Linux Audio

Check our new training course

Loading...
v4.10.11
   1/*
   2 * Kprobes-based tracing events
   3 *
   4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5 *
   6 * This program is free software; you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License version 2 as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it will be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write to the Free Software
  17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18 */
  19
  20#include <linux/module.h>
  21#include <linux/uaccess.h>
  22
  23#include "trace_probe.h"
  24
  25#define KPROBE_EVENT_SYSTEM "kprobes"
  26
  27/**
  28 * Kprobe event core functions
  29 */
  30struct trace_kprobe {
  31	struct list_head	list;
  32	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
  33	unsigned long __percpu *nhit;
  34	const char		*symbol;	/* symbol name */
  35	struct trace_probe	tp;
  36};
  37
  38#define SIZEOF_TRACE_KPROBE(n)				\
  39	(offsetof(struct trace_kprobe, tp.args) +	\
  40	(sizeof(struct probe_arg) * (n)))
  41
  42
  43static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  44{
  45	return tk->rp.handler != NULL;
  46}
  47
  48static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  49{
  50	return tk->symbol ? tk->symbol : "unknown";
  51}
  52
  53static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  54{
  55	return tk->rp.kp.offset;
  56}
  57
  58static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  59{
  60	return !!(kprobe_gone(&tk->rp.kp));
  61}
  62
  63static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  64						 struct module *mod)
  65{
  66	int len = strlen(mod->name);
  67	const char *name = trace_kprobe_symbol(tk);
  68	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  69}
  70
  71static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  72{
  73	return !!strchr(trace_kprobe_symbol(tk), ':');
  74}
  75
  76static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
  77{
  78	unsigned long nhit = 0;
  79	int cpu;
  80
  81	for_each_possible_cpu(cpu)
  82		nhit += *per_cpu_ptr(tk->nhit, cpu);
  83
  84	return nhit;
  85}
  86
  87static int register_kprobe_event(struct trace_kprobe *tk);
  88static int unregister_kprobe_event(struct trace_kprobe *tk);
  89
  90static DEFINE_MUTEX(probe_lock);
  91static LIST_HEAD(probe_list);
  92
  93static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  94static int kretprobe_dispatcher(struct kretprobe_instance *ri,
  95				struct pt_regs *regs);
  96
  97/* Memory fetching by symbol */
  98struct symbol_cache {
  99	char		*symbol;
 100	long		offset;
 101	unsigned long	addr;
 102};
 103
 104unsigned long update_symbol_cache(struct symbol_cache *sc)
 105{
 106	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 107
 108	if (sc->addr)
 109		sc->addr += sc->offset;
 110
 111	return sc->addr;
 112}
 113
 114void free_symbol_cache(struct symbol_cache *sc)
 115{
 116	kfree(sc->symbol);
 117	kfree(sc);
 118}
 119
 120struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 121{
 122	struct symbol_cache *sc;
 123
 124	if (!sym || strlen(sym) == 0)
 125		return NULL;
 126
 127	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 128	if (!sc)
 129		return NULL;
 130
 131	sc->symbol = kstrdup(sym, GFP_KERNEL);
 132	if (!sc->symbol) {
 133		kfree(sc);
 134		return NULL;
 135	}
 136	sc->offset = offset;
 137	update_symbol_cache(sc);
 138
 139	return sc;
 140}
 141
 142/*
 143 * Kprobes-specific fetch functions
 144 */
 145#define DEFINE_FETCH_stack(type)					\
 146static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,		\
 147					  void *offset, void *dest)	\
 148{									\
 149	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
 150				(unsigned int)((unsigned long)offset));	\
 151}									\
 152NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 153
 154DEFINE_BASIC_FETCH_FUNCS(stack)
 155/* No string on the stack entry */
 156#define fetch_stack_string	NULL
 157#define fetch_stack_string_size	NULL
 158
 159#define DEFINE_FETCH_memory(type)					\
 160static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
 161					  void *addr, void *dest)	\
 162{									\
 163	type retval;							\
 164	if (probe_kernel_address(addr, retval))				\
 165		*(type *)dest = 0;					\
 166	else								\
 167		*(type *)dest = retval;					\
 168}									\
 169NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 170
 171DEFINE_BASIC_FETCH_FUNCS(memory)
 172/*
 173 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 174 * length and relative data location.
 175 */
 176static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 177					    void *addr, void *dest)
 178{
 
 179	int maxlen = get_rloc_len(*(u32 *)dest);
 180	u8 *dst = get_rloc_data(dest);
 181	long ret;
 
 182
 183	if (!maxlen)
 184		return;
 185
 186	/*
 187	 * Try to get string again, since the string can be changed while
 188	 * probing.
 189	 */
 190	ret = strncpy_from_unsafe(dst, addr, maxlen);
 
 
 
 
 
 
 
 
 
 191
 192	if (ret < 0) {	/* Failed to fetch string */
 193		dst[0] = '\0';
 194		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 195	} else {
 196		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 
 197	}
 198}
 199NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 200
 201/* Return the length of string -- including null terminal byte */
 202static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 203						 void *addr, void *dest)
 204{
 205	mm_segment_t old_fs;
 206	int ret, len = 0;
 207	u8 c;
 208
 209	old_fs = get_fs();
 210	set_fs(KERNEL_DS);
 211	pagefault_disable();
 212
 213	do {
 214		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 215		len++;
 216	} while (c && ret == 0 && len < MAX_STRING_SIZE);
 217
 218	pagefault_enable();
 219	set_fs(old_fs);
 220
 221	if (ret < 0)	/* Failed to check the length */
 222		*(u32 *)dest = 0;
 223	else
 224		*(u32 *)dest = len;
 225}
 226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 227
 228#define DEFINE_FETCH_symbol(type)					\
 229void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 
 230{									\
 231	struct symbol_cache *sc = data;					\
 232	if (sc->addr)							\
 233		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
 234	else								\
 235		*(type *)dest = 0;					\
 236}									\
 237NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 238
 239DEFINE_BASIC_FETCH_FUNCS(symbol)
 240DEFINE_FETCH_symbol(string)
 241DEFINE_FETCH_symbol(string_size)
 242
 243/* kprobes don't support file_offset fetch methods */
 244#define fetch_file_offset_u8		NULL
 245#define fetch_file_offset_u16		NULL
 246#define fetch_file_offset_u32		NULL
 247#define fetch_file_offset_u64		NULL
 248#define fetch_file_offset_string	NULL
 249#define fetch_file_offset_string_size	NULL
 250
 251/* Fetch type information table */
 252static const struct fetch_type kprobes_fetch_type_table[] = {
 253	/* Special types */
 254	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 255					sizeof(u32), 1, "__data_loc char[]"),
 256	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 257					string_size, sizeof(u32), 0, "u32"),
 258	/* Basic types */
 259	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 260	ASSIGN_FETCH_TYPE(u16, u16, 0),
 261	ASSIGN_FETCH_TYPE(u32, u32, 0),
 262	ASSIGN_FETCH_TYPE(u64, u64, 0),
 263	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 264	ASSIGN_FETCH_TYPE(s16, u16, 1),
 265	ASSIGN_FETCH_TYPE(s32, u32, 1),
 266	ASSIGN_FETCH_TYPE(s64, u64, 1),
 267	ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 268	ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 269	ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 270	ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 271
 272	ASSIGN_FETCH_TYPE_END
 273};
 274
 275/*
 276 * Allocate new trace_probe and initialize it (including kprobes).
 277 */
 278static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 279					     const char *event,
 280					     void *addr,
 281					     const char *symbol,
 282					     unsigned long offs,
 283					     int nargs, bool is_return)
 284{
 285	struct trace_kprobe *tk;
 286	int ret = -ENOMEM;
 287
 288	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 289	if (!tk)
 290		return ERR_PTR(ret);
 291
 292	tk->nhit = alloc_percpu(unsigned long);
 293	if (!tk->nhit)
 294		goto error;
 295
 296	if (symbol) {
 297		tk->symbol = kstrdup(symbol, GFP_KERNEL);
 298		if (!tk->symbol)
 299			goto error;
 300		tk->rp.kp.symbol_name = tk->symbol;
 301		tk->rp.kp.offset = offs;
 302	} else
 303		tk->rp.kp.addr = addr;
 304
 305	if (is_return)
 306		tk->rp.handler = kretprobe_dispatcher;
 307	else
 308		tk->rp.kp.pre_handler = kprobe_dispatcher;
 309
 310	if (!event || !is_good_name(event)) {
 311		ret = -EINVAL;
 312		goto error;
 313	}
 314
 315	tk->tp.call.class = &tk->tp.class;
 316	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 317	if (!tk->tp.call.name)
 318		goto error;
 319
 320	if (!group || !is_good_name(group)) {
 321		ret = -EINVAL;
 322		goto error;
 323	}
 324
 325	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 326	if (!tk->tp.class.system)
 327		goto error;
 328
 329	INIT_LIST_HEAD(&tk->list);
 330	INIT_LIST_HEAD(&tk->tp.files);
 331	return tk;
 332error:
 333	kfree(tk->tp.call.name);
 334	kfree(tk->symbol);
 335	free_percpu(tk->nhit);
 336	kfree(tk);
 337	return ERR_PTR(ret);
 338}
 339
 340static void free_trace_kprobe(struct trace_kprobe *tk)
 341{
 342	int i;
 343
 344	for (i = 0; i < tk->tp.nr_args; i++)
 345		traceprobe_free_probe_arg(&tk->tp.args[i]);
 346
 347	kfree(tk->tp.call.class->system);
 348	kfree(tk->tp.call.name);
 349	kfree(tk->symbol);
 350	free_percpu(tk->nhit);
 351	kfree(tk);
 352}
 353
 354static struct trace_kprobe *find_trace_kprobe(const char *event,
 355					      const char *group)
 356{
 357	struct trace_kprobe *tk;
 358
 359	list_for_each_entry(tk, &probe_list, list)
 360		if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 361		    strcmp(tk->tp.call.class->system, group) == 0)
 362			return tk;
 363	return NULL;
 364}
 365
 366/*
 367 * Enable trace_probe
 368 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 369 */
 370static int
 371enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 372{
 373	int ret = 0;
 374
 375	if (file) {
 376		struct event_file_link *link;
 377
 378		link = kmalloc(sizeof(*link), GFP_KERNEL);
 379		if (!link) {
 380			ret = -ENOMEM;
 381			goto out;
 382		}
 383
 384		link->file = file;
 385		list_add_tail_rcu(&link->list, &tk->tp.files);
 386
 387		tk->tp.flags |= TP_FLAG_TRACE;
 388	} else
 389		tk->tp.flags |= TP_FLAG_PROFILE;
 390
 391	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 392		if (trace_kprobe_is_return(tk))
 393			ret = enable_kretprobe(&tk->rp);
 394		else
 395			ret = enable_kprobe(&tk->rp.kp);
 396	}
 397 out:
 398	return ret;
 399}
 400
 401/*
 402 * Disable trace_probe
 403 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 404 */
 405static int
 406disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 407{
 408	struct event_file_link *link = NULL;
 409	int wait = 0;
 410	int ret = 0;
 411
 412	if (file) {
 413		link = find_event_file_link(&tk->tp, file);
 414		if (!link) {
 415			ret = -EINVAL;
 416			goto out;
 417		}
 418
 419		list_del_rcu(&link->list);
 420		wait = 1;
 421		if (!list_empty(&tk->tp.files))
 422			goto out;
 423
 424		tk->tp.flags &= ~TP_FLAG_TRACE;
 425	} else
 426		tk->tp.flags &= ~TP_FLAG_PROFILE;
 427
 428	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 429		if (trace_kprobe_is_return(tk))
 430			disable_kretprobe(&tk->rp);
 431		else
 432			disable_kprobe(&tk->rp.kp);
 433		wait = 1;
 434	}
 435 out:
 436	if (wait) {
 437		/*
 438		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
 439		 * to ensure disabled (all running handlers are finished).
 440		 * This is not only for kfree(), but also the caller,
 441		 * trace_remove_event_call() supposes it for releasing
 442		 * event_call related objects, which will be accessed in
 443		 * the kprobe_trace_func/kretprobe_trace_func.
 444		 */
 445		synchronize_sched();
 446		kfree(link);	/* Ignored if link == NULL */
 447	}
 448
 449	return ret;
 450}
 451
 452/* Internal register function - just handle k*probes and flags */
 453static int __register_trace_kprobe(struct trace_kprobe *tk)
 454{
 455	int i, ret;
 456
 457	if (trace_probe_is_registered(&tk->tp))
 458		return -EINVAL;
 459
 460	for (i = 0; i < tk->tp.nr_args; i++)
 461		traceprobe_update_arg(&tk->tp.args[i]);
 462
 463	/* Set/clear disabled flag according to tp->flag */
 464	if (trace_probe_is_enabled(&tk->tp))
 465		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 466	else
 467		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 468
 469	if (trace_kprobe_is_return(tk))
 470		ret = register_kretprobe(&tk->rp);
 471	else
 472		ret = register_kprobe(&tk->rp.kp);
 473
 474	if (ret == 0)
 475		tk->tp.flags |= TP_FLAG_REGISTERED;
 476	else {
 477		pr_warn("Could not insert probe at %s+%lu: %d\n",
 478			trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 479		if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 480			pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 
 481			ret = 0;
 482		} else if (ret == -EILSEQ) {
 483			pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 484				tk->rp.kp.addr);
 
 485			ret = -EINVAL;
 486		}
 487	}
 488
 489	return ret;
 490}
 491
 492/* Internal unregister function - just handle k*probes and flags */
 493static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 494{
 495	if (trace_probe_is_registered(&tk->tp)) {
 496		if (trace_kprobe_is_return(tk))
 497			unregister_kretprobe(&tk->rp);
 498		else
 499			unregister_kprobe(&tk->rp.kp);
 500		tk->tp.flags &= ~TP_FLAG_REGISTERED;
 501		/* Cleanup kprobe for reuse */
 502		if (tk->rp.kp.symbol_name)
 503			tk->rp.kp.addr = NULL;
 504	}
 505}
 506
 507/* Unregister a trace_probe and probe_event: call with locking probe_lock */
 508static int unregister_trace_kprobe(struct trace_kprobe *tk)
 509{
 510	/* Enabled event can not be unregistered */
 511	if (trace_probe_is_enabled(&tk->tp))
 512		return -EBUSY;
 513
 514	/* Will fail if probe is being used by ftrace or perf */
 515	if (unregister_kprobe_event(tk))
 516		return -EBUSY;
 517
 518	__unregister_trace_kprobe(tk);
 519	list_del(&tk->list);
 520
 521	return 0;
 522}
 523
 524/* Register a trace_probe and probe_event */
 525static int register_trace_kprobe(struct trace_kprobe *tk)
 526{
 527	struct trace_kprobe *old_tk;
 528	int ret;
 529
 530	mutex_lock(&probe_lock);
 531
 532	/* Delete old (same name) event if exist */
 533	old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 534			tk->tp.call.class->system);
 535	if (old_tk) {
 536		ret = unregister_trace_kprobe(old_tk);
 537		if (ret < 0)
 538			goto end;
 539		free_trace_kprobe(old_tk);
 540	}
 541
 542	/* Register new event */
 543	ret = register_kprobe_event(tk);
 544	if (ret) {
 545		pr_warn("Failed to register probe event(%d)\n", ret);
 546		goto end;
 547	}
 548
 549	/* Register k*probe */
 550	ret = __register_trace_kprobe(tk);
 551	if (ret < 0)
 552		unregister_kprobe_event(tk);
 553	else
 554		list_add_tail(&tk->list, &probe_list);
 555
 556end:
 557	mutex_unlock(&probe_lock);
 558	return ret;
 559}
 560
 561/* Module notifier call back, checking event on the module */
 562static int trace_kprobe_module_callback(struct notifier_block *nb,
 563				       unsigned long val, void *data)
 564{
 565	struct module *mod = data;
 566	struct trace_kprobe *tk;
 567	int ret;
 568
 569	if (val != MODULE_STATE_COMING)
 570		return NOTIFY_DONE;
 571
 572	/* Update probes on coming module */
 573	mutex_lock(&probe_lock);
 574	list_for_each_entry(tk, &probe_list, list) {
 575		if (trace_kprobe_within_module(tk, mod)) {
 576			/* Don't need to check busy - this should have gone. */
 577			__unregister_trace_kprobe(tk);
 578			ret = __register_trace_kprobe(tk);
 579			if (ret)
 580				pr_warn("Failed to re-register probe %s on %s: %d\n",
 581					trace_event_name(&tk->tp.call),
 582					mod->name, ret);
 
 583		}
 584	}
 585	mutex_unlock(&probe_lock);
 586
 587	return NOTIFY_DONE;
 588}
 589
 590static struct notifier_block trace_kprobe_module_nb = {
 591	.notifier_call = trace_kprobe_module_callback,
 592	.priority = 1	/* Invoked after kprobe module callback */
 593};
 594
 595static int create_trace_kprobe(int argc, char **argv)
 596{
 597	/*
 598	 * Argument syntax:
 599	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 600	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 601	 * Fetch args:
 602	 *  $retval	: fetch return value
 603	 *  $stack	: fetch stack address
 604	 *  $stackN	: fetch Nth of stack (N:0-)
 605	 *  $comm       : fetch current task comm
 606	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
 607	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 608	 *  %REG	: fetch register REG
 609	 * Dereferencing memory fetch:
 610	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 611	 * Alias name of args:
 612	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 613	 * Type of args:
 614	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 615	 */
 616	struct trace_kprobe *tk;
 617	int i, ret = 0;
 618	bool is_return = false, is_delete = false;
 619	char *symbol = NULL, *event = NULL, *group = NULL;
 620	char *arg;
 621	unsigned long offset = 0;
 622	void *addr = NULL;
 623	char buf[MAX_EVENT_NAME_LEN];
 624
 625	/* argc must be >= 1 */
 626	if (argv[0][0] == 'p')
 627		is_return = false;
 628	else if (argv[0][0] == 'r')
 629		is_return = true;
 630	else if (argv[0][0] == '-')
 631		is_delete = true;
 632	else {
 633		pr_info("Probe definition must be started with 'p', 'r' or"
 634			" '-'.\n");
 635		return -EINVAL;
 636	}
 637
 638	if (argv[0][1] == ':') {
 639		event = &argv[0][2];
 640		if (strchr(event, '/')) {
 641			group = event;
 642			event = strchr(group, '/') + 1;
 643			event[-1] = '\0';
 644			if (strlen(group) == 0) {
 645				pr_info("Group name is not specified\n");
 646				return -EINVAL;
 647			}
 648		}
 649		if (strlen(event) == 0) {
 650			pr_info("Event name is not specified\n");
 651			return -EINVAL;
 652		}
 653	}
 654	if (!group)
 655		group = KPROBE_EVENT_SYSTEM;
 656
 657	if (is_delete) {
 658		if (!event) {
 659			pr_info("Delete command needs an event name.\n");
 660			return -EINVAL;
 661		}
 662		mutex_lock(&probe_lock);
 663		tk = find_trace_kprobe(event, group);
 664		if (!tk) {
 665			mutex_unlock(&probe_lock);
 666			pr_info("Event %s/%s doesn't exist.\n", group, event);
 667			return -ENOENT;
 668		}
 669		/* delete an event */
 670		ret = unregister_trace_kprobe(tk);
 671		if (ret == 0)
 672			free_trace_kprobe(tk);
 673		mutex_unlock(&probe_lock);
 674		return ret;
 675	}
 676
 677	if (argc < 2) {
 678		pr_info("Probe point is not specified.\n");
 679		return -EINVAL;
 680	}
 681	if (isdigit(argv[1][0])) {
 682		if (is_return) {
 683			pr_info("Return probe point must be a symbol.\n");
 684			return -EINVAL;
 685		}
 686		/* an address specified */
 687		ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 688		if (ret) {
 689			pr_info("Failed to parse address.\n");
 690			return ret;
 691		}
 692	} else {
 693		/* a symbol specified */
 694		symbol = argv[1];
 695		/* TODO: support .init module functions */
 696		ret = traceprobe_split_symbol_offset(symbol, &offset);
 697		if (ret) {
 698			pr_info("Failed to parse symbol.\n");
 699			return ret;
 700		}
 701		if (offset && is_return) {
 702			pr_info("Return probe must be used without offset.\n");
 703			return -EINVAL;
 704		}
 705	}
 706	argc -= 2; argv += 2;
 707
 708	/* setup a probe */
 709	if (!event) {
 710		/* Make a new event name */
 711		if (symbol)
 712			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 713				 is_return ? 'r' : 'p', symbol, offset);
 714		else
 715			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 716				 is_return ? 'r' : 'p', addr);
 717		event = buf;
 718	}
 719	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
 720			       is_return);
 721	if (IS_ERR(tk)) {
 722		pr_info("Failed to allocate trace_probe.(%d)\n",
 723			(int)PTR_ERR(tk));
 724		return PTR_ERR(tk);
 725	}
 726
 727	/* parse arguments */
 728	ret = 0;
 729	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 730		struct probe_arg *parg = &tk->tp.args[i];
 731
 732		/* Increment count for freeing args in error case */
 733		tk->tp.nr_args++;
 734
 735		/* Parse argument name */
 736		arg = strchr(argv[i], '=');
 737		if (arg) {
 738			*arg++ = '\0';
 739			parg->name = kstrdup(argv[i], GFP_KERNEL);
 740		} else {
 741			arg = argv[i];
 742			/* If argument name is omitted, set "argN" */
 743			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 744			parg->name = kstrdup(buf, GFP_KERNEL);
 745		}
 746
 747		if (!parg->name) {
 748			pr_info("Failed to allocate argument[%d] name.\n", i);
 749			ret = -ENOMEM;
 750			goto error;
 751		}
 752
 753		if (!is_good_name(parg->name)) {
 754			pr_info("Invalid argument[%d] name: %s\n",
 755				i, parg->name);
 756			ret = -EINVAL;
 757			goto error;
 758		}
 759
 760		if (traceprobe_conflict_field_name(parg->name,
 761							tk->tp.args, i)) {
 762			pr_info("Argument[%d] name '%s' conflicts with "
 763				"another field.\n", i, argv[i]);
 764			ret = -EINVAL;
 765			goto error;
 766		}
 767
 768		/* Parse fetch argument */
 769		ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 770						is_return, true,
 771						kprobes_fetch_type_table);
 772		if (ret) {
 773			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 774			goto error;
 775		}
 776	}
 777
 778	ret = register_trace_kprobe(tk);
 779	if (ret)
 780		goto error;
 781	return 0;
 782
 783error:
 784	free_trace_kprobe(tk);
 785	return ret;
 786}
 787
 788static int release_all_trace_kprobes(void)
 789{
 790	struct trace_kprobe *tk;
 791	int ret = 0;
 792
 793	mutex_lock(&probe_lock);
 794	/* Ensure no probe is in use. */
 795	list_for_each_entry(tk, &probe_list, list)
 796		if (trace_probe_is_enabled(&tk->tp)) {
 797			ret = -EBUSY;
 798			goto end;
 799		}
 800	/* TODO: Use batch unregistration */
 801	while (!list_empty(&probe_list)) {
 802		tk = list_entry(probe_list.next, struct trace_kprobe, list);
 803		ret = unregister_trace_kprobe(tk);
 804		if (ret)
 805			goto end;
 806		free_trace_kprobe(tk);
 807	}
 808
 809end:
 810	mutex_unlock(&probe_lock);
 811
 812	return ret;
 813}
 814
 815/* Probes listing interfaces */
 816static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 817{
 818	mutex_lock(&probe_lock);
 819	return seq_list_start(&probe_list, *pos);
 820}
 821
 822static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 823{
 824	return seq_list_next(v, &probe_list, pos);
 825}
 826
 827static void probes_seq_stop(struct seq_file *m, void *v)
 828{
 829	mutex_unlock(&probe_lock);
 830}
 831
 832static int probes_seq_show(struct seq_file *m, void *v)
 833{
 834	struct trace_kprobe *tk = v;
 835	int i;
 836
 837	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 838	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 839			trace_event_name(&tk->tp.call));
 840
 841	if (!tk->symbol)
 842		seq_printf(m, " 0x%p", tk->rp.kp.addr);
 843	else if (tk->rp.kp.offset)
 844		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 845			   tk->rp.kp.offset);
 846	else
 847		seq_printf(m, " %s", trace_kprobe_symbol(tk));
 848
 849	for (i = 0; i < tk->tp.nr_args; i++)
 850		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 851	seq_putc(m, '\n');
 852
 853	return 0;
 854}
 855
 856static const struct seq_operations probes_seq_op = {
 857	.start  = probes_seq_start,
 858	.next   = probes_seq_next,
 859	.stop   = probes_seq_stop,
 860	.show   = probes_seq_show
 861};
 862
 863static int probes_open(struct inode *inode, struct file *file)
 864{
 865	int ret;
 866
 867	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 868		ret = release_all_trace_kprobes();
 869		if (ret < 0)
 870			return ret;
 871	}
 872
 873	return seq_open(file, &probes_seq_op);
 874}
 875
 876static ssize_t probes_write(struct file *file, const char __user *buffer,
 877			    size_t count, loff_t *ppos)
 878{
 879	return traceprobe_probes_write(file, buffer, count, ppos,
 880			create_trace_kprobe);
 881}
 882
 883static const struct file_operations kprobe_events_ops = {
 884	.owner          = THIS_MODULE,
 885	.open           = probes_open,
 886	.read           = seq_read,
 887	.llseek         = seq_lseek,
 888	.release        = seq_release,
 889	.write		= probes_write,
 890};
 891
 892/* Probes profiling interfaces */
 893static int probes_profile_seq_show(struct seq_file *m, void *v)
 894{
 895	struct trace_kprobe *tk = v;
 896
 897	seq_printf(m, "  %-44s %15lu %15lu\n",
 898		   trace_event_name(&tk->tp.call),
 899		   trace_kprobe_nhit(tk),
 900		   tk->rp.kp.nmissed);
 901
 902	return 0;
 903}
 904
 905static const struct seq_operations profile_seq_op = {
 906	.start  = probes_seq_start,
 907	.next   = probes_seq_next,
 908	.stop   = probes_seq_stop,
 909	.show   = probes_profile_seq_show
 910};
 911
 912static int profile_open(struct inode *inode, struct file *file)
 913{
 914	return seq_open(file, &profile_seq_op);
 915}
 916
 917static const struct file_operations kprobe_profile_ops = {
 918	.owner          = THIS_MODULE,
 919	.open           = profile_open,
 920	.read           = seq_read,
 921	.llseek         = seq_lseek,
 922	.release        = seq_release,
 923};
 924
 925/* Kprobe handler */
 926static nokprobe_inline void
 927__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 928		    struct trace_event_file *trace_file)
 929{
 930	struct kprobe_trace_entry_head *entry;
 931	struct ring_buffer_event *event;
 932	struct ring_buffer *buffer;
 933	int size, dsize, pc;
 934	unsigned long irq_flags;
 935	struct trace_event_call *call = &tk->tp.call;
 936
 937	WARN_ON(call != trace_file->event_call);
 938
 939	if (trace_trigger_soft_disabled(trace_file))
 940		return;
 941
 942	local_save_flags(irq_flags);
 943	pc = preempt_count();
 944
 945	dsize = __get_data_size(&tk->tp, regs);
 946	size = sizeof(*entry) + tk->tp.size + dsize;
 947
 948	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 949						call->event.type,
 950						size, irq_flags, pc);
 951	if (!event)
 952		return;
 953
 954	entry = ring_buffer_event_data(event);
 955	entry->ip = (unsigned long)tk->rp.kp.addr;
 956	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 957
 958	event_trigger_unlock_commit_regs(trace_file, buffer, event,
 959					 entry, irq_flags, pc, regs);
 960}
 961
 962static void
 963kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 964{
 965	struct event_file_link *link;
 966
 967	list_for_each_entry_rcu(link, &tk->tp.files, list)
 968		__kprobe_trace_func(tk, regs, link->file);
 969}
 970NOKPROBE_SYMBOL(kprobe_trace_func);
 971
 972/* Kretprobe handler */
 973static nokprobe_inline void
 974__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 975		       struct pt_regs *regs,
 976		       struct trace_event_file *trace_file)
 977{
 978	struct kretprobe_trace_entry_head *entry;
 979	struct ring_buffer_event *event;
 980	struct ring_buffer *buffer;
 981	int size, pc, dsize;
 982	unsigned long irq_flags;
 983	struct trace_event_call *call = &tk->tp.call;
 984
 985	WARN_ON(call != trace_file->event_call);
 986
 987	if (trace_trigger_soft_disabled(trace_file))
 988		return;
 989
 990	local_save_flags(irq_flags);
 991	pc = preempt_count();
 992
 993	dsize = __get_data_size(&tk->tp, regs);
 994	size = sizeof(*entry) + tk->tp.size + dsize;
 995
 996	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 997						call->event.type,
 998						size, irq_flags, pc);
 999	if (!event)
1000		return;
1001
1002	entry = ring_buffer_event_data(event);
1003	entry->func = (unsigned long)tk->rp.kp.addr;
1004	entry->ret_ip = (unsigned long)ri->ret_addr;
1005	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1006
1007	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1008					 entry, irq_flags, pc, regs);
1009}
1010
1011static void
1012kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1013		     struct pt_regs *regs)
1014{
1015	struct event_file_link *link;
1016
1017	list_for_each_entry_rcu(link, &tk->tp.files, list)
1018		__kretprobe_trace_func(tk, ri, regs, link->file);
1019}
1020NOKPROBE_SYMBOL(kretprobe_trace_func);
1021
1022/* Event entry printers */
1023static enum print_line_t
1024print_kprobe_event(struct trace_iterator *iter, int flags,
1025		   struct trace_event *event)
1026{
1027	struct kprobe_trace_entry_head *field;
1028	struct trace_seq *s = &iter->seq;
1029	struct trace_probe *tp;
1030	u8 *data;
1031	int i;
1032
1033	field = (struct kprobe_trace_entry_head *)iter->ent;
1034	tp = container_of(event, struct trace_probe, call.event);
1035
1036	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
 
1037
1038	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1039		goto out;
1040
1041	trace_seq_putc(s, ')');
 
1042
1043	data = (u8 *)&field[1];
1044	for (i = 0; i < tp->nr_args; i++)
1045		if (!tp->args[i].type->print(s, tp->args[i].name,
1046					     data + tp->args[i].offset, field))
1047			goto out;
1048
1049	trace_seq_putc(s, '\n');
1050 out:
1051	return trace_handle_return(s);
 
 
 
1052}
1053
1054static enum print_line_t
1055print_kretprobe_event(struct trace_iterator *iter, int flags,
1056		      struct trace_event *event)
1057{
1058	struct kretprobe_trace_entry_head *field;
1059	struct trace_seq *s = &iter->seq;
1060	struct trace_probe *tp;
1061	u8 *data;
1062	int i;
1063
1064	field = (struct kretprobe_trace_entry_head *)iter->ent;
1065	tp = container_of(event, struct trace_probe, call.event);
1066
1067	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
 
1068
1069	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1070		goto out;
1071
1072	trace_seq_puts(s, " <- ");
 
1073
1074	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1075		goto out;
1076
1077	trace_seq_putc(s, ')');
 
1078
1079	data = (u8 *)&field[1];
1080	for (i = 0; i < tp->nr_args; i++)
1081		if (!tp->args[i].type->print(s, tp->args[i].name,
1082					     data + tp->args[i].offset, field))
1083			goto out;
1084
1085	trace_seq_putc(s, '\n');
 
1086
1087 out:
1088	return trace_handle_return(s);
 
1089}
1090
1091
1092static int kprobe_event_define_fields(struct trace_event_call *event_call)
1093{
1094	int ret, i;
1095	struct kprobe_trace_entry_head field;
1096	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1097
1098	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1099	/* Set argument names as fields */
1100	for (i = 0; i < tk->tp.nr_args; i++) {
1101		struct probe_arg *parg = &tk->tp.args[i];
1102
1103		ret = trace_define_field(event_call, parg->type->fmttype,
1104					 parg->name,
1105					 sizeof(field) + parg->offset,
1106					 parg->type->size,
1107					 parg->type->is_signed,
1108					 FILTER_OTHER);
1109		if (ret)
1110			return ret;
1111	}
1112	return 0;
1113}
1114
1115static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1116{
1117	int ret, i;
1118	struct kretprobe_trace_entry_head field;
1119	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1120
1121	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1122	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1123	/* Set argument names as fields */
1124	for (i = 0; i < tk->tp.nr_args; i++) {
1125		struct probe_arg *parg = &tk->tp.args[i];
1126
1127		ret = trace_define_field(event_call, parg->type->fmttype,
1128					 parg->name,
1129					 sizeof(field) + parg->offset,
1130					 parg->type->size,
1131					 parg->type->is_signed,
1132					 FILTER_OTHER);
1133		if (ret)
1134			return ret;
1135	}
1136	return 0;
1137}
1138
1139#ifdef CONFIG_PERF_EVENTS
1140
1141/* Kprobe profile handler */
1142static void
1143kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1144{
1145	struct trace_event_call *call = &tk->tp.call;
1146	struct bpf_prog *prog = call->prog;
1147	struct kprobe_trace_entry_head *entry;
1148	struct hlist_head *head;
1149	int size, __size, dsize;
1150	int rctx;
1151
1152	if (prog && !trace_call_bpf(prog, regs))
1153		return;
1154
1155	head = this_cpu_ptr(call->perf_events);
1156	if (hlist_empty(head))
1157		return;
1158
1159	dsize = __get_data_size(&tk->tp, regs);
1160	__size = sizeof(*entry) + tk->tp.size + dsize;
1161	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1162	size -= sizeof(u32);
1163
1164	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1165	if (!entry)
1166		return;
1167
1168	entry->ip = (unsigned long)tk->rp.kp.addr;
1169	memset(&entry[1], 0, dsize);
1170	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1171	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1172			      head, NULL);
1173}
1174NOKPROBE_SYMBOL(kprobe_perf_func);
1175
1176/* Kretprobe profile handler */
1177static void
1178kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1179		    struct pt_regs *regs)
1180{
1181	struct trace_event_call *call = &tk->tp.call;
1182	struct bpf_prog *prog = call->prog;
1183	struct kretprobe_trace_entry_head *entry;
1184	struct hlist_head *head;
1185	int size, __size, dsize;
1186	int rctx;
1187
1188	if (prog && !trace_call_bpf(prog, regs))
1189		return;
1190
1191	head = this_cpu_ptr(call->perf_events);
1192	if (hlist_empty(head))
1193		return;
1194
1195	dsize = __get_data_size(&tk->tp, regs);
1196	__size = sizeof(*entry) + tk->tp.size + dsize;
1197	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1198	size -= sizeof(u32);
1199
1200	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1201	if (!entry)
1202		return;
1203
1204	entry->func = (unsigned long)tk->rp.kp.addr;
1205	entry->ret_ip = (unsigned long)ri->ret_addr;
1206	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1207	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1208			      head, NULL);
1209}
1210NOKPROBE_SYMBOL(kretprobe_perf_func);
1211#endif	/* CONFIG_PERF_EVENTS */
1212
1213/*
1214 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1215 *
1216 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1217 * lockless, but we can't race with this __init function.
1218 */
1219static int kprobe_register(struct trace_event_call *event,
1220			   enum trace_reg type, void *data)
 
1221{
1222	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1223	struct trace_event_file *file = data;
1224
1225	switch (type) {
1226	case TRACE_REG_REGISTER:
1227		return enable_trace_kprobe(tk, file);
1228	case TRACE_REG_UNREGISTER:
1229		return disable_trace_kprobe(tk, file);
1230
1231#ifdef CONFIG_PERF_EVENTS
1232	case TRACE_REG_PERF_REGISTER:
1233		return enable_trace_kprobe(tk, NULL);
1234	case TRACE_REG_PERF_UNREGISTER:
1235		return disable_trace_kprobe(tk, NULL);
1236	case TRACE_REG_PERF_OPEN:
1237	case TRACE_REG_PERF_CLOSE:
1238	case TRACE_REG_PERF_ADD:
1239	case TRACE_REG_PERF_DEL:
1240		return 0;
1241#endif
1242	}
1243	return 0;
1244}
1245
1246static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
 
1247{
1248	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1249
1250	raw_cpu_inc(*tk->nhit);
1251
1252	if (tk->tp.flags & TP_FLAG_TRACE)
1253		kprobe_trace_func(tk, regs);
1254#ifdef CONFIG_PERF_EVENTS
1255	if (tk->tp.flags & TP_FLAG_PROFILE)
1256		kprobe_perf_func(tk, regs);
1257#endif
1258	return 0;	/* We don't tweek kernel, so just return 0 */
1259}
1260NOKPROBE_SYMBOL(kprobe_dispatcher);
1261
1262static int
1263kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1264{
1265	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1266
1267	raw_cpu_inc(*tk->nhit);
1268
1269	if (tk->tp.flags & TP_FLAG_TRACE)
1270		kretprobe_trace_func(tk, ri, regs);
1271#ifdef CONFIG_PERF_EVENTS
1272	if (tk->tp.flags & TP_FLAG_PROFILE)
1273		kretprobe_perf_func(tk, ri, regs);
1274#endif
1275	return 0;	/* We don't tweek kernel, so just return 0 */
1276}
1277NOKPROBE_SYMBOL(kretprobe_dispatcher);
1278
1279static struct trace_event_functions kretprobe_funcs = {
1280	.trace		= print_kretprobe_event
1281};
1282
1283static struct trace_event_functions kprobe_funcs = {
1284	.trace		= print_kprobe_event
1285};
1286
1287static int register_kprobe_event(struct trace_kprobe *tk)
1288{
1289	struct trace_event_call *call = &tk->tp.call;
1290	int ret;
1291
1292	/* Initialize trace_event_call */
1293	INIT_LIST_HEAD(&call->class->fields);
1294	if (trace_kprobe_is_return(tk)) {
1295		call->event.funcs = &kretprobe_funcs;
1296		call->class->define_fields = kretprobe_event_define_fields;
1297	} else {
1298		call->event.funcs = &kprobe_funcs;
1299		call->class->define_fields = kprobe_event_define_fields;
1300	}
1301	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1302		return -ENOMEM;
1303	ret = register_trace_event(&call->event);
1304	if (!ret) {
1305		kfree(call->print_fmt);
1306		return -ENODEV;
1307	}
1308	call->flags = TRACE_EVENT_FL_KPROBE;
1309	call->class->reg = kprobe_register;
1310	call->data = tk;
1311	ret = trace_add_event_call(call);
1312	if (ret) {
1313		pr_info("Failed to register kprobe event: %s\n",
1314			trace_event_name(call));
1315		kfree(call->print_fmt);
1316		unregister_trace_event(&call->event);
1317	}
1318	return ret;
1319}
1320
1321static int unregister_kprobe_event(struct trace_kprobe *tk)
1322{
1323	int ret;
1324
1325	/* tp->event is unregistered in trace_remove_event_call() */
1326	ret = trace_remove_event_call(&tk->tp.call);
1327	if (!ret)
1328		kfree(tk->tp.call.print_fmt);
1329	return ret;
1330}
1331
1332/* Make a tracefs interface for controlling probe points */
1333static __init int init_kprobe_trace(void)
1334{
1335	struct dentry *d_tracer;
1336	struct dentry *entry;
1337
1338	if (register_module_notifier(&trace_kprobe_module_nb))
1339		return -EINVAL;
1340
1341	d_tracer = tracing_init_dentry();
1342	if (IS_ERR(d_tracer))
1343		return 0;
1344
1345	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1346				    NULL, &kprobe_events_ops);
1347
1348	/* Event list interface */
1349	if (!entry)
1350		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
 
1351
1352	/* Profile interface */
1353	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1354				    NULL, &kprobe_profile_ops);
1355
1356	if (!entry)
1357		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
 
1358	return 0;
1359}
1360fs_initcall(init_kprobe_trace);
1361
1362
1363#ifdef CONFIG_FTRACE_STARTUP_TEST
 
1364/*
1365 * The "__used" keeps gcc from removing the function symbol
1366 * from the kallsyms table. 'noinline' makes sure that there
1367 * isn't an inlined version used by the test method below
1368 */
1369static __used __init noinline int
1370kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1371{
1372	return a1 + a2 + a3 + a4 + a5 + a6;
1373}
1374
1375static __init struct trace_event_file *
1376find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1377{
1378	struct trace_event_file *file;
1379
1380	list_for_each_entry(file, &tr->events, list)
1381		if (file->event_call == &tk->tp.call)
1382			return file;
1383
1384	return NULL;
1385}
1386
1387/*
1388 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1389 * stage, we can do this lockless.
1390 */
1391static __init int kprobe_trace_self_tests_init(void)
1392{
1393	int ret, warn = 0;
1394	int (*target)(int, int, int, int, int, int);
1395	struct trace_kprobe *tk;
1396	struct trace_event_file *file;
1397
1398	if (tracing_is_disabled())
1399		return -ENODEV;
1400
1401	target = kprobe_trace_selftest_target;
1402
1403	pr_info("Testing kprobe tracing: ");
1404
1405	ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1406				  "$stack $stack0 +0($stack)",
1407				  create_trace_kprobe);
1408	if (WARN_ON_ONCE(ret)) {
1409		pr_warn("error on probing function entry.\n");
1410		warn++;
1411	} else {
1412		/* Enable trace point */
1413		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1414		if (WARN_ON_ONCE(tk == NULL)) {
1415			pr_warn("error on getting new probe.\n");
1416			warn++;
1417		} else {
1418			file = find_trace_probe_file(tk, top_trace_array());
1419			if (WARN_ON_ONCE(file == NULL)) {
1420				pr_warn("error on getting probe file.\n");
1421				warn++;
1422			} else
1423				enable_trace_kprobe(tk, file);
1424		}
1425	}
1426
1427	ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1428				  "$retval", create_trace_kprobe);
1429	if (WARN_ON_ONCE(ret)) {
1430		pr_warn("error on probing function return.\n");
1431		warn++;
1432	} else {
1433		/* Enable trace point */
1434		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1435		if (WARN_ON_ONCE(tk == NULL)) {
1436			pr_warn("error on getting 2nd new probe.\n");
1437			warn++;
1438		} else {
1439			file = find_trace_probe_file(tk, top_trace_array());
1440			if (WARN_ON_ONCE(file == NULL)) {
1441				pr_warn("error on getting probe file.\n");
1442				warn++;
1443			} else
1444				enable_trace_kprobe(tk, file);
1445		}
1446	}
1447
1448	if (warn)
1449		goto end;
1450
1451	ret = target(1, 2, 3, 4, 5, 6);
1452
1453	/*
1454	 * Not expecting an error here, the check is only to prevent the
1455	 * optimizer from removing the call to target() as otherwise there
1456	 * are no side-effects and the call is never performed.
1457	 */
1458	if (ret != 21)
1459		warn++;
1460
1461	/* Disable trace points before removing it */
1462	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1463	if (WARN_ON_ONCE(tk == NULL)) {
1464		pr_warn("error on getting test probe.\n");
1465		warn++;
1466	} else {
1467		if (trace_kprobe_nhit(tk) != 1) {
1468			pr_warn("incorrect number of testprobe hits\n");
1469			warn++;
1470		}
1471
1472		file = find_trace_probe_file(tk, top_trace_array());
1473		if (WARN_ON_ONCE(file == NULL)) {
1474			pr_warn("error on getting probe file.\n");
1475			warn++;
1476		} else
1477			disable_trace_kprobe(tk, file);
1478	}
1479
1480	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1481	if (WARN_ON_ONCE(tk == NULL)) {
1482		pr_warn("error on getting 2nd test probe.\n");
1483		warn++;
1484	} else {
1485		if (trace_kprobe_nhit(tk) != 1) {
1486			pr_warn("incorrect number of testprobe2 hits\n");
1487			warn++;
1488		}
1489
1490		file = find_trace_probe_file(tk, top_trace_array());
1491		if (WARN_ON_ONCE(file == NULL)) {
1492			pr_warn("error on getting probe file.\n");
1493			warn++;
1494		} else
1495			disable_trace_kprobe(tk, file);
1496	}
1497
1498	ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1499	if (WARN_ON_ONCE(ret)) {
1500		pr_warn("error on deleting a probe.\n");
1501		warn++;
1502	}
1503
1504	ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1505	if (WARN_ON_ONCE(ret)) {
1506		pr_warn("error on deleting a probe.\n");
1507		warn++;
1508	}
1509
1510end:
1511	release_all_trace_kprobes();
1512	if (warn)
1513		pr_cont("NG: Some tests are failed. Please check them.\n");
1514	else
1515		pr_cont("OK\n");
1516	return 0;
1517}
1518
1519late_initcall(kprobe_trace_self_tests_init);
1520
1521#endif
v3.15
   1/*
   2 * Kprobes-based tracing events
   3 *
   4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5 *
   6 * This program is free software; you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License version 2 as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it will be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write to the Free Software
  17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18 */
  19
  20#include <linux/module.h>
  21#include <linux/uaccess.h>
  22
  23#include "trace_probe.h"
  24
  25#define KPROBE_EVENT_SYSTEM "kprobes"
  26
  27/**
  28 * Kprobe event core functions
  29 */
  30struct trace_kprobe {
  31	struct list_head	list;
  32	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
  33	unsigned long 		nhit;
  34	const char		*symbol;	/* symbol name */
  35	struct trace_probe	tp;
  36};
  37
  38#define SIZEOF_TRACE_KPROBE(n)				\
  39	(offsetof(struct trace_kprobe, tp.args) +	\
  40	(sizeof(struct probe_arg) * (n)))
  41
  42
  43static __kprobes bool trace_kprobe_is_return(struct trace_kprobe *tk)
  44{
  45	return tk->rp.handler != NULL;
  46}
  47
  48static __kprobes const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  49{
  50	return tk->symbol ? tk->symbol : "unknown";
  51}
  52
  53static __kprobes unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  54{
  55	return tk->rp.kp.offset;
  56}
  57
  58static __kprobes bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  59{
  60	return !!(kprobe_gone(&tk->rp.kp));
  61}
  62
  63static __kprobes bool trace_kprobe_within_module(struct trace_kprobe *tk,
  64						 struct module *mod)
  65{
  66	int len = strlen(mod->name);
  67	const char *name = trace_kprobe_symbol(tk);
  68	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  69}
  70
  71static __kprobes bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  72{
  73	return !!strchr(trace_kprobe_symbol(tk), ':');
  74}
  75
 
 
 
 
 
 
 
 
 
 
 
  76static int register_kprobe_event(struct trace_kprobe *tk);
  77static int unregister_kprobe_event(struct trace_kprobe *tk);
  78
  79static DEFINE_MUTEX(probe_lock);
  80static LIST_HEAD(probe_list);
  81
  82static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  83static int kretprobe_dispatcher(struct kretprobe_instance *ri,
  84				struct pt_regs *regs);
  85
  86/* Memory fetching by symbol */
  87struct symbol_cache {
  88	char		*symbol;
  89	long		offset;
  90	unsigned long	addr;
  91};
  92
  93unsigned long update_symbol_cache(struct symbol_cache *sc)
  94{
  95	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
  96
  97	if (sc->addr)
  98		sc->addr += sc->offset;
  99
 100	return sc->addr;
 101}
 102
 103void free_symbol_cache(struct symbol_cache *sc)
 104{
 105	kfree(sc->symbol);
 106	kfree(sc);
 107}
 108
 109struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 110{
 111	struct symbol_cache *sc;
 112
 113	if (!sym || strlen(sym) == 0)
 114		return NULL;
 115
 116	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 117	if (!sc)
 118		return NULL;
 119
 120	sc->symbol = kstrdup(sym, GFP_KERNEL);
 121	if (!sc->symbol) {
 122		kfree(sc);
 123		return NULL;
 124	}
 125	sc->offset = offset;
 126	update_symbol_cache(sc);
 127
 128	return sc;
 129}
 130
 131/*
 132 * Kprobes-specific fetch functions
 133 */
 134#define DEFINE_FETCH_stack(type)					\
 135static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
 136					  void *offset, void *dest)	\
 137{									\
 138	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
 139				(unsigned int)((unsigned long)offset));	\
 140}
 
 
 141DEFINE_BASIC_FETCH_FUNCS(stack)
 142/* No string on the stack entry */
 143#define fetch_stack_string	NULL
 144#define fetch_stack_string_size	NULL
 145
 146#define DEFINE_FETCH_memory(type)					\
 147static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
 148					  void *addr, void *dest)	\
 149{									\
 150	type retval;							\
 151	if (probe_kernel_address(addr, retval))				\
 152		*(type *)dest = 0;					\
 153	else								\
 154		*(type *)dest = retval;					\
 155}
 
 
 156DEFINE_BASIC_FETCH_FUNCS(memory)
 157/*
 158 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 159 * length and relative data location.
 160 */
 161static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 162						      void *addr, void *dest)
 163{
 164	long ret;
 165	int maxlen = get_rloc_len(*(u32 *)dest);
 166	u8 *dst = get_rloc_data(dest);
 167	u8 *src = addr;
 168	mm_segment_t old_fs = get_fs();
 169
 170	if (!maxlen)
 171		return;
 172
 173	/*
 174	 * Try to get string again, since the string can be changed while
 175	 * probing.
 176	 */
 177	set_fs(KERNEL_DS);
 178	pagefault_disable();
 179
 180	do
 181		ret = __copy_from_user_inatomic(dst++, src++, 1);
 182	while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
 183
 184	dst[-1] = '\0';
 185	pagefault_enable();
 186	set_fs(old_fs);
 187
 188	if (ret < 0) {	/* Failed to fetch string */
 189		((u8 *)get_rloc_data(dest))[0] = '\0';
 190		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 191	} else {
 192		*(u32 *)dest = make_data_rloc(src - (u8 *)addr,
 193					      get_rloc_offs(*(u32 *)dest));
 194	}
 195}
 
 196
 197/* Return the length of string -- including null terminal byte */
 198static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 199							void *addr, void *dest)
 200{
 201	mm_segment_t old_fs;
 202	int ret, len = 0;
 203	u8 c;
 204
 205	old_fs = get_fs();
 206	set_fs(KERNEL_DS);
 207	pagefault_disable();
 208
 209	do {
 210		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 211		len++;
 212	} while (c && ret == 0 && len < MAX_STRING_SIZE);
 213
 214	pagefault_enable();
 215	set_fs(old_fs);
 216
 217	if (ret < 0)	/* Failed to check the length */
 218		*(u32 *)dest = 0;
 219	else
 220		*(u32 *)dest = len;
 221}
 
 222
 223#define DEFINE_FETCH_symbol(type)					\
 224__kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs,	\
 225					  void *data, void *dest)	\
 226{									\
 227	struct symbol_cache *sc = data;					\
 228	if (sc->addr)							\
 229		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
 230	else								\
 231		*(type *)dest = 0;					\
 232}
 
 
 233DEFINE_BASIC_FETCH_FUNCS(symbol)
 234DEFINE_FETCH_symbol(string)
 235DEFINE_FETCH_symbol(string_size)
 236
 237/* kprobes don't support file_offset fetch methods */
 238#define fetch_file_offset_u8		NULL
 239#define fetch_file_offset_u16		NULL
 240#define fetch_file_offset_u32		NULL
 241#define fetch_file_offset_u64		NULL
 242#define fetch_file_offset_string	NULL
 243#define fetch_file_offset_string_size	NULL
 244
 245/* Fetch type information table */
 246const struct fetch_type kprobes_fetch_type_table[] = {
 247	/* Special types */
 248	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 249					sizeof(u32), 1, "__data_loc char[]"),
 250	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 251					string_size, sizeof(u32), 0, "u32"),
 252	/* Basic types */
 253	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 254	ASSIGN_FETCH_TYPE(u16, u16, 0),
 255	ASSIGN_FETCH_TYPE(u32, u32, 0),
 256	ASSIGN_FETCH_TYPE(u64, u64, 0),
 257	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 258	ASSIGN_FETCH_TYPE(s16, u16, 1),
 259	ASSIGN_FETCH_TYPE(s32, u32, 1),
 260	ASSIGN_FETCH_TYPE(s64, u64, 1),
 
 
 
 
 261
 262	ASSIGN_FETCH_TYPE_END
 263};
 264
 265/*
 266 * Allocate new trace_probe and initialize it (including kprobes).
 267 */
 268static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 269					     const char *event,
 270					     void *addr,
 271					     const char *symbol,
 272					     unsigned long offs,
 273					     int nargs, bool is_return)
 274{
 275	struct trace_kprobe *tk;
 276	int ret = -ENOMEM;
 277
 278	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 279	if (!tk)
 280		return ERR_PTR(ret);
 281
 
 
 
 
 282	if (symbol) {
 283		tk->symbol = kstrdup(symbol, GFP_KERNEL);
 284		if (!tk->symbol)
 285			goto error;
 286		tk->rp.kp.symbol_name = tk->symbol;
 287		tk->rp.kp.offset = offs;
 288	} else
 289		tk->rp.kp.addr = addr;
 290
 291	if (is_return)
 292		tk->rp.handler = kretprobe_dispatcher;
 293	else
 294		tk->rp.kp.pre_handler = kprobe_dispatcher;
 295
 296	if (!event || !is_good_name(event)) {
 297		ret = -EINVAL;
 298		goto error;
 299	}
 300
 301	tk->tp.call.class = &tk->tp.class;
 302	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 303	if (!tk->tp.call.name)
 304		goto error;
 305
 306	if (!group || !is_good_name(group)) {
 307		ret = -EINVAL;
 308		goto error;
 309	}
 310
 311	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 312	if (!tk->tp.class.system)
 313		goto error;
 314
 315	INIT_LIST_HEAD(&tk->list);
 316	INIT_LIST_HEAD(&tk->tp.files);
 317	return tk;
 318error:
 319	kfree(tk->tp.call.name);
 320	kfree(tk->symbol);
 
 321	kfree(tk);
 322	return ERR_PTR(ret);
 323}
 324
 325static void free_trace_kprobe(struct trace_kprobe *tk)
 326{
 327	int i;
 328
 329	for (i = 0; i < tk->tp.nr_args; i++)
 330		traceprobe_free_probe_arg(&tk->tp.args[i]);
 331
 332	kfree(tk->tp.call.class->system);
 333	kfree(tk->tp.call.name);
 334	kfree(tk->symbol);
 
 335	kfree(tk);
 336}
 337
 338static struct trace_kprobe *find_trace_kprobe(const char *event,
 339					      const char *group)
 340{
 341	struct trace_kprobe *tk;
 342
 343	list_for_each_entry(tk, &probe_list, list)
 344		if (strcmp(ftrace_event_name(&tk->tp.call), event) == 0 &&
 345		    strcmp(tk->tp.call.class->system, group) == 0)
 346			return tk;
 347	return NULL;
 348}
 349
 350/*
 351 * Enable trace_probe
 352 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 353 */
 354static int
 355enable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
 356{
 357	int ret = 0;
 358
 359	if (file) {
 360		struct event_file_link *link;
 361
 362		link = kmalloc(sizeof(*link), GFP_KERNEL);
 363		if (!link) {
 364			ret = -ENOMEM;
 365			goto out;
 366		}
 367
 368		link->file = file;
 369		list_add_tail_rcu(&link->list, &tk->tp.files);
 370
 371		tk->tp.flags |= TP_FLAG_TRACE;
 372	} else
 373		tk->tp.flags |= TP_FLAG_PROFILE;
 374
 375	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 376		if (trace_kprobe_is_return(tk))
 377			ret = enable_kretprobe(&tk->rp);
 378		else
 379			ret = enable_kprobe(&tk->rp.kp);
 380	}
 381 out:
 382	return ret;
 383}
 384
 385/*
 386 * Disable trace_probe
 387 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 388 */
 389static int
 390disable_trace_kprobe(struct trace_kprobe *tk, struct ftrace_event_file *file)
 391{
 392	struct event_file_link *link = NULL;
 393	int wait = 0;
 394	int ret = 0;
 395
 396	if (file) {
 397		link = find_event_file_link(&tk->tp, file);
 398		if (!link) {
 399			ret = -EINVAL;
 400			goto out;
 401		}
 402
 403		list_del_rcu(&link->list);
 404		wait = 1;
 405		if (!list_empty(&tk->tp.files))
 406			goto out;
 407
 408		tk->tp.flags &= ~TP_FLAG_TRACE;
 409	} else
 410		tk->tp.flags &= ~TP_FLAG_PROFILE;
 411
 412	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 413		if (trace_kprobe_is_return(tk))
 414			disable_kretprobe(&tk->rp);
 415		else
 416			disable_kprobe(&tk->rp.kp);
 417		wait = 1;
 418	}
 419 out:
 420	if (wait) {
 421		/*
 422		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
 423		 * to ensure disabled (all running handlers are finished).
 424		 * This is not only for kfree(), but also the caller,
 425		 * trace_remove_event_call() supposes it for releasing
 426		 * event_call related objects, which will be accessed in
 427		 * the kprobe_trace_func/kretprobe_trace_func.
 428		 */
 429		synchronize_sched();
 430		kfree(link);	/* Ignored if link == NULL */
 431	}
 432
 433	return ret;
 434}
 435
 436/* Internal register function - just handle k*probes and flags */
 437static int __register_trace_kprobe(struct trace_kprobe *tk)
 438{
 439	int i, ret;
 440
 441	if (trace_probe_is_registered(&tk->tp))
 442		return -EINVAL;
 443
 444	for (i = 0; i < tk->tp.nr_args; i++)
 445		traceprobe_update_arg(&tk->tp.args[i]);
 446
 447	/* Set/clear disabled flag according to tp->flag */
 448	if (trace_probe_is_enabled(&tk->tp))
 449		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 450	else
 451		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 452
 453	if (trace_kprobe_is_return(tk))
 454		ret = register_kretprobe(&tk->rp);
 455	else
 456		ret = register_kprobe(&tk->rp.kp);
 457
 458	if (ret == 0)
 459		tk->tp.flags |= TP_FLAG_REGISTERED;
 460	else {
 461		pr_warning("Could not insert probe at %s+%lu: %d\n",
 462			   trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 463		if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 464			pr_warning("This probe might be able to register after"
 465				   "target module is loaded. Continue.\n");
 466			ret = 0;
 467		} else if (ret == -EILSEQ) {
 468			pr_warning("Probing address(0x%p) is not an "
 469				   "instruction boundary.\n",
 470				   tk->rp.kp.addr);
 471			ret = -EINVAL;
 472		}
 473	}
 474
 475	return ret;
 476}
 477
 478/* Internal unregister function - just handle k*probes and flags */
 479static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 480{
 481	if (trace_probe_is_registered(&tk->tp)) {
 482		if (trace_kprobe_is_return(tk))
 483			unregister_kretprobe(&tk->rp);
 484		else
 485			unregister_kprobe(&tk->rp.kp);
 486		tk->tp.flags &= ~TP_FLAG_REGISTERED;
 487		/* Cleanup kprobe for reuse */
 488		if (tk->rp.kp.symbol_name)
 489			tk->rp.kp.addr = NULL;
 490	}
 491}
 492
 493/* Unregister a trace_probe and probe_event: call with locking probe_lock */
 494static int unregister_trace_kprobe(struct trace_kprobe *tk)
 495{
 496	/* Enabled event can not be unregistered */
 497	if (trace_probe_is_enabled(&tk->tp))
 498		return -EBUSY;
 499
 500	/* Will fail if probe is being used by ftrace or perf */
 501	if (unregister_kprobe_event(tk))
 502		return -EBUSY;
 503
 504	__unregister_trace_kprobe(tk);
 505	list_del(&tk->list);
 506
 507	return 0;
 508}
 509
 510/* Register a trace_probe and probe_event */
 511static int register_trace_kprobe(struct trace_kprobe *tk)
 512{
 513	struct trace_kprobe *old_tk;
 514	int ret;
 515
 516	mutex_lock(&probe_lock);
 517
 518	/* Delete old (same name) event if exist */
 519	old_tk = find_trace_kprobe(ftrace_event_name(&tk->tp.call),
 520			tk->tp.call.class->system);
 521	if (old_tk) {
 522		ret = unregister_trace_kprobe(old_tk);
 523		if (ret < 0)
 524			goto end;
 525		free_trace_kprobe(old_tk);
 526	}
 527
 528	/* Register new event */
 529	ret = register_kprobe_event(tk);
 530	if (ret) {
 531		pr_warning("Failed to register probe event(%d)\n", ret);
 532		goto end;
 533	}
 534
 535	/* Register k*probe */
 536	ret = __register_trace_kprobe(tk);
 537	if (ret < 0)
 538		unregister_kprobe_event(tk);
 539	else
 540		list_add_tail(&tk->list, &probe_list);
 541
 542end:
 543	mutex_unlock(&probe_lock);
 544	return ret;
 545}
 546
 547/* Module notifier call back, checking event on the module */
 548static int trace_kprobe_module_callback(struct notifier_block *nb,
 549				       unsigned long val, void *data)
 550{
 551	struct module *mod = data;
 552	struct trace_kprobe *tk;
 553	int ret;
 554
 555	if (val != MODULE_STATE_COMING)
 556		return NOTIFY_DONE;
 557
 558	/* Update probes on coming module */
 559	mutex_lock(&probe_lock);
 560	list_for_each_entry(tk, &probe_list, list) {
 561		if (trace_kprobe_within_module(tk, mod)) {
 562			/* Don't need to check busy - this should have gone. */
 563			__unregister_trace_kprobe(tk);
 564			ret = __register_trace_kprobe(tk);
 565			if (ret)
 566				pr_warning("Failed to re-register probe %s on"
 567					   "%s: %d\n",
 568					   ftrace_event_name(&tk->tp.call),
 569					   mod->name, ret);
 570		}
 571	}
 572	mutex_unlock(&probe_lock);
 573
 574	return NOTIFY_DONE;
 575}
 576
 577static struct notifier_block trace_kprobe_module_nb = {
 578	.notifier_call = trace_kprobe_module_callback,
 579	.priority = 1	/* Invoked after kprobe module callback */
 580};
 581
 582static int create_trace_kprobe(int argc, char **argv)
 583{
 584	/*
 585	 * Argument syntax:
 586	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 587	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 588	 * Fetch args:
 589	 *  $retval	: fetch return value
 590	 *  $stack	: fetch stack address
 591	 *  $stackN	: fetch Nth of stack (N:0-)
 
 592	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
 593	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 594	 *  %REG	: fetch register REG
 595	 * Dereferencing memory fetch:
 596	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 597	 * Alias name of args:
 598	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 599	 * Type of args:
 600	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 601	 */
 602	struct trace_kprobe *tk;
 603	int i, ret = 0;
 604	bool is_return = false, is_delete = false;
 605	char *symbol = NULL, *event = NULL, *group = NULL;
 606	char *arg;
 607	unsigned long offset = 0;
 608	void *addr = NULL;
 609	char buf[MAX_EVENT_NAME_LEN];
 610
 611	/* argc must be >= 1 */
 612	if (argv[0][0] == 'p')
 613		is_return = false;
 614	else if (argv[0][0] == 'r')
 615		is_return = true;
 616	else if (argv[0][0] == '-')
 617		is_delete = true;
 618	else {
 619		pr_info("Probe definition must be started with 'p', 'r' or"
 620			" '-'.\n");
 621		return -EINVAL;
 622	}
 623
 624	if (argv[0][1] == ':') {
 625		event = &argv[0][2];
 626		if (strchr(event, '/')) {
 627			group = event;
 628			event = strchr(group, '/') + 1;
 629			event[-1] = '\0';
 630			if (strlen(group) == 0) {
 631				pr_info("Group name is not specified\n");
 632				return -EINVAL;
 633			}
 634		}
 635		if (strlen(event) == 0) {
 636			pr_info("Event name is not specified\n");
 637			return -EINVAL;
 638		}
 639	}
 640	if (!group)
 641		group = KPROBE_EVENT_SYSTEM;
 642
 643	if (is_delete) {
 644		if (!event) {
 645			pr_info("Delete command needs an event name.\n");
 646			return -EINVAL;
 647		}
 648		mutex_lock(&probe_lock);
 649		tk = find_trace_kprobe(event, group);
 650		if (!tk) {
 651			mutex_unlock(&probe_lock);
 652			pr_info("Event %s/%s doesn't exist.\n", group, event);
 653			return -ENOENT;
 654		}
 655		/* delete an event */
 656		ret = unregister_trace_kprobe(tk);
 657		if (ret == 0)
 658			free_trace_kprobe(tk);
 659		mutex_unlock(&probe_lock);
 660		return ret;
 661	}
 662
 663	if (argc < 2) {
 664		pr_info("Probe point is not specified.\n");
 665		return -EINVAL;
 666	}
 667	if (isdigit(argv[1][0])) {
 668		if (is_return) {
 669			pr_info("Return probe point must be a symbol.\n");
 670			return -EINVAL;
 671		}
 672		/* an address specified */
 673		ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 674		if (ret) {
 675			pr_info("Failed to parse address.\n");
 676			return ret;
 677		}
 678	} else {
 679		/* a symbol specified */
 680		symbol = argv[1];
 681		/* TODO: support .init module functions */
 682		ret = traceprobe_split_symbol_offset(symbol, &offset);
 683		if (ret) {
 684			pr_info("Failed to parse symbol.\n");
 685			return ret;
 686		}
 687		if (offset && is_return) {
 688			pr_info("Return probe must be used without offset.\n");
 689			return -EINVAL;
 690		}
 691	}
 692	argc -= 2; argv += 2;
 693
 694	/* setup a probe */
 695	if (!event) {
 696		/* Make a new event name */
 697		if (symbol)
 698			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 699				 is_return ? 'r' : 'p', symbol, offset);
 700		else
 701			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 702				 is_return ? 'r' : 'p', addr);
 703		event = buf;
 704	}
 705	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
 706			       is_return);
 707	if (IS_ERR(tk)) {
 708		pr_info("Failed to allocate trace_probe.(%d)\n",
 709			(int)PTR_ERR(tk));
 710		return PTR_ERR(tk);
 711	}
 712
 713	/* parse arguments */
 714	ret = 0;
 715	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 716		struct probe_arg *parg = &tk->tp.args[i];
 717
 718		/* Increment count for freeing args in error case */
 719		tk->tp.nr_args++;
 720
 721		/* Parse argument name */
 722		arg = strchr(argv[i], '=');
 723		if (arg) {
 724			*arg++ = '\0';
 725			parg->name = kstrdup(argv[i], GFP_KERNEL);
 726		} else {
 727			arg = argv[i];
 728			/* If argument name is omitted, set "argN" */
 729			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 730			parg->name = kstrdup(buf, GFP_KERNEL);
 731		}
 732
 733		if (!parg->name) {
 734			pr_info("Failed to allocate argument[%d] name.\n", i);
 735			ret = -ENOMEM;
 736			goto error;
 737		}
 738
 739		if (!is_good_name(parg->name)) {
 740			pr_info("Invalid argument[%d] name: %s\n",
 741				i, parg->name);
 742			ret = -EINVAL;
 743			goto error;
 744		}
 745
 746		if (traceprobe_conflict_field_name(parg->name,
 747							tk->tp.args, i)) {
 748			pr_info("Argument[%d] name '%s' conflicts with "
 749				"another field.\n", i, argv[i]);
 750			ret = -EINVAL;
 751			goto error;
 752		}
 753
 754		/* Parse fetch argument */
 755		ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 756						is_return, true);
 
 757		if (ret) {
 758			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 759			goto error;
 760		}
 761	}
 762
 763	ret = register_trace_kprobe(tk);
 764	if (ret)
 765		goto error;
 766	return 0;
 767
 768error:
 769	free_trace_kprobe(tk);
 770	return ret;
 771}
 772
 773static int release_all_trace_kprobes(void)
 774{
 775	struct trace_kprobe *tk;
 776	int ret = 0;
 777
 778	mutex_lock(&probe_lock);
 779	/* Ensure no probe is in use. */
 780	list_for_each_entry(tk, &probe_list, list)
 781		if (trace_probe_is_enabled(&tk->tp)) {
 782			ret = -EBUSY;
 783			goto end;
 784		}
 785	/* TODO: Use batch unregistration */
 786	while (!list_empty(&probe_list)) {
 787		tk = list_entry(probe_list.next, struct trace_kprobe, list);
 788		ret = unregister_trace_kprobe(tk);
 789		if (ret)
 790			goto end;
 791		free_trace_kprobe(tk);
 792	}
 793
 794end:
 795	mutex_unlock(&probe_lock);
 796
 797	return ret;
 798}
 799
 800/* Probes listing interfaces */
 801static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 802{
 803	mutex_lock(&probe_lock);
 804	return seq_list_start(&probe_list, *pos);
 805}
 806
 807static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 808{
 809	return seq_list_next(v, &probe_list, pos);
 810}
 811
 812static void probes_seq_stop(struct seq_file *m, void *v)
 813{
 814	mutex_unlock(&probe_lock);
 815}
 816
 817static int probes_seq_show(struct seq_file *m, void *v)
 818{
 819	struct trace_kprobe *tk = v;
 820	int i;
 821
 822	seq_printf(m, "%c", trace_kprobe_is_return(tk) ? 'r' : 'p');
 823	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 824			ftrace_event_name(&tk->tp.call));
 825
 826	if (!tk->symbol)
 827		seq_printf(m, " 0x%p", tk->rp.kp.addr);
 828	else if (tk->rp.kp.offset)
 829		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 830			   tk->rp.kp.offset);
 831	else
 832		seq_printf(m, " %s", trace_kprobe_symbol(tk));
 833
 834	for (i = 0; i < tk->tp.nr_args; i++)
 835		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 836	seq_printf(m, "\n");
 837
 838	return 0;
 839}
 840
 841static const struct seq_operations probes_seq_op = {
 842	.start  = probes_seq_start,
 843	.next   = probes_seq_next,
 844	.stop   = probes_seq_stop,
 845	.show   = probes_seq_show
 846};
 847
 848static int probes_open(struct inode *inode, struct file *file)
 849{
 850	int ret;
 851
 852	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 853		ret = release_all_trace_kprobes();
 854		if (ret < 0)
 855			return ret;
 856	}
 857
 858	return seq_open(file, &probes_seq_op);
 859}
 860
 861static ssize_t probes_write(struct file *file, const char __user *buffer,
 862			    size_t count, loff_t *ppos)
 863{
 864	return traceprobe_probes_write(file, buffer, count, ppos,
 865			create_trace_kprobe);
 866}
 867
 868static const struct file_operations kprobe_events_ops = {
 869	.owner          = THIS_MODULE,
 870	.open           = probes_open,
 871	.read           = seq_read,
 872	.llseek         = seq_lseek,
 873	.release        = seq_release,
 874	.write		= probes_write,
 875};
 876
 877/* Probes profiling interfaces */
 878static int probes_profile_seq_show(struct seq_file *m, void *v)
 879{
 880	struct trace_kprobe *tk = v;
 881
 882	seq_printf(m, "  %-44s %15lu %15lu\n",
 883		   ftrace_event_name(&tk->tp.call), tk->nhit,
 
 884		   tk->rp.kp.nmissed);
 885
 886	return 0;
 887}
 888
 889static const struct seq_operations profile_seq_op = {
 890	.start  = probes_seq_start,
 891	.next   = probes_seq_next,
 892	.stop   = probes_seq_stop,
 893	.show   = probes_profile_seq_show
 894};
 895
 896static int profile_open(struct inode *inode, struct file *file)
 897{
 898	return seq_open(file, &profile_seq_op);
 899}
 900
 901static const struct file_operations kprobe_profile_ops = {
 902	.owner          = THIS_MODULE,
 903	.open           = profile_open,
 904	.read           = seq_read,
 905	.llseek         = seq_lseek,
 906	.release        = seq_release,
 907};
 908
 909/* Kprobe handler */
 910static __kprobes void
 911__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 912		    struct ftrace_event_file *ftrace_file)
 913{
 914	struct kprobe_trace_entry_head *entry;
 915	struct ring_buffer_event *event;
 916	struct ring_buffer *buffer;
 917	int size, dsize, pc;
 918	unsigned long irq_flags;
 919	struct ftrace_event_call *call = &tk->tp.call;
 920
 921	WARN_ON(call != ftrace_file->event_call);
 922
 923	if (ftrace_trigger_soft_disabled(ftrace_file))
 924		return;
 925
 926	local_save_flags(irq_flags);
 927	pc = preempt_count();
 928
 929	dsize = __get_data_size(&tk->tp, regs);
 930	size = sizeof(*entry) + tk->tp.size + dsize;
 931
 932	event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 933						call->event.type,
 934						size, irq_flags, pc);
 935	if (!event)
 936		return;
 937
 938	entry = ring_buffer_event_data(event);
 939	entry->ip = (unsigned long)tk->rp.kp.addr;
 940	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 941
 942	event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
 943					 entry, irq_flags, pc, regs);
 944}
 945
 946static __kprobes void
 947kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 948{
 949	struct event_file_link *link;
 950
 951	list_for_each_entry_rcu(link, &tk->tp.files, list)
 952		__kprobe_trace_func(tk, regs, link->file);
 953}
 
 954
 955/* Kretprobe handler */
 956static __kprobes void
 957__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 958		       struct pt_regs *regs,
 959		       struct ftrace_event_file *ftrace_file)
 960{
 961	struct kretprobe_trace_entry_head *entry;
 962	struct ring_buffer_event *event;
 963	struct ring_buffer *buffer;
 964	int size, pc, dsize;
 965	unsigned long irq_flags;
 966	struct ftrace_event_call *call = &tk->tp.call;
 967
 968	WARN_ON(call != ftrace_file->event_call);
 969
 970	if (ftrace_trigger_soft_disabled(ftrace_file))
 971		return;
 972
 973	local_save_flags(irq_flags);
 974	pc = preempt_count();
 975
 976	dsize = __get_data_size(&tk->tp, regs);
 977	size = sizeof(*entry) + tk->tp.size + dsize;
 978
 979	event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 980						call->event.type,
 981						size, irq_flags, pc);
 982	if (!event)
 983		return;
 984
 985	entry = ring_buffer_event_data(event);
 986	entry->func = (unsigned long)tk->rp.kp.addr;
 987	entry->ret_ip = (unsigned long)ri->ret_addr;
 988	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 989
 990	event_trigger_unlock_commit_regs(ftrace_file, buffer, event,
 991					 entry, irq_flags, pc, regs);
 992}
 993
 994static __kprobes void
 995kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 996		     struct pt_regs *regs)
 997{
 998	struct event_file_link *link;
 999
1000	list_for_each_entry_rcu(link, &tk->tp.files, list)
1001		__kretprobe_trace_func(tk, ri, regs, link->file);
1002}
 
1003
1004/* Event entry printers */
1005static enum print_line_t
1006print_kprobe_event(struct trace_iterator *iter, int flags,
1007		   struct trace_event *event)
1008{
1009	struct kprobe_trace_entry_head *field;
1010	struct trace_seq *s = &iter->seq;
1011	struct trace_probe *tp;
1012	u8 *data;
1013	int i;
1014
1015	field = (struct kprobe_trace_entry_head *)iter->ent;
1016	tp = container_of(event, struct trace_probe, call.event);
1017
1018	if (!trace_seq_printf(s, "%s: (", ftrace_event_name(&tp->call)))
1019		goto partial;
1020
1021	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1022		goto partial;
1023
1024	if (!trace_seq_puts(s, ")"))
1025		goto partial;
1026
1027	data = (u8 *)&field[1];
1028	for (i = 0; i < tp->nr_args; i++)
1029		if (!tp->args[i].type->print(s, tp->args[i].name,
1030					     data + tp->args[i].offset, field))
1031			goto partial;
1032
1033	if (!trace_seq_puts(s, "\n"))
1034		goto partial;
1035
1036	return TRACE_TYPE_HANDLED;
1037partial:
1038	return TRACE_TYPE_PARTIAL_LINE;
1039}
1040
1041static enum print_line_t
1042print_kretprobe_event(struct trace_iterator *iter, int flags,
1043		      struct trace_event *event)
1044{
1045	struct kretprobe_trace_entry_head *field;
1046	struct trace_seq *s = &iter->seq;
1047	struct trace_probe *tp;
1048	u8 *data;
1049	int i;
1050
1051	field = (struct kretprobe_trace_entry_head *)iter->ent;
1052	tp = container_of(event, struct trace_probe, call.event);
1053
1054	if (!trace_seq_printf(s, "%s: (", ftrace_event_name(&tp->call)))
1055		goto partial;
1056
1057	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1058		goto partial;
1059
1060	if (!trace_seq_puts(s, " <- "))
1061		goto partial;
1062
1063	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1064		goto partial;
1065
1066	if (!trace_seq_puts(s, ")"))
1067		goto partial;
1068
1069	data = (u8 *)&field[1];
1070	for (i = 0; i < tp->nr_args; i++)
1071		if (!tp->args[i].type->print(s, tp->args[i].name,
1072					     data + tp->args[i].offset, field))
1073			goto partial;
1074
1075	if (!trace_seq_puts(s, "\n"))
1076		goto partial;
1077
1078	return TRACE_TYPE_HANDLED;
1079partial:
1080	return TRACE_TYPE_PARTIAL_LINE;
1081}
1082
1083
1084static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1085{
1086	int ret, i;
1087	struct kprobe_trace_entry_head field;
1088	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1089
1090	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1091	/* Set argument names as fields */
1092	for (i = 0; i < tk->tp.nr_args; i++) {
1093		struct probe_arg *parg = &tk->tp.args[i];
1094
1095		ret = trace_define_field(event_call, parg->type->fmttype,
1096					 parg->name,
1097					 sizeof(field) + parg->offset,
1098					 parg->type->size,
1099					 parg->type->is_signed,
1100					 FILTER_OTHER);
1101		if (ret)
1102			return ret;
1103	}
1104	return 0;
1105}
1106
1107static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1108{
1109	int ret, i;
1110	struct kretprobe_trace_entry_head field;
1111	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1112
1113	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1114	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1115	/* Set argument names as fields */
1116	for (i = 0; i < tk->tp.nr_args; i++) {
1117		struct probe_arg *parg = &tk->tp.args[i];
1118
1119		ret = trace_define_field(event_call, parg->type->fmttype,
1120					 parg->name,
1121					 sizeof(field) + parg->offset,
1122					 parg->type->size,
1123					 parg->type->is_signed,
1124					 FILTER_OTHER);
1125		if (ret)
1126			return ret;
1127	}
1128	return 0;
1129}
1130
1131#ifdef CONFIG_PERF_EVENTS
1132
1133/* Kprobe profile handler */
1134static __kprobes void
1135kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1136{
1137	struct ftrace_event_call *call = &tk->tp.call;
 
1138	struct kprobe_trace_entry_head *entry;
1139	struct hlist_head *head;
1140	int size, __size, dsize;
1141	int rctx;
1142
 
 
 
1143	head = this_cpu_ptr(call->perf_events);
1144	if (hlist_empty(head))
1145		return;
1146
1147	dsize = __get_data_size(&tk->tp, regs);
1148	__size = sizeof(*entry) + tk->tp.size + dsize;
1149	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1150	size -= sizeof(u32);
1151
1152	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1153	if (!entry)
1154		return;
1155
1156	entry->ip = (unsigned long)tk->rp.kp.addr;
1157	memset(&entry[1], 0, dsize);
1158	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1159	perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 
1160}
 
1161
1162/* Kretprobe profile handler */
1163static __kprobes void
1164kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1165		    struct pt_regs *regs)
1166{
1167	struct ftrace_event_call *call = &tk->tp.call;
 
1168	struct kretprobe_trace_entry_head *entry;
1169	struct hlist_head *head;
1170	int size, __size, dsize;
1171	int rctx;
1172
 
 
 
1173	head = this_cpu_ptr(call->perf_events);
1174	if (hlist_empty(head))
1175		return;
1176
1177	dsize = __get_data_size(&tk->tp, regs);
1178	__size = sizeof(*entry) + tk->tp.size + dsize;
1179	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1180	size -= sizeof(u32);
1181
1182	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1183	if (!entry)
1184		return;
1185
1186	entry->func = (unsigned long)tk->rp.kp.addr;
1187	entry->ret_ip = (unsigned long)ri->ret_addr;
1188	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1189	perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 
1190}
 
1191#endif	/* CONFIG_PERF_EVENTS */
1192
1193/*
1194 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1195 *
1196 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1197 * lockless, but we can't race with this __init function.
1198 */
1199static __kprobes
1200int kprobe_register(struct ftrace_event_call *event,
1201		    enum trace_reg type, void *data)
1202{
1203	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1204	struct ftrace_event_file *file = data;
1205
1206	switch (type) {
1207	case TRACE_REG_REGISTER:
1208		return enable_trace_kprobe(tk, file);
1209	case TRACE_REG_UNREGISTER:
1210		return disable_trace_kprobe(tk, file);
1211
1212#ifdef CONFIG_PERF_EVENTS
1213	case TRACE_REG_PERF_REGISTER:
1214		return enable_trace_kprobe(tk, NULL);
1215	case TRACE_REG_PERF_UNREGISTER:
1216		return disable_trace_kprobe(tk, NULL);
1217	case TRACE_REG_PERF_OPEN:
1218	case TRACE_REG_PERF_CLOSE:
1219	case TRACE_REG_PERF_ADD:
1220	case TRACE_REG_PERF_DEL:
1221		return 0;
1222#endif
1223	}
1224	return 0;
1225}
1226
1227static __kprobes
1228int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1229{
1230	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1231
1232	tk->nhit++;
1233
1234	if (tk->tp.flags & TP_FLAG_TRACE)
1235		kprobe_trace_func(tk, regs);
1236#ifdef CONFIG_PERF_EVENTS
1237	if (tk->tp.flags & TP_FLAG_PROFILE)
1238		kprobe_perf_func(tk, regs);
1239#endif
1240	return 0;	/* We don't tweek kernel, so just return 0 */
1241}
 
1242
1243static __kprobes
1244int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1245{
1246	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1247
1248	tk->nhit++;
1249
1250	if (tk->tp.flags & TP_FLAG_TRACE)
1251		kretprobe_trace_func(tk, ri, regs);
1252#ifdef CONFIG_PERF_EVENTS
1253	if (tk->tp.flags & TP_FLAG_PROFILE)
1254		kretprobe_perf_func(tk, ri, regs);
1255#endif
1256	return 0;	/* We don't tweek kernel, so just return 0 */
1257}
 
1258
1259static struct trace_event_functions kretprobe_funcs = {
1260	.trace		= print_kretprobe_event
1261};
1262
1263static struct trace_event_functions kprobe_funcs = {
1264	.trace		= print_kprobe_event
1265};
1266
1267static int register_kprobe_event(struct trace_kprobe *tk)
1268{
1269	struct ftrace_event_call *call = &tk->tp.call;
1270	int ret;
1271
1272	/* Initialize ftrace_event_call */
1273	INIT_LIST_HEAD(&call->class->fields);
1274	if (trace_kprobe_is_return(tk)) {
1275		call->event.funcs = &kretprobe_funcs;
1276		call->class->define_fields = kretprobe_event_define_fields;
1277	} else {
1278		call->event.funcs = &kprobe_funcs;
1279		call->class->define_fields = kprobe_event_define_fields;
1280	}
1281	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1282		return -ENOMEM;
1283	ret = register_ftrace_event(&call->event);
1284	if (!ret) {
1285		kfree(call->print_fmt);
1286		return -ENODEV;
1287	}
1288	call->flags = 0;
1289	call->class->reg = kprobe_register;
1290	call->data = tk;
1291	ret = trace_add_event_call(call);
1292	if (ret) {
1293		pr_info("Failed to register kprobe event: %s\n",
1294			ftrace_event_name(call));
1295		kfree(call->print_fmt);
1296		unregister_ftrace_event(&call->event);
1297	}
1298	return ret;
1299}
1300
1301static int unregister_kprobe_event(struct trace_kprobe *tk)
1302{
1303	int ret;
1304
1305	/* tp->event is unregistered in trace_remove_event_call() */
1306	ret = trace_remove_event_call(&tk->tp.call);
1307	if (!ret)
1308		kfree(tk->tp.call.print_fmt);
1309	return ret;
1310}
1311
1312/* Make a debugfs interface for controlling probe points */
1313static __init int init_kprobe_trace(void)
1314{
1315	struct dentry *d_tracer;
1316	struct dentry *entry;
1317
1318	if (register_module_notifier(&trace_kprobe_module_nb))
1319		return -EINVAL;
1320
1321	d_tracer = tracing_init_dentry();
1322	if (!d_tracer)
1323		return 0;
1324
1325	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1326				    NULL, &kprobe_events_ops);
1327
1328	/* Event list interface */
1329	if (!entry)
1330		pr_warning("Could not create debugfs "
1331			   "'kprobe_events' entry\n");
1332
1333	/* Profile interface */
1334	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1335				    NULL, &kprobe_profile_ops);
1336
1337	if (!entry)
1338		pr_warning("Could not create debugfs "
1339			   "'kprobe_profile' entry\n");
1340	return 0;
1341}
1342fs_initcall(init_kprobe_trace);
1343
1344
1345#ifdef CONFIG_FTRACE_STARTUP_TEST
1346
1347/*
1348 * The "__used" keeps gcc from removing the function symbol
1349 * from the kallsyms table.
 
1350 */
1351static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1352					       int a4, int a5, int a6)
1353{
1354	return a1 + a2 + a3 + a4 + a5 + a6;
1355}
1356
1357static struct ftrace_event_file *
1358find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1359{
1360	struct ftrace_event_file *file;
1361
1362	list_for_each_entry(file, &tr->events, list)
1363		if (file->event_call == &tk->tp.call)
1364			return file;
1365
1366	return NULL;
1367}
1368
1369/*
1370 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1371 * stage, we can do this lockless.
1372 */
1373static __init int kprobe_trace_self_tests_init(void)
1374{
1375	int ret, warn = 0;
1376	int (*target)(int, int, int, int, int, int);
1377	struct trace_kprobe *tk;
1378	struct ftrace_event_file *file;
 
 
 
1379
1380	target = kprobe_trace_selftest_target;
1381
1382	pr_info("Testing kprobe tracing: ");
1383
1384	ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1385				  "$stack $stack0 +0($stack)",
1386				  create_trace_kprobe);
1387	if (WARN_ON_ONCE(ret)) {
1388		pr_warn("error on probing function entry.\n");
1389		warn++;
1390	} else {
1391		/* Enable trace point */
1392		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1393		if (WARN_ON_ONCE(tk == NULL)) {
1394			pr_warn("error on getting new probe.\n");
1395			warn++;
1396		} else {
1397			file = find_trace_probe_file(tk, top_trace_array());
1398			if (WARN_ON_ONCE(file == NULL)) {
1399				pr_warn("error on getting probe file.\n");
1400				warn++;
1401			} else
1402				enable_trace_kprobe(tk, file);
1403		}
1404	}
1405
1406	ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1407				  "$retval", create_trace_kprobe);
1408	if (WARN_ON_ONCE(ret)) {
1409		pr_warn("error on probing function return.\n");
1410		warn++;
1411	} else {
1412		/* Enable trace point */
1413		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1414		if (WARN_ON_ONCE(tk == NULL)) {
1415			pr_warn("error on getting 2nd new probe.\n");
1416			warn++;
1417		} else {
1418			file = find_trace_probe_file(tk, top_trace_array());
1419			if (WARN_ON_ONCE(file == NULL)) {
1420				pr_warn("error on getting probe file.\n");
1421				warn++;
1422			} else
1423				enable_trace_kprobe(tk, file);
1424		}
1425	}
1426
1427	if (warn)
1428		goto end;
1429
1430	ret = target(1, 2, 3, 4, 5, 6);
1431
 
 
 
 
 
 
 
 
1432	/* Disable trace points before removing it */
1433	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1434	if (WARN_ON_ONCE(tk == NULL)) {
1435		pr_warn("error on getting test probe.\n");
1436		warn++;
1437	} else {
 
 
 
 
 
1438		file = find_trace_probe_file(tk, top_trace_array());
1439		if (WARN_ON_ONCE(file == NULL)) {
1440			pr_warn("error on getting probe file.\n");
1441			warn++;
1442		} else
1443			disable_trace_kprobe(tk, file);
1444	}
1445
1446	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1447	if (WARN_ON_ONCE(tk == NULL)) {
1448		pr_warn("error on getting 2nd test probe.\n");
1449		warn++;
1450	} else {
 
 
 
 
 
1451		file = find_trace_probe_file(tk, top_trace_array());
1452		if (WARN_ON_ONCE(file == NULL)) {
1453			pr_warn("error on getting probe file.\n");
1454			warn++;
1455		} else
1456			disable_trace_kprobe(tk, file);
1457	}
1458
1459	ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1460	if (WARN_ON_ONCE(ret)) {
1461		pr_warn("error on deleting a probe.\n");
1462		warn++;
1463	}
1464
1465	ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1466	if (WARN_ON_ONCE(ret)) {
1467		pr_warn("error on deleting a probe.\n");
1468		warn++;
1469	}
1470
1471end:
1472	release_all_trace_kprobes();
1473	if (warn)
1474		pr_cont("NG: Some tests are failed. Please check them.\n");
1475	else
1476		pr_cont("OK\n");
1477	return 0;
1478}
1479
1480late_initcall(kprobe_trace_self_tests_init);
1481
1482#endif