucb);
1542	return 0;
1543}
1544
1545static struct trace_event_functions uprobe_funcs = {
1546	.trace		= print_uprobe_event
1547};
1548
1549static struct trace_event_fields uprobe_fields_array[] = {
1550	{ .type = TRACE_FUNCTION_TYPE,
1551	  .define_fields = uprobe_event_define_fields },
1552	{}
1553};
1554
1555static inline void init_trace_event_call(struct trace_uprobe *tu)
1556{
1557	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1558	call->event.funcs = &uprobe_funcs;
1559	call->class->fields_array = uprobe_fields_array;
1560
1561	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1562	call->class->reg = trace_uprobe_register;
1563}
1564
1565static int register_uprobe_event(struct trace_uprobe *tu)
1566{
1567	init_trace_event_call(tu);
1568
1569	return trace_probe_register_event_call(&tu->tp);
1570}
1571
1572static int unregister_uprobe_event(struct trace_uprobe *tu)
1573{
1574	return trace_probe_unregister_event_call(&tu->tp);
1575}
1576
1577#ifdef CONFIG_PERF_EVENTS
1578struct trace_event_call *
1579create_local_trace_uprobe(char *name, unsigned long offs,
1580			  unsigned long ref_ctr_offset, bool is_return)
1581{
1582	enum probe_print_type ptype;
1583	struct trace_uprobe *tu;
1584	struct path path;
1585	int ret;
1586
1587	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1588	if (ret)
1589		return ERR_PTR(ret);
 
1590
1591	if (!d_is_reg(path.dentry)) {
1592		path_put(&path);
1593		return ERR_PTR(-EINVAL);
1594	}
1595
1596	/*
1597	 * local trace_kprobes are not added to dyn_event, so they are never
1598	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1599	 * duplicated name "DUMMY_EVENT" here.
1600	 */
1601	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1602				is_return);
1603
1604	if (IS_ERR(tu)) {
1605		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1606			(int)PTR_ERR(tu));
1607		path_put(&path);
1608		return ERR_CAST(tu);
1609	}
 
 
 
 
1610
1611	tu->offset = offs;
1612	tu->path = path;
1613	tu->ref_ctr_offset = ref_ctr_offset;
1614	tu->filename = kstrdup(name, GFP_KERNEL);
1615	if (!tu->filename) {
1616		ret = -ENOMEM;
1617		goto error;
1618	}
1619
1620	init_trace_event_call(tu);
1621
1622	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1623	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1624		ret = -ENOMEM;
1625		goto error;
1626	}
1627
1628	return trace_probe_event_call(&tu->tp);
1629error:
1630	free_trace_uprobe(tu);
1631	return ERR_PTR(ret);
1632}
1633
1634void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1635{
1636	struct trace_uprobe *tu;
1637
1638	tu = trace_uprobe_primary_from_call(event_call);
1639
1640	free_trace_uprobe(tu);
 
 
 
 
1641}
1642#endif /* CONFIG_PERF_EVENTS */
1643
1644/* Make a trace interface for controlling probe points */
1645static __init int init_uprobe_trace(void)
1646{
1647	int ret;
1648
1649	ret = dyn_event_register(&trace_uprobe_ops);
1650	if (ret)
1651		return ret;
1652
1653	ret = tracing_init_dentry();
1654	if (ret)
1655		return 0;
1656
1657	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1658				    NULL, &uprobe_events_ops);
1659	/* Profile interface */
1660	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1661				    NULL, &uprobe_profile_ops);
1662	return 0;
1663}
1664
1665fs_initcall(init_uprobe_trace);