Linux Audio

Check our new training course

Loading...
v4.6
 
   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 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 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}									\
 141NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 142
 143DEFINE_BASIC_FETCH_FUNCS(stack)
 144/* No string on the stack entry */
 145#define fetch_stack_string	NULL
 146#define fetch_stack_string_size	NULL
 147
 148#define DEFINE_FETCH_memory(type)					\
 149static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
 150					  void *addr, void *dest)	\
 151{									\
 152	type retval;							\
 153	if (probe_kernel_address(addr, retval))				\
 154		*(type *)dest = 0;					\
 155	else								\
 156		*(type *)dest = retval;					\
 157}									\
 158NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 159
 160DEFINE_BASIC_FETCH_FUNCS(memory)
 161/*
 162 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 163 * length and relative data location.
 164 */
 165static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 166					    void *addr, void *dest)
 167{
 168	int maxlen = get_rloc_len(*(u32 *)dest);
 169	u8 *dst = get_rloc_data(dest);
 170	long ret;
 171
 172	if (!maxlen)
 173		return;
 174
 175	/*
 176	 * Try to get string again, since the string can be changed while
 177	 * probing.
 178	 */
 179	ret = strncpy_from_unsafe(dst, addr, maxlen);
 180
 181	if (ret < 0) {	/* Failed to fetch string */
 182		dst[0] = '\0';
 183		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 
 
 184	} else {
 185		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 186	}
 
 187}
 188NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 189
 190/* Return the length of string -- including null terminal byte */
 191static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 192						 void *addr, void *dest)
 193{
 194	mm_segment_t old_fs;
 195	int ret, len = 0;
 196	u8 c;
 197
 198	old_fs = get_fs();
 199	set_fs(KERNEL_DS);
 200	pagefault_disable();
 201
 202	do {
 203		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 204		len++;
 205	} while (c && ret == 0 && len < MAX_STRING_SIZE);
 206
 207	pagefault_enable();
 208	set_fs(old_fs);
 
 209
 210	if (ret < 0)	/* Failed to check the length */
 211		*(u32 *)dest = 0;
 212	else
 213		*(u32 *)dest = len;
 214}
 215NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 216
 217#define DEFINE_FETCH_symbol(type)					\
 218void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 219{									\
 220	struct symbol_cache *sc = data;					\
 221	if (sc->addr)							\
 222		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
 223	else								\
 224		*(type *)dest = 0;					\
 225}									\
 226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 227
 228DEFINE_BASIC_FETCH_FUNCS(symbol)
 229DEFINE_FETCH_symbol(string)
 230DEFINE_FETCH_symbol(string_size)
 231
 232/* kprobes don't support file_offset fetch methods */
 233#define fetch_file_offset_u8		NULL
 234#define fetch_file_offset_u16		NULL
 235#define fetch_file_offset_u32		NULL
 236#define fetch_file_offset_u64		NULL
 237#define fetch_file_offset_string	NULL
 238#define fetch_file_offset_string_size	NULL
 239
 240/* Fetch type information table */
 241static const struct fetch_type kprobes_fetch_type_table[] = {
 242	/* Special types */
 243	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 244					sizeof(u32), 1, "__data_loc char[]"),
 245	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 246					string_size, sizeof(u32), 0, "u32"),
 247	/* Basic types */
 248	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 249	ASSIGN_FETCH_TYPE(u16, u16, 0),
 250	ASSIGN_FETCH_TYPE(u32, u32, 0),
 251	ASSIGN_FETCH_TYPE(u64, u64, 0),
 252	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 253	ASSIGN_FETCH_TYPE(s16, u16, 1),
 254	ASSIGN_FETCH_TYPE(s32, u32, 1),
 255	ASSIGN_FETCH_TYPE(s64, u64, 1),
 256
 257	ASSIGN_FETCH_TYPE_END
 258};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 259
 260/*
 261 * Allocate new trace_probe and initialize it (including kprobes).
 262 */
 263static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 264					     const char *event,
 265					     void *addr,
 266					     const char *symbol,
 267					     unsigned long offs,
 
 268					     int nargs, bool is_return)
 269{
 270	struct trace_kprobe *tk;
 271	int ret = -ENOMEM;
 272
 273	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 274	if (!tk)
 275		return ERR_PTR(ret);
 276
 277	tk->nhit = alloc_percpu(unsigned long);
 278	if (!tk->nhit)
 279		goto error;
 280
 281	if (symbol) {
 282		tk->symbol = kstrdup(symbol, GFP_KERNEL);
 283		if (!tk->symbol)
 284			goto error;
 285		tk->rp.kp.symbol_name = tk->symbol;
 286		tk->rp.kp.offset = offs;
 287	} else
 288		tk->rp.kp.addr = addr;
 289
 290	if (is_return)
 291		tk->rp.handler = kretprobe_dispatcher;
 292	else
 293		tk->rp.kp.pre_handler = kprobe_dispatcher;
 294
 295	if (!event || !is_good_name(event)) {
 296		ret = -EINVAL;
 297		goto error;
 298	}
 299
 300	tk->tp.call.class = &tk->tp.class;
 301	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 302	if (!tk->tp.call.name)
 303		goto error;
 304
 305	if (!group || !is_good_name(group)) {
 306		ret = -EINVAL;
 307		goto error;
 308	}
 309
 310	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 311	if (!tk->tp.class.system)
 312		goto error;
 313
 314	INIT_LIST_HEAD(&tk->list);
 315	INIT_LIST_HEAD(&tk->tp.files);
 316	return tk;
 317error:
 318	kfree(tk->tp.call.name);
 319	kfree(tk->symbol);
 320	free_percpu(tk->nhit);
 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	free_percpu(tk->nhit);
 336	kfree(tk);
 337}
 338
 339static struct trace_kprobe *find_trace_kprobe(const char *event,
 340					      const char *group)
 341{
 
 342	struct trace_kprobe *tk;
 343
 344	list_for_each_entry(tk, &probe_list, list)
 345		if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 346		    strcmp(tk->tp.call.class->system, group) == 0)
 347			return tk;
 348	return NULL;
 
 
 
 
 349}
 350
 351/*
 352 * Enable trace_probe
 353 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 354 */
 355static int
 356enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 357{
 
 
 
 358	int ret = 0;
 359
 360	if (file) {
 361		struct event_file_link *link;
 
 
 362
 363		link = kmalloc(sizeof(*link), GFP_KERNEL);
 364		if (!link) {
 365			ret = -ENOMEM;
 366			goto out;
 367		}
 
 
 368
 369		link->file = file;
 370		list_add_tail_rcu(&link->list, &tk->tp.files);
 371
 372		tk->tp.flags |= TP_FLAG_TRACE;
 373	} else
 374		tk->tp.flags |= TP_FLAG_PROFILE;
 
 
 
 
 
 
 375
 376	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 377		if (trace_kprobe_is_return(tk))
 378			ret = enable_kretprobe(&tk->rp);
 
 
 
 379		else
 380			ret = enable_kprobe(&tk->rp.kp);
 381	}
 382 out:
 383	return ret;
 384}
 385
 386/*
 387 * Disable trace_probe
 388 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 389 */
 390static int
 391disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 392{
 393	struct event_file_link *link = NULL;
 394	int wait = 0;
 395	int ret = 0;
 396
 397	if (file) {
 398		link = find_event_file_link(&tk->tp, file);
 399		if (!link) {
 400			ret = -EINVAL;
 401			goto out;
 402		}
 403
 404		list_del_rcu(&link->list);
 405		wait = 1;
 406		if (!list_empty(&tk->tp.files))
 
 407			goto out;
 408
 409		tk->tp.flags &= ~TP_FLAG_TRACE;
 410	} else
 411		tk->tp.flags &= ~TP_FLAG_PROFILE;
 
 
 
 412
 413	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 414		if (trace_kprobe_is_return(tk))
 415			disable_kretprobe(&tk->rp);
 416		else
 417			disable_kprobe(&tk->rp.kp);
 418		wait = 1;
 419	}
 420 out:
 421	if (wait) {
 422		/*
 423		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
 424		 * to ensure disabled (all running handlers are finished).
 425		 * This is not only for kfree(), but also the caller,
 426		 * trace_remove_event_call() supposes it for releasing
 427		 * event_call related objects, which will be accessed in
 428		 * the kprobe_trace_func/kretprobe_trace_func.
 429		 */
 430		synchronize_sched();
 431		kfree(link);	/* Ignored if link == NULL */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 432	}
 433
 434	return ret;
 435}
 
 
 
 436
 437/* Internal register function - just handle k*probes and flags */
 438static int __register_trace_kprobe(struct trace_kprobe *tk)
 439{
 440	int i, ret;
 441
 442	if (trace_probe_is_registered(&tk->tp))
 
 
 
 
 443		return -EINVAL;
 444
 445	for (i = 0; i < tk->tp.nr_args; i++)
 446		traceprobe_update_arg(&tk->tp.args[i]);
 
 
 
 
 
 
 
 
 
 447
 448	/* Set/clear disabled flag according to tp->flag */
 449	if (trace_probe_is_enabled(&tk->tp))
 450		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 451	else
 452		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 453
 454	if (trace_kprobe_is_return(tk))
 455		ret = register_kretprobe(&tk->rp);
 456	else
 457		ret = register_kprobe(&tk->rp.kp);
 458
 459	if (ret == 0)
 460		tk->tp.flags |= TP_FLAG_REGISTERED;
 461	else {
 462		pr_warn("Could not insert probe at %s+%lu: %d\n",
 463			trace_kprobe_symbol(tk), trace_kprobe_offset(tk), ret);
 464		if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 465			pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 466			ret = 0;
 467		} else if (ret == -EILSEQ) {
 468			pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 469				tk->rp.kp.addr);
 470			ret = -EINVAL;
 471		}
 472	}
 473
 474	return ret;
 475}
 476
 477/* Internal unregister function - just handle k*probes and flags */
 478static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 479{
 480	if (trace_probe_is_registered(&tk->tp)) {
 481		if (trace_kprobe_is_return(tk))
 482			unregister_kretprobe(&tk->rp);
 483		else
 484			unregister_kprobe(&tk->rp.kp);
 485		tk->tp.flags &= ~TP_FLAG_REGISTERED;
 486		/* Cleanup kprobe for reuse */
 
 487		if (tk->rp.kp.symbol_name)
 488			tk->rp.kp.addr = NULL;
 489	}
 490}
 491
 492/* Unregister a trace_probe and probe_event: call with locking probe_lock */
 493static int unregister_trace_kprobe(struct trace_kprobe *tk)
 494{
 
 
 
 
 495	/* Enabled event can not be unregistered */
 496	if (trace_probe_is_enabled(&tk->tp))
 497		return -EBUSY;
 498
 499	/* Will fail if probe is being used by ftrace or perf */
 500	if (unregister_kprobe_event(tk))
 501		return -EBUSY;
 502
 
 503	__unregister_trace_kprobe(tk);
 504	list_del(&tk->list);
 
 505
 506	return 0;
 507}
 508
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 509/* Register a trace_probe and probe_event */
 510static int register_trace_kprobe(struct trace_kprobe *tk)
 511{
 512	struct trace_kprobe *old_tk;
 513	int ret;
 514
 515	mutex_lock(&probe_lock);
 516
 517	/* Delete old (same name) event if exist */
 518	old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 519			tk->tp.call.class->system);
 520	if (old_tk) {
 521		ret = unregister_trace_kprobe(old_tk);
 522		if (ret < 0)
 523			goto end;
 524		free_trace_kprobe(old_tk);
 
 
 
 
 525	}
 526
 527	/* Register new event */
 528	ret = register_kprobe_event(tk);
 529	if (ret) {
 530		pr_warn("Failed to register probe event(%d)\n", ret);
 531		goto end;
 532	}
 533
 534	/* Register k*probe */
 535	ret = __register_trace_kprobe(tk);
 
 
 
 
 
 536	if (ret < 0)
 537		unregister_kprobe_event(tk);
 538	else
 539		list_add_tail(&tk->list, &probe_list);
 540
 541end:
 542	mutex_unlock(&probe_lock);
 543	return ret;
 544}
 545
 546/* Module notifier call back, checking event on the module */
 547static int trace_kprobe_module_callback(struct notifier_block *nb,
 548				       unsigned long val, void *data)
 549{
 550	struct module *mod = data;
 
 551	struct trace_kprobe *tk;
 552	int ret;
 553
 554	if (val != MODULE_STATE_COMING)
 555		return NOTIFY_DONE;
 556
 557	/* Update probes on coming module */
 558	mutex_lock(&probe_lock);
 559	list_for_each_entry(tk, &probe_list, list) {
 560		if (trace_kprobe_within_module(tk, mod)) {
 561			/* Don't need to check busy - this should have gone. */
 562			__unregister_trace_kprobe(tk);
 563			ret = __register_trace_kprobe(tk);
 564			if (ret)
 565				pr_warn("Failed to re-register probe %s on %s: %d\n",
 566					trace_event_name(&tk->tp.call),
 567					mod->name, ret);
 568		}
 569	}
 570	mutex_unlock(&probe_lock);
 571
 572	return NOTIFY_DONE;
 573}
 574
 575static struct notifier_block trace_kprobe_module_nb = {
 576	.notifier_call = trace_kprobe_module_callback,
 577	.priority = 1	/* Invoked after kprobe module callback */
 578};
 579
 580static int create_trace_kprobe(int argc, char **argv)
 
 
 
 
 
 
 
 
 581{
 582	/*
 583	 * Argument syntax:
 584	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 585	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 
 
 586	 * Fetch args:
 587	 *  $retval	: fetch return value
 588	 *  $stack	: fetch stack address
 589	 *  $stackN	: fetch Nth of stack (N:0-)
 
 590	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
 591	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 592	 *  %REG	: fetch register REG
 593	 * Dereferencing memory fetch:
 594	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 595	 * Alias name of args:
 596	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 597	 * Type of args:
 598	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 599	 */
 600	struct trace_kprobe *tk;
 601	int i, ret = 0;
 602	bool is_return = false, is_delete = false;
 603	char *symbol = NULL, *event = NULL, *group = NULL;
 604	char *arg;
 605	unsigned long offset = 0;
 
 606	void *addr = NULL;
 607	char buf[MAX_EVENT_NAME_LEN];
 
 608
 609	/* argc must be >= 1 */
 610	if (argv[0][0] == 'p')
 611		is_return = false;
 612	else if (argv[0][0] == 'r')
 613		is_return = true;
 614	else if (argv[0][0] == '-')
 615		is_delete = true;
 616	else {
 617		pr_info("Probe definition must be started with 'p', 'r' or"
 618			" '-'.\n");
 619		return -EINVAL;
 620	}
 621
 622	if (argv[0][1] == ':') {
 623		event = &argv[0][2];
 624		if (strchr(event, '/')) {
 625			group = event;
 626			event = strchr(group, '/') + 1;
 627			event[-1] = '\0';
 628			if (strlen(group) == 0) {
 629				pr_info("Group name is not specified\n");
 630				return -EINVAL;
 631			}
 
 
 632		}
 633		if (strlen(event) == 0) {
 634			pr_info("Event name is not specified\n");
 635			return -EINVAL;
 
 
 
 
 636		}
 637	}
 638	if (!group)
 639		group = KPROBE_EVENT_SYSTEM;
 640
 641	if (is_delete) {
 642		if (!event) {
 643			pr_info("Delete command needs an event name.\n");
 644			return -EINVAL;
 645		}
 646		mutex_lock(&probe_lock);
 647		tk = find_trace_kprobe(event, group);
 648		if (!tk) {
 649			mutex_unlock(&probe_lock);
 650			pr_info("Event %s/%s doesn't exist.\n", group, event);
 651			return -ENOENT;
 652		}
 653		/* delete an event */
 654		ret = unregister_trace_kprobe(tk);
 655		if (ret == 0)
 656			free_trace_kprobe(tk);
 657		mutex_unlock(&probe_lock);
 658		return ret;
 659	}
 660
 661	if (argc < 2) {
 662		pr_info("Probe point is not specified.\n");
 663		return -EINVAL;
 664	}
 665	if (isdigit(argv[1][0])) {
 666		if (is_return) {
 667			pr_info("Return probe point must be a symbol.\n");
 668			return -EINVAL;
 669		}
 670		/* an address specified */
 671		ret = kstrtoul(&argv[1][0], 0, (unsigned long *)&addr);
 672		if (ret) {
 673			pr_info("Failed to parse address.\n");
 674			return ret;
 675		}
 676	} else {
 677		/* a symbol specified */
 678		symbol = argv[1];
 
 
 679		/* TODO: support .init module functions */
 680		ret = traceprobe_split_symbol_offset(symbol, &offset);
 681		if (ret) {
 682			pr_info("Failed to parse symbol.\n");
 683			return ret;
 684		}
 685		if (offset && is_return) {
 686			pr_info("Return probe must be used without offset.\n");
 687			return -EINVAL;
 
 
 688		}
 689	}
 690	argc -= 2; argv += 2;
 691
 692	/* setup a probe */
 693	if (!event) {
 
 
 
 
 
 694		/* Make a new event name */
 695		if (symbol)
 696			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 697				 is_return ? 'r' : 'p', symbol, offset);
 698		else
 699			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 700				 is_return ? 'r' : 'p', addr);
 
 701		event = buf;
 702	}
 703	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, argc,
 704			       is_return);
 
 
 705	if (IS_ERR(tk)) {
 706		pr_info("Failed to allocate trace_probe.(%d)\n",
 707			(int)PTR_ERR(tk));
 708		return PTR_ERR(tk);
 
 709	}
 
 710
 711	/* parse arguments */
 712	ret = 0;
 713	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 714		struct probe_arg *parg = &tk->tp.args[i];
 715
 716		/* Increment count for freeing args in error case */
 717		tk->tp.nr_args++;
 718
 719		/* Parse argument name */
 720		arg = strchr(argv[i], '=');
 721		if (arg) {
 722			*arg++ = '\0';
 723			parg->name = kstrdup(argv[i], GFP_KERNEL);
 724		} else {
 725			arg = argv[i];
 726			/* If argument name is omitted, set "argN" */
 727			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 728			parg->name = kstrdup(buf, GFP_KERNEL);
 729		}
 730
 731		if (!parg->name) {
 732			pr_info("Failed to allocate argument[%d] name.\n", i);
 733			ret = -ENOMEM;
 734			goto error;
 735		}
 736
 737		if (!is_good_name(parg->name)) {
 738			pr_info("Invalid argument[%d] name: %s\n",
 739				i, parg->name);
 740			ret = -EINVAL;
 741			goto error;
 742		}
 743
 744		if (traceprobe_conflict_field_name(parg->name,
 745							tk->tp.args, i)) {
 746			pr_info("Argument[%d] name '%s' conflicts with "
 747				"another field.\n", i, argv[i]);
 748			ret = -EINVAL;
 749			goto error;
 750		}
 751
 752		/* Parse fetch argument */
 753		ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 754						is_return, true,
 755						kprobes_fetch_type_table);
 756		if (ret) {
 757			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 758			goto error;
 759		}
 760	}
 761
 
 
 
 
 762	ret = register_trace_kprobe(tk);
 763	if (ret)
 
 
 
 
 
 
 
 764		goto error;
 765	return 0;
 
 
 
 
 
 766
 
 
 767error:
 768	free_trace_kprobe(tk);
 769	return ret;
 770}
 771
 772static int release_all_trace_kprobes(void)
 773{
 774	struct trace_kprobe *tk;
 775	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 776
 777	mutex_lock(&probe_lock);
 778	/* Ensure no probe is in use. */
 779	list_for_each_entry(tk, &probe_list, list)
 780		if (trace_probe_is_enabled(&tk->tp)) {
 781			ret = -EBUSY;
 782			goto end;
 
 
 
 
 
 783		}
 784	/* TODO: Use batch unregistration */
 785	while (!list_empty(&probe_list)) {
 786		tk = list_entry(probe_list.next, struct trace_kprobe, list);
 787		ret = unregister_trace_kprobe(tk);
 788		if (ret)
 789			goto end;
 790		free_trace_kprobe(tk);
 791	}
 792
 793end:
 794	mutex_unlock(&probe_lock);
 795
 796	return ret;
 797}
 
 798
 799/* Probes listing interfaces */
 800static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 801{
 802	mutex_lock(&probe_lock);
 803	return seq_list_start(&probe_list, *pos);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 804}
 
 805
 806static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 
 
 
 
 
 
 
 
 
 807{
 808	return seq_list_next(v, &probe_list, pos);
 
 
 
 
 809}
 
 810
 811static void probes_seq_stop(struct seq_file *m, void *v)
 812{
 813	mutex_unlock(&probe_lock);
 
 
 
 
 
 814}
 815
 816static int probes_seq_show(struct seq_file *m, void *v)
 817{
 818	struct trace_kprobe *tk = v;
 819	int i;
 820
 821	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 822	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 823			trace_event_name(&tk->tp.call));
 
 
 824
 825	if (!tk->symbol)
 826		seq_printf(m, " 0x%p", tk->rp.kp.addr);
 827	else if (tk->rp.kp.offset)
 828		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 829			   tk->rp.kp.offset);
 830	else
 831		seq_printf(m, " %s", trace_kprobe_symbol(tk));
 832
 833	for (i = 0; i < tk->tp.nr_args; i++)
 834		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 835	seq_putc(m, '\n');
 836
 837	return 0;
 838}
 839
 
 
 
 
 
 
 
 
 
 
 840static const struct seq_operations probes_seq_op = {
 841	.start  = probes_seq_start,
 842	.next   = probes_seq_next,
 843	.stop   = probes_seq_stop,
 844	.show   = probes_seq_show
 845};
 846
 847static int probes_open(struct inode *inode, struct file *file)
 848{
 849	int ret;
 850
 
 
 
 
 851	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 852		ret = release_all_trace_kprobes();
 853		if (ret < 0)
 854			return ret;
 855	}
 856
 857	return seq_open(file, &probes_seq_op);
 858}
 859
 860static ssize_t probes_write(struct file *file, const char __user *buffer,
 861			    size_t count, loff_t *ppos)
 862{
 863	return traceprobe_probes_write(file, buffer, count, ppos,
 864			create_trace_kprobe);
 865}
 866
 867static const struct file_operations kprobe_events_ops = {
 868	.owner          = THIS_MODULE,
 869	.open           = probes_open,
 870	.read           = seq_read,
 871	.llseek         = seq_lseek,
 872	.release        = seq_release,
 873	.write		= probes_write,
 874};
 875
 876/* Probes profiling interfaces */
 877static int probes_profile_seq_show(struct seq_file *m, void *v)
 878{
 879	struct trace_kprobe *tk = v;
 880	unsigned long nhit = 0;
 881	int cpu;
 882
 883	for_each_possible_cpu(cpu)
 884		nhit += *per_cpu_ptr(tk->nhit, cpu);
 885
 
 886	seq_printf(m, "  %-44s %15lu %15lu\n",
 887		   trace_event_name(&tk->tp.call), nhit,
 
 888		   tk->rp.kp.nmissed);
 889
 890	return 0;
 891}
 892
 893static const struct seq_operations profile_seq_op = {
 894	.start  = probes_seq_start,
 895	.next   = probes_seq_next,
 896	.stop   = probes_seq_stop,
 897	.show   = probes_profile_seq_show
 898};
 899
 900static int profile_open(struct inode *inode, struct file *file)
 901{
 
 
 
 
 
 
 902	return seq_open(file, &profile_seq_op);
 903}
 904
 905static const struct file_operations kprobe_profile_ops = {
 906	.owner          = THIS_MODULE,
 907	.open           = profile_open,
 908	.read           = seq_read,
 909	.llseek         = seq_lseek,
 910	.release        = seq_release,
 911};
 912
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 913/* Kprobe handler */
 914static nokprobe_inline void
 915__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 916		    struct trace_event_file *trace_file)
 917{
 918	struct kprobe_trace_entry_head *entry;
 919	struct ring_buffer_event *event;
 920	struct ring_buffer *buffer;
 921	int size, dsize, pc;
 922	unsigned long irq_flags;
 923	struct trace_event_call *call = &tk->tp.call;
 924
 925	WARN_ON(call != trace_file->event_call);
 926
 927	if (trace_trigger_soft_disabled(trace_file))
 928		return;
 929
 930	local_save_flags(irq_flags);
 931	pc = preempt_count();
 
 932
 933	dsize = __get_data_size(&tk->tp, regs);
 934	size = sizeof(*entry) + tk->tp.size + dsize;
 935
 936	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 937						call->event.type,
 938						size, irq_flags, pc);
 939	if (!event)
 
 
 940		return;
 941
 942	entry = ring_buffer_event_data(event);
 
 943	entry->ip = (unsigned long)tk->rp.kp.addr;
 944	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 945
 946	event_trigger_unlock_commit_regs(trace_file, buffer, event,
 947					 entry, irq_flags, pc, regs);
 948}
 949
 950static void
 951kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
 952{
 953	struct event_file_link *link;
 954
 955	list_for_each_entry_rcu(link, &tk->tp.files, list)
 956		__kprobe_trace_func(tk, regs, link->file);
 957}
 958NOKPROBE_SYMBOL(kprobe_trace_func);
 959
 960/* Kretprobe handler */
 961static nokprobe_inline void
 962__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
 963		       struct pt_regs *regs,
 964		       struct trace_event_file *trace_file)
 965{
 966	struct kretprobe_trace_entry_head *entry;
 967	struct ring_buffer_event *event;
 968	struct ring_buffer *buffer;
 969	int size, pc, dsize;
 970	unsigned long irq_flags;
 971	struct trace_event_call *call = &tk->tp.call;
 972
 973	WARN_ON(call != trace_file->event_call);
 974
 975	if (trace_trigger_soft_disabled(trace_file))
 976		return;
 977
 978	local_save_flags(irq_flags);
 979	pc = preempt_count();
 
 980
 981	dsize = __get_data_size(&tk->tp, regs);
 982	size = sizeof(*entry) + tk->tp.size + dsize;
 983
 984	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 985						call->event.type,
 986						size, irq_flags, pc);
 987	if (!event)
 988		return;
 989
 990	entry = ring_buffer_event_data(event);
 
 991	entry->func = (unsigned long)tk->rp.kp.addr;
 992	entry->ret_ip = (unsigned long)ri->ret_addr;
 993	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
 994
 995	event_trigger_unlock_commit_regs(trace_file, buffer, event,
 996					 entry, irq_flags, pc, regs);
 997}
 998
 999static void
1000kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1001		     struct pt_regs *regs)
1002{
1003	struct event_file_link *link;
1004
1005	list_for_each_entry_rcu(link, &tk->tp.files, list)
1006		__kretprobe_trace_func(tk, ri, regs, link->file);
1007}
1008NOKPROBE_SYMBOL(kretprobe_trace_func);
1009
1010/* Event entry printers */
1011static enum print_line_t
1012print_kprobe_event(struct trace_iterator *iter, int flags,
1013		   struct trace_event *event)
1014{
1015	struct kprobe_trace_entry_head *field;
1016	struct trace_seq *s = &iter->seq;
1017	struct trace_probe *tp;
1018	u8 *data;
1019	int i;
1020
1021	field = (struct kprobe_trace_entry_head *)iter->ent;
1022	tp = container_of(event, struct trace_probe, call.event);
 
 
 
1023
1024	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1025
1026	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1027		goto out;
1028
1029	trace_seq_putc(s, ')');
1030
1031	data = (u8 *)&field[1];
1032	for (i = 0; i < tp->nr_args; i++)
1033		if (!tp->args[i].type->print(s, tp->args[i].name,
1034					     data + tp->args[i].offset, field))
1035			goto out;
1036
1037	trace_seq_putc(s, '\n');
1038 out:
1039	return trace_handle_return(s);
1040}
1041
1042static enum print_line_t
1043print_kretprobe_event(struct trace_iterator *iter, int flags,
1044		      struct trace_event *event)
1045{
1046	struct kretprobe_trace_entry_head *field;
1047	struct trace_seq *s = &iter->seq;
1048	struct trace_probe *tp;
1049	u8 *data;
1050	int i;
1051
1052	field = (struct kretprobe_trace_entry_head *)iter->ent;
1053	tp = container_of(event, struct trace_probe, call.event);
 
 
 
1054
1055	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1056
1057	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1058		goto out;
1059
1060	trace_seq_puts(s, " <- ");
1061
1062	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1063		goto out;
1064
1065	trace_seq_putc(s, ')');
1066
1067	data = (u8 *)&field[1];
1068	for (i = 0; i < tp->nr_args; i++)
1069		if (!tp->args[i].type->print(s, tp->args[i].name,
1070					     data + tp->args[i].offset, field))
1071			goto out;
1072
1073	trace_seq_putc(s, '\n');
1074
1075 out:
1076	return trace_handle_return(s);
1077}
1078
1079
1080static int kprobe_event_define_fields(struct trace_event_call *event_call)
1081{
1082	int ret, i;
1083	struct kprobe_trace_entry_head field;
1084	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
 
 
 
 
1085
1086	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1087	/* Set argument names as fields */
1088	for (i = 0; i < tk->tp.nr_args; i++) {
1089		struct probe_arg *parg = &tk->tp.args[i];
1090
1091		ret = trace_define_field(event_call, parg->type->fmttype,
1092					 parg->name,
1093					 sizeof(field) + parg->offset,
1094					 parg->type->size,
1095					 parg->type->is_signed,
1096					 FILTER_OTHER);
1097		if (ret)
1098			return ret;
1099	}
1100	return 0;
1101}
1102
1103static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1104{
1105	int ret, i;
1106	struct kretprobe_trace_entry_head field;
1107	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
 
 
 
 
1108
1109	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1110	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1111	/* Set argument names as fields */
1112	for (i = 0; i < tk->tp.nr_args; i++) {
1113		struct probe_arg *parg = &tk->tp.args[i];
1114
1115		ret = trace_define_field(event_call, parg->type->fmttype,
1116					 parg->name,
1117					 sizeof(field) + parg->offset,
1118					 parg->type->size,
1119					 parg->type->is_signed,
1120					 FILTER_OTHER);
1121		if (ret)
1122			return ret;
1123	}
1124	return 0;
1125}
1126
1127#ifdef CONFIG_PERF_EVENTS
1128
1129/* Kprobe profile handler */
1130static void
1131kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1132{
1133	struct trace_event_call *call = &tk->tp.call;
1134	struct bpf_prog *prog = call->prog;
1135	struct kprobe_trace_entry_head *entry;
1136	struct hlist_head *head;
1137	int size, __size, dsize;
1138	int rctx;
1139
1140	if (prog && !trace_call_bpf(prog, regs))
1141		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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, NULL, &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}
1161NOKPROBE_SYMBOL(kprobe_perf_func);
1162
1163/* Kretprobe profile handler */
1164static void
1165kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1166		    struct pt_regs *regs)
1167{
1168	struct trace_event_call *call = &tk->tp.call;
1169	struct bpf_prog *prog = call->prog;
1170	struct kretprobe_trace_entry_head *entry;
1171	struct hlist_head *head;
1172	int size, __size, dsize;
1173	int rctx;
1174
1175	if (prog && !trace_call_bpf(prog, regs))
1176		return;
1177
1178	head = this_cpu_ptr(call->perf_events);
1179	if (hlist_empty(head))
1180		return;
1181
1182	dsize = __get_data_size(&tk->tp, regs);
1183	__size = sizeof(*entry) + tk->tp.size + dsize;
1184	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1185	size -= sizeof(u32);
1186
1187	entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1188	if (!entry)
1189		return;
1190
1191	entry->func = (unsigned long)tk->rp.kp.addr;
1192	entry->ret_ip = (unsigned long)ri->ret_addr;
1193	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1194	perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 
1195}
1196NOKPROBE_SYMBOL(kretprobe_perf_func);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1197#endif	/* CONFIG_PERF_EVENTS */
1198
1199/*
1200 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1201 *
1202 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1203 * lockless, but we can't race with this __init function.
1204 */
1205static int kprobe_register(struct trace_event_call *event,
1206			   enum trace_reg type, void *data)
1207{
1208	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1209	struct trace_event_file *file = data;
1210
1211	switch (type) {
1212	case TRACE_REG_REGISTER:
1213		return enable_trace_kprobe(tk, file);
1214	case TRACE_REG_UNREGISTER:
1215		return disable_trace_kprobe(tk, file);
1216
1217#ifdef CONFIG_PERF_EVENTS
1218	case TRACE_REG_PERF_REGISTER:
1219		return enable_trace_kprobe(tk, NULL);
1220	case TRACE_REG_PERF_UNREGISTER:
1221		return disable_trace_kprobe(tk, NULL);
1222	case TRACE_REG_PERF_OPEN:
1223	case TRACE_REG_PERF_CLOSE:
1224	case TRACE_REG_PERF_ADD:
1225	case TRACE_REG_PERF_DEL:
1226		return 0;
1227#endif
1228	}
1229	return 0;
1230}
1231
1232static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1233{
1234	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
 
1235
1236	raw_cpu_inc(*tk->nhit);
1237
1238	if (tk->tp.flags & TP_FLAG_TRACE)
1239		kprobe_trace_func(tk, regs);
1240#ifdef CONFIG_PERF_EVENTS
1241	if (tk->tp.flags & TP_FLAG_PROFILE)
1242		kprobe_perf_func(tk, regs);
1243#endif
1244	return 0;	/* We don't tweek kernel, so just return 0 */
1245}
1246NOKPROBE_SYMBOL(kprobe_dispatcher);
1247
1248static int
1249kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1250{
1251	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1252
1253	raw_cpu_inc(*tk->nhit);
1254
1255	if (tk->tp.flags & TP_FLAG_TRACE)
1256		kretprobe_trace_func(tk, ri, regs);
1257#ifdef CONFIG_PERF_EVENTS
1258	if (tk->tp.flags & TP_FLAG_PROFILE)
1259		kretprobe_perf_func(tk, ri, regs);
1260#endif
1261	return 0;	/* We don't tweek kernel, so just return 0 */
1262}
1263NOKPROBE_SYMBOL(kretprobe_dispatcher);
1264
1265static struct trace_event_functions kretprobe_funcs = {
1266	.trace		= print_kretprobe_event
1267};
1268
1269static struct trace_event_functions kprobe_funcs = {
1270	.trace		= print_kprobe_event
1271};
1272
1273static int register_kprobe_event(struct trace_kprobe *tk)
 
 
 
 
 
 
 
 
 
 
 
 
1274{
1275	struct trace_event_call *call = &tk->tp.call;
1276	int ret;
1277
1278	/* Initialize trace_event_call */
1279	INIT_LIST_HEAD(&call->class->fields);
1280	if (trace_kprobe_is_return(tk)) {
1281		call->event.funcs = &kretprobe_funcs;
1282		call->class->define_fields = kretprobe_event_define_fields;
1283	} else {
1284		call->event.funcs = &kprobe_funcs;
1285		call->class->define_fields = kprobe_event_define_fields;
1286	}
1287	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1288		return -ENOMEM;
1289	ret = register_trace_event(&call->event);
1290	if (!ret) {
1291		kfree(call->print_fmt);
1292		return -ENODEV;
1293	}
 
1294	call->flags = TRACE_EVENT_FL_KPROBE;
1295	call->class->reg = kprobe_register;
1296	call->data = tk;
1297	ret = trace_add_event_call(call);
1298	if (ret) {
1299		pr_info("Failed to register kprobe event: %s\n",
1300			trace_event_name(call));
1301		kfree(call->print_fmt);
1302		unregister_trace_event(&call->event);
1303	}
1304	return ret;
1305}
1306
1307static int unregister_kprobe_event(struct trace_kprobe *tk)
1308{
 
 
 
 
 
 
 
 
 
 
1309	int ret;
 
1310
1311	/* tp->event is unregistered in trace_remove_event_call() */
1312	ret = trace_remove_event_call(&tk->tp.call);
1313	if (!ret)
1314		kfree(tk->tp.call.print_fmt);
1315	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1316}
 
1317
1318/* Make a tracefs interface for controlling probe points */
1319static __init int init_kprobe_trace(void)
1320{
1321	struct dentry *d_tracer;
1322	struct dentry *entry;
1323
1324	if (register_module_notifier(&trace_kprobe_module_nb))
1325		return -EINVAL;
1326
1327	d_tracer = tracing_init_dentry();
1328	if (IS_ERR(d_tracer))
1329		return 0;
1330
1331	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1332				    NULL, &kprobe_events_ops);
1333
1334	/* Event list interface */
1335	if (!entry)
1336		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1337
1338	/* Profile interface */
1339	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1340				    NULL, &kprobe_profile_ops);
1341
1342	if (!entry)
1343		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
 
 
 
1344	return 0;
1345}
1346fs_initcall(init_kprobe_trace);
1347
1348
1349#ifdef CONFIG_FTRACE_STARTUP_TEST
1350
1351/*
1352 * The "__used" keeps gcc from removing the function symbol
1353 * from the kallsyms table.
1354 */
1355static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1356					       int a4, int a5, int a6)
1357{
1358	return a1 + a2 + a3 + a4 + a5 + a6;
1359}
1360
1361static struct trace_event_file *
1362find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1363{
1364	struct trace_event_file *file;
1365
1366	list_for_each_entry(file, &tr->events, list)
1367		if (file->event_call == &tk->tp.call)
1368			return file;
1369
1370	return NULL;
1371}
1372
1373/*
1374 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1375 * stage, we can do this lockless.
1376 */
1377static __init int kprobe_trace_self_tests_init(void)
1378{
1379	int ret, warn = 0;
1380	int (*target)(int, int, int, int, int, int);
1381	struct trace_kprobe *tk;
1382	struct trace_event_file *file;
1383
1384	if (tracing_is_disabled())
1385		return -ENODEV;
1386
 
 
 
 
 
1387	target = kprobe_trace_selftest_target;
1388
1389	pr_info("Testing kprobe tracing: ");
1390
1391	ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1392				  "$stack $stack0 +0($stack)",
1393				  create_trace_kprobe);
1394	if (WARN_ON_ONCE(ret)) {
1395		pr_warn("error on probing function entry.\n");
1396		warn++;
1397	} else {
1398		/* Enable trace point */
1399		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1400		if (WARN_ON_ONCE(tk == NULL)) {
1401			pr_warn("error on getting new probe.\n");
1402			warn++;
1403		} else {
1404			file = find_trace_probe_file(tk, top_trace_array());
1405			if (WARN_ON_ONCE(file == NULL)) {
1406				pr_warn("error on getting probe file.\n");
1407				warn++;
1408			} else
1409				enable_trace_kprobe(tk, file);
 
1410		}
1411	}
1412
1413	ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1414				  "$retval", create_trace_kprobe);
1415	if (WARN_ON_ONCE(ret)) {
1416		pr_warn("error on probing function return.\n");
1417		warn++;
1418	} else {
1419		/* Enable trace point */
1420		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1421		if (WARN_ON_ONCE(tk == NULL)) {
1422			pr_warn("error on getting 2nd new probe.\n");
1423			warn++;
1424		} else {
1425			file = find_trace_probe_file(tk, top_trace_array());
1426			if (WARN_ON_ONCE(file == NULL)) {
1427				pr_warn("error on getting probe file.\n");
1428				warn++;
1429			} else
1430				enable_trace_kprobe(tk, file);
 
1431		}
1432	}
1433
1434	if (warn)
1435		goto end;
1436
1437	ret = target(1, 2, 3, 4, 5, 6);
1438
 
 
 
 
 
 
 
 
1439	/* Disable trace points before removing it */
1440	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1441	if (WARN_ON_ONCE(tk == NULL)) {
1442		pr_warn("error on getting test probe.\n");
1443		warn++;
1444	} else {
 
 
 
 
 
1445		file = find_trace_probe_file(tk, top_trace_array());
1446		if (WARN_ON_ONCE(file == NULL)) {
1447			pr_warn("error on getting probe file.\n");
1448			warn++;
1449		} else
1450			disable_trace_kprobe(tk, file);
 
1451	}
1452
1453	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1454	if (WARN_ON_ONCE(tk == NULL)) {
1455		pr_warn("error on getting 2nd test probe.\n");
1456		warn++;
1457	} else {
 
 
 
 
 
1458		file = find_trace_probe_file(tk, top_trace_array());
1459		if (WARN_ON_ONCE(file == NULL)) {
1460			pr_warn("error on getting probe file.\n");
1461			warn++;
1462		} else
1463			disable_trace_kprobe(tk, file);
 
1464	}
1465
1466	ret = traceprobe_command("-:testprobe", create_trace_kprobe);
1467	if (WARN_ON_ONCE(ret)) {
1468		pr_warn("error on deleting a probe.\n");
1469		warn++;
1470	}
1471
1472	ret = traceprobe_command("-:testprobe2", create_trace_kprobe);
1473	if (WARN_ON_ONCE(ret)) {
1474		pr_warn("error on deleting a probe.\n");
1475		warn++;
1476	}
1477
1478end:
1479	release_all_trace_kprobes();
 
 
 
 
 
 
 
 
 
1480	if (warn)
1481		pr_cont("NG: Some tests are failed. Please check them.\n");
1482	else
1483		pr_cont("OK\n");
1484	return 0;
1485}
1486
1487late_initcall(kprobe_trace_self_tests_init);
1488
1489#endif
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Kprobes-based tracing events
   4 *
   5 * Created by Masami Hiramatsu <mhiramat@redhat.com>
   6 *
 
 
 
 
 
 
 
 
 
 
 
 
   7 */
   8#define pr_fmt(fmt)	"trace_kprobe: " fmt
   9
  10#include <linux/security.h>
  11#include <linux/module.h>
  12#include <linux/uaccess.h>
  13#include <linux/rculist.h>
  14#include <linux/error-injection.h>
  15
  16#include <asm/setup.h>  /* for COMMAND_LINE_SIZE */
  17
  18#include "trace_dynevent.h"
  19#include "trace_kprobe_selftest.h"
  20#include "trace_probe.h"
  21#include "trace_probe_tmpl.h"
  22
  23#define KPROBE_EVENT_SYSTEM "kprobes"
  24#define KRETPROBE_MAXACTIVE_MAX 4096
  25
  26/* Kprobe early definition from command line */
  27static char kprobe_boot_events_buf[COMMAND_LINE_SIZE] __initdata;
  28static bool kprobe_boot_events_enabled __initdata;
  29
  30static int __init set_kprobe_boot_events(char *str)
  31{
  32	strlcpy(kprobe_boot_events_buf, str, COMMAND_LINE_SIZE);
  33	return 0;
  34}
  35__setup("kprobe_event=", set_kprobe_boot_events);
  36
  37static int trace_kprobe_create(int argc, const char **argv);
  38static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev);
  39static int trace_kprobe_release(struct dyn_event *ev);
  40static bool trace_kprobe_is_busy(struct dyn_event *ev);
  41static bool trace_kprobe_match(const char *system, const char *event,
  42			int argc, const char **argv, struct dyn_event *ev);
  43
  44static struct dyn_event_operations trace_kprobe_ops = {
  45	.create = trace_kprobe_create,
  46	.show = trace_kprobe_show,
  47	.is_busy = trace_kprobe_is_busy,
  48	.free = trace_kprobe_release,
  49	.match = trace_kprobe_match,
  50};
  51
  52/*
  53 * Kprobe event core functions
  54 */
  55struct trace_kprobe {
  56	struct dyn_event	devent;
  57	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
  58	unsigned long __percpu *nhit;
  59	const char		*symbol;	/* symbol name */
  60	struct trace_probe	tp;
  61};
  62
  63static bool is_trace_kprobe(struct dyn_event *ev)
  64{
  65	return ev->ops == &trace_kprobe_ops;
  66}
  67
  68static struct trace_kprobe *to_trace_kprobe(struct dyn_event *ev)
  69{
  70	return container_of(ev, struct trace_kprobe, devent);
  71}
  72
  73/**
  74 * for_each_trace_kprobe - iterate over the trace_kprobe list
  75 * @pos:	the struct trace_kprobe * for each entry
  76 * @dpos:	the struct dyn_event * to use as a loop cursor
  77 */
  78#define for_each_trace_kprobe(pos, dpos)	\
  79	for_each_dyn_event(dpos)		\
  80		if (is_trace_kprobe(dpos) && (pos = to_trace_kprobe(dpos)))
  81
  82#define SIZEOF_TRACE_KPROBE(n)				\
  83	(offsetof(struct trace_kprobe, tp.args) +	\
  84	(sizeof(struct probe_arg) * (n)))
  85
 
  86static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  87{
  88	return tk->rp.handler != NULL;
  89}
  90
  91static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  92{
  93	return tk->symbol ? tk->symbol : "unknown";
  94}
  95
  96static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  97{
  98	return tk->rp.kp.offset;
  99}
 100
 101static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
 102{
 103	return !!(kprobe_gone(&tk->rp.kp));
 104}
 105
 106static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
 107						 struct module *mod)
 108{
 109	int len = strlen(mod->name);
 110	const char *name = trace_kprobe_symbol(tk);
 111	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
 112}
 113
 114static nokprobe_inline bool trace_kprobe_module_exist(struct trace_kprobe *tk)
 115{
 116	char *p;
 117	bool ret;
 118
 119	if (!tk->symbol)
 120		return false;
 121	p = strchr(tk->symbol, ':');
 122	if (!p)
 123		return true;
 124	*p = '\0';
 125	mutex_lock(&module_mutex);
 126	ret = !!find_module(tk->symbol);
 127	mutex_unlock(&module_mutex);
 128	*p = ':';
 129
 130	return ret;
 131}
 132
 133static bool trace_kprobe_is_busy(struct dyn_event *ev)
 134{
 135	struct trace_kprobe *tk = to_trace_kprobe(ev);
 136
 137	return trace_probe_is_enabled(&tk->tp);
 138}
 
 
 
 
 139
 140static bool trace_kprobe_match_command_head(struct trace_kprobe *tk,
 141					    int argc, const char **argv)
 142{
 143	char buf[MAX_ARGSTR_LEN + 1];
 144
 145	if (!argc)
 146		return true;
 147
 148	if (!tk->symbol)
 149		snprintf(buf, sizeof(buf), "0x%p", tk->rp.kp.addr);
 150	else if (tk->rp.kp.offset)
 151		snprintf(buf, sizeof(buf), "%s+%u",
 152			 trace_kprobe_symbol(tk), tk->rp.kp.offset);
 153	else
 154		snprintf(buf, sizeof(buf), "%s", trace_kprobe_symbol(tk));
 155	if (strcmp(buf, argv[0]))
 156		return false;
 157	argc--; argv++;
 158
 159	return trace_probe_match_command_args(&tk->tp, argc, argv);
 
 
 
 160}
 161
 162static bool trace_kprobe_match(const char *system, const char *event,
 163			int argc, const char **argv, struct dyn_event *ev)
 164{
 165	struct trace_kprobe *tk = to_trace_kprobe(ev);
 166
 167	return strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 168	    (!system || strcmp(trace_probe_group_name(&tk->tp), system) == 0) &&
 169	    trace_kprobe_match_command_head(tk, argc, argv);
 170}
 171
 172static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 173{
 174	unsigned long nhit = 0;
 175	int cpu;
 176
 177	for_each_possible_cpu(cpu)
 178		nhit += *per_cpu_ptr(tk->nhit, cpu);
 
 
 
 
 
 179
 180	return nhit;
 181}
 182
 183static nokprobe_inline bool trace_kprobe_is_registered(struct trace_kprobe *tk)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 184{
 185	return !(list_empty(&tk->rp.kp.list) &&
 186		 hlist_unhashed(&tk->rp.kp.hlist));
 187}
 
 
 
 188
 189/* Return 0 if it fails to find the symbol address */
 190static nokprobe_inline
 191unsigned long trace_kprobe_address(struct trace_kprobe *tk)
 192{
 193	unsigned long addr;
 194
 195	if (tk->symbol) {
 196		addr = (unsigned long)
 197			kallsyms_lookup_name(trace_kprobe_symbol(tk));
 198		if (addr)
 199			addr += tk->rp.kp.offset;
 200	} else {
 201		addr = (unsigned long)tk->rp.kp.addr;
 202	}
 203	return addr;
 204}
 
 205
 206static nokprobe_inline struct trace_kprobe *
 207trace_kprobe_primary_from_call(struct trace_event_call *call)
 
 208{
 209	struct trace_probe *tp;
 
 
 210
 211	tp = trace_probe_primary_from_call(call);
 212	if (WARN_ON_ONCE(!tp))
 213		return NULL;
 214
 215	return container_of(tp, struct trace_kprobe, tp);
 216}
 
 
 217
 218bool trace_kprobe_on_func_entry(struct trace_event_call *call)
 219{
 220	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 221
 222	return tk ? kprobe_on_func_entry(tk->rp.kp.addr,
 223			tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 224			tk->rp.kp.addr ? 0 : tk->rp.kp.offset) : false;
 
 225}
 
 226
 227bool trace_kprobe_error_injectable(struct trace_event_call *call)
 228{
 229	struct trace_kprobe *tk = trace_kprobe_primary_from_call(call);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 230
 231	return tk ? within_error_injection_list(trace_kprobe_address(tk)) :
 232	       false;
 233}
 234
 235static int register_kprobe_event(struct trace_kprobe *tk);
 236static int unregister_kprobe_event(struct trace_kprobe *tk);
 237
 238static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 239static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 240				struct pt_regs *regs);
 241
 242static void free_trace_kprobe(struct trace_kprobe *tk)
 243{
 244	if (tk) {
 245		trace_probe_cleanup(&tk->tp);
 246		kfree(tk->symbol);
 247		free_percpu(tk->nhit);
 248		kfree(tk);
 249	}
 250}
 251
 252/*
 253 * Allocate new trace_probe and initialize it (including kprobes).
 254 */
 255static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 256					     const char *event,
 257					     void *addr,
 258					     const char *symbol,
 259					     unsigned long offs,
 260					     int maxactive,
 261					     int nargs, bool is_return)
 262{
 263	struct trace_kprobe *tk;
 264	int ret = -ENOMEM;
 265
 266	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 267	if (!tk)
 268		return ERR_PTR(ret);
 269
 270	tk->nhit = alloc_percpu(unsigned long);
 271	if (!tk->nhit)
 272		goto error;
 273
 274	if (symbol) {
 275		tk->symbol = kstrdup(symbol, GFP_KERNEL);
 276		if (!tk->symbol)
 277			goto error;
 278		tk->rp.kp.symbol_name = tk->symbol;
 279		tk->rp.kp.offset = offs;
 280	} else
 281		tk->rp.kp.addr = addr;
 282
 283	if (is_return)
 284		tk->rp.handler = kretprobe_dispatcher;
 285	else
 286		tk->rp.kp.pre_handler = kprobe_dispatcher;
 287
 288	tk->rp.maxactive = maxactive;
 289	INIT_HLIST_NODE(&tk->rp.kp.hlist);
 290	INIT_LIST_HEAD(&tk->rp.kp.list);
 
 
 
 
 
 
 291
 292	ret = trace_probe_init(&tk->tp, event, group, false);
 293	if (ret < 0)
 
 
 
 
 
 294		goto error;
 295
 296	dyn_event_init(&tk->devent, &trace_kprobe_ops);
 
 297	return tk;
 298error:
 299	free_trace_kprobe(tk);
 
 
 
 300	return ERR_PTR(ret);
 301}
 302
 303static struct trace_kprobe *find_trace_kprobe(const char *event,
 304					      const char *group)
 305{
 306	struct dyn_event *pos;
 307	struct trace_kprobe *tk;
 308
 309	for_each_trace_kprobe(tk, pos)
 310		if (strcmp(trace_probe_name(&tk->tp), event) == 0 &&
 311		    strcmp(trace_probe_group_name(&tk->tp), group) == 0)
 312			return tk;
 313	return NULL;
 314}
 315
 316static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 317{
 318	int ret = 0;
 319
 320	if (trace_kprobe_is_registered(tk) && !trace_kprobe_has_gone(tk)) {
 321		if (trace_kprobe_is_return(tk))
 322			ret = enable_kretprobe(&tk->rp);
 323		else
 324			ret = enable_kprobe(&tk->rp.kp);
 325	}
 326
 327	return ret;
 
 
 
 
 328}
 329
 330static void __disable_trace_kprobe(struct trace_probe *tp)
 
 331{
 332	struct trace_probe *pos;
 333	struct trace_kprobe *tk;
 334
 335	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 336		tk = container_of(pos, struct trace_kprobe, tp);
 337		if (!trace_kprobe_is_registered(tk))
 338			continue;
 339		if (trace_kprobe_is_return(tk))
 340			disable_kretprobe(&tk->rp);
 341		else
 342			disable_kprobe(&tk->rp.kp);
 343	}
 344}
 345
 346/*
 347 * Enable trace_probe
 348 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 349 */
 350static int enable_trace_kprobe(struct trace_event_call *call,
 351				struct trace_event_file *file)
 352{
 353	struct trace_probe *pos, *tp;
 354	struct trace_kprobe *tk;
 355	bool enabled;
 356	int ret = 0;
 357
 358	tp = trace_probe_primary_from_call(call);
 359	if (WARN_ON_ONCE(!tp))
 360		return -ENODEV;
 361	enabled = trace_probe_is_enabled(tp);
 362
 363	/* This also changes "enabled" state */
 364	if (file) {
 365		ret = trace_probe_add_file(tp, file);
 366		if (ret)
 367			return ret;
 368	} else
 369		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
 370
 371	if (enabled)
 372		return 0;
 373
 374	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 375		tk = container_of(pos, struct trace_kprobe, tp);
 376		if (trace_kprobe_has_gone(tk))
 377			continue;
 378		ret = __enable_trace_kprobe(tk);
 379		if (ret)
 380			break;
 381		enabled = true;
 382	}
 383
 384	if (ret) {
 385		/* Failed to enable one of them. Roll back all */
 386		if (enabled)
 387			__disable_trace_kprobe(tp);
 388		if (file)
 389			trace_probe_remove_file(tp, file);
 390		else
 391			trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 392	}
 393
 394	return ret;
 395}
 396
 397/*
 398 * Disable trace_probe
 399 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 400 */
 401static int disable_trace_kprobe(struct trace_event_call *call,
 402				struct trace_event_file *file)
 403{
 404	struct trace_probe *tp;
 
 
 405
 406	tp = trace_probe_primary_from_call(call);
 407	if (WARN_ON_ONCE(!tp))
 408		return -ENODEV;
 
 
 
 409
 410	if (file) {
 411		if (!trace_probe_get_file_link(tp, file))
 412			return -ENOENT;
 413		if (!trace_probe_has_single_file(tp))
 414			goto out;
 415		trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 
 416	} else
 417		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 418
 419	if (!trace_probe_is_enabled(tp))
 420		__disable_trace_kprobe(tp);
 421
 
 
 
 
 
 
 
 422 out:
 423	if (file)
 424		/*
 425		 * Synchronization is done in below function. For perf event,
 426		 * file == NULL and perf_trace_event_unreg() calls
 427		 * tracepoint_synchronize_unregister() to ensure synchronize
 428		 * event. We don't need to care about it.
 
 
 429		 */
 430		trace_probe_remove_file(tp, file);
 431
 432	return 0;
 433}
 434
 435#if defined(CONFIG_KPROBES_ON_FTRACE) && \
 436	!defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 437static bool __within_notrace_func(unsigned long addr)
 438{
 439	unsigned long offset, size;
 440
 441	if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 442		return false;
 443
 444	/* Get the entry address of the target function */
 445	addr -= offset;
 446
 447	/*
 448	 * Since ftrace_location_range() does inclusive range check, we need
 449	 * to subtract 1 byte from the end address.
 450	 */
 451	return !ftrace_location_range(addr, addr + size - 1);
 452}
 453
 454static bool within_notrace_func(struct trace_kprobe *tk)
 455{
 456	unsigned long addr = trace_kprobe_address(tk);
 457	char symname[KSYM_NAME_LEN], *p;
 458
 459	if (!__within_notrace_func(addr))
 460		return false;
 461
 462	/* Check if the address is on a suffixed-symbol */
 463	if (!lookup_symbol_name(addr, symname)) {
 464		p = strchr(symname, '.');
 465		if (!p)
 466			return true;
 467		*p = '\0';
 468		addr = (unsigned long)kprobe_lookup_name(symname, 0);
 469		if (addr)
 470			return __within_notrace_func(addr);
 471	}
 472
 473	return true;
 474}
 475#else
 476#define within_notrace_func(tk)	(false)
 477#endif
 478
 479/* Internal register function - just handle k*probes and flags */
 480static int __register_trace_kprobe(struct trace_kprobe *tk)
 481{
 482	int i, ret;
 483
 484	ret = security_locked_down(LOCKDOWN_KPROBES);
 485	if (ret)
 486		return ret;
 487
 488	if (trace_kprobe_is_registered(tk))
 489		return -EINVAL;
 490
 491	if (within_notrace_func(tk)) {
 492		pr_warn("Could not probe notrace function %s\n",
 493			trace_kprobe_symbol(tk));
 494		return -EINVAL;
 495	}
 496
 497	for (i = 0; i < tk->tp.nr_args; i++) {
 498		ret = traceprobe_update_arg(&tk->tp.args[i]);
 499		if (ret)
 500			return ret;
 501	}
 502
 503	/* Set/clear disabled flag according to tp->flag */
 504	if (trace_probe_is_enabled(&tk->tp))
 505		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 506	else
 507		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 508
 509	if (trace_kprobe_is_return(tk))
 510		ret = register_kretprobe(&tk->rp);
 511	else
 512		ret = register_kprobe(&tk->rp.kp);
 513
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 514	return ret;
 515}
 516
 517/* Internal unregister function - just handle k*probes and flags */
 518static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 519{
 520	if (trace_kprobe_is_registered(tk)) {
 521		if (trace_kprobe_is_return(tk))
 522			unregister_kretprobe(&tk->rp);
 523		else
 524			unregister_kprobe(&tk->rp.kp);
 525		/* Cleanup kprobe for reuse and mark it unregistered */
 526		INIT_HLIST_NODE(&tk->rp.kp.hlist);
 527		INIT_LIST_HEAD(&tk->rp.kp.list);
 528		if (tk->rp.kp.symbol_name)
 529			tk->rp.kp.addr = NULL;
 530	}
 531}
 532
 533/* Unregister a trace_probe and probe_event */
 534static int unregister_trace_kprobe(struct trace_kprobe *tk)
 535{
 536	/* If other probes are on the event, just unregister kprobe */
 537	if (trace_probe_has_sibling(&tk->tp))
 538		goto unreg;
 539
 540	/* Enabled event can not be unregistered */
 541	if (trace_probe_is_enabled(&tk->tp))
 542		return -EBUSY;
 543
 544	/* Will fail if probe is being used by ftrace or perf */
 545	if (unregister_kprobe_event(tk))
 546		return -EBUSY;
 547
 548unreg:
 549	__unregister_trace_kprobe(tk);
 550	dyn_event_remove(&tk->devent);
 551	trace_probe_unlink(&tk->tp);
 552
 553	return 0;
 554}
 555
 556static bool trace_kprobe_has_same_kprobe(struct trace_kprobe *orig,
 557					 struct trace_kprobe *comp)
 558{
 559	struct trace_probe_event *tpe = orig->tp.event;
 560	struct trace_probe *pos;
 561	int i;
 562
 563	list_for_each_entry(pos, &tpe->probes, list) {
 564		orig = container_of(pos, struct trace_kprobe, tp);
 565		if (strcmp(trace_kprobe_symbol(orig),
 566			   trace_kprobe_symbol(comp)) ||
 567		    trace_kprobe_offset(orig) != trace_kprobe_offset(comp))
 568			continue;
 569
 570		/*
 571		 * trace_probe_compare_arg_type() ensured that nr_args and
 572		 * each argument name and type are same. Let's compare comm.
 573		 */
 574		for (i = 0; i < orig->tp.nr_args; i++) {
 575			if (strcmp(orig->tp.args[i].comm,
 576				   comp->tp.args[i].comm))
 577				break;
 578		}
 579
 580		if (i == orig->tp.nr_args)
 581			return true;
 582	}
 583
 584	return false;
 585}
 586
 587static int append_trace_kprobe(struct trace_kprobe *tk, struct trace_kprobe *to)
 588{
 589	int ret;
 590
 591	ret = trace_probe_compare_arg_type(&tk->tp, &to->tp);
 592	if (ret) {
 593		/* Note that argument starts index = 2 */
 594		trace_probe_log_set_index(ret + 1);
 595		trace_probe_log_err(0, DIFF_ARG_TYPE);
 596		return -EEXIST;
 597	}
 598	if (trace_kprobe_has_same_kprobe(to, tk)) {
 599		trace_probe_log_set_index(0);
 600		trace_probe_log_err(0, SAME_PROBE);
 601		return -EEXIST;
 602	}
 603
 604	/* Append to existing event */
 605	ret = trace_probe_append(&tk->tp, &to->tp);
 606	if (ret)
 607		return ret;
 608
 609	/* Register k*probe */
 610	ret = __register_trace_kprobe(tk);
 611	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 612		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 613		ret = 0;
 614	}
 615
 616	if (ret)
 617		trace_probe_unlink(&tk->tp);
 618	else
 619		dyn_event_add(&tk->devent);
 620
 621	return ret;
 622}
 623
 624/* Register a trace_probe and probe_event */
 625static int register_trace_kprobe(struct trace_kprobe *tk)
 626{
 627	struct trace_kprobe *old_tk;
 628	int ret;
 629
 630	mutex_lock(&event_mutex);
 631
 632	old_tk = find_trace_kprobe(trace_probe_name(&tk->tp),
 633				   trace_probe_group_name(&tk->tp));
 
 634	if (old_tk) {
 635		if (trace_kprobe_is_return(tk) != trace_kprobe_is_return(old_tk)) {
 636			trace_probe_log_set_index(0);
 637			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 638			ret = -EEXIST;
 639		} else {
 640			ret = append_trace_kprobe(tk, old_tk);
 641		}
 642		goto end;
 643	}
 644
 645	/* Register new event */
 646	ret = register_kprobe_event(tk);
 647	if (ret) {
 648		pr_warn("Failed to register probe event(%d)\n", ret);
 649		goto end;
 650	}
 651
 652	/* Register k*probe */
 653	ret = __register_trace_kprobe(tk);
 654	if (ret == -ENOENT && !trace_kprobe_module_exist(tk)) {
 655		pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 656		ret = 0;
 657	}
 658
 659	if (ret < 0)
 660		unregister_kprobe_event(tk);
 661	else
 662		dyn_event_add(&tk->devent);
 663
 664end:
 665	mutex_unlock(&event_mutex);
 666	return ret;
 667}
 668
 669/* Module notifier call back, checking event on the module */
 670static int trace_kprobe_module_callback(struct notifier_block *nb,
 671				       unsigned long val, void *data)
 672{
 673	struct module *mod = data;
 674	struct dyn_event *pos;
 675	struct trace_kprobe *tk;
 676	int ret;
 677
 678	if (val != MODULE_STATE_COMING)
 679		return NOTIFY_DONE;
 680
 681	/* Update probes on coming module */
 682	mutex_lock(&event_mutex);
 683	for_each_trace_kprobe(tk, pos) {
 684		if (trace_kprobe_within_module(tk, mod)) {
 685			/* Don't need to check busy - this should have gone. */
 686			__unregister_trace_kprobe(tk);
 687			ret = __register_trace_kprobe(tk);
 688			if (ret)
 689				pr_warn("Failed to re-register probe %s on %s: %d\n",
 690					trace_probe_name(&tk->tp),
 691					mod->name, ret);
 692		}
 693	}
 694	mutex_unlock(&event_mutex);
 695
 696	return NOTIFY_DONE;
 697}
 698
 699static struct notifier_block trace_kprobe_module_nb = {
 700	.notifier_call = trace_kprobe_module_callback,
 701	.priority = 1	/* Invoked after kprobe module callback */
 702};
 703
 704/* Convert certain expected symbols into '_' when generating event names */
 705static inline void sanitize_event_name(char *name)
 706{
 707	while (*name++ != '\0')
 708		if (*name == ':' || *name == '.')
 709			*name = '_';
 710}
 711
 712static int trace_kprobe_create(int argc, const char *argv[])
 713{
 714	/*
 715	 * Argument syntax:
 716	 *  - Add kprobe:
 717	 *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 718	 *  - Add kretprobe:
 719	 *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 720	 * Fetch args:
 721	 *  $retval	: fetch return value
 722	 *  $stack	: fetch stack address
 723	 *  $stackN	: fetch Nth of stack (N:0-)
 724	 *  $comm       : fetch current task comm
 725	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
 726	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 727	 *  %REG	: fetch register REG
 728	 * Dereferencing memory fetch:
 729	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 730	 * Alias name of args:
 731	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 732	 * Type of args:
 733	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 734	 */
 735	struct trace_kprobe *tk = NULL;
 736	int i, len, ret = 0;
 737	bool is_return = false;
 738	char *symbol = NULL, *tmp = NULL;
 739	const char *event = NULL, *group = KPROBE_EVENT_SYSTEM;
 740	int maxactive = 0;
 741	long offset = 0;
 742	void *addr = NULL;
 743	char buf[MAX_EVENT_NAME_LEN];
 744	unsigned int flags = TPARG_FL_KERNEL;
 745
 746	switch (argv[0][0]) {
 747	case 'r':
 
 
 748		is_return = true;
 749		flags |= TPARG_FL_RETURN;
 750		break;
 751	case 'p':
 752		break;
 753	default:
 754		return -ECANCELED;
 755	}
 756	if (argc < 2)
 757		return -ECANCELED;
 758
 759	trace_probe_log_init("trace_kprobe", argc, argv);
 760
 761	event = strchr(&argv[0][1], ':');
 762	if (event)
 763		event++;
 764
 765	if (isdigit(argv[0][1])) {
 766		if (!is_return) {
 767			trace_probe_log_err(1, MAXACT_NO_KPROBE);
 768			goto parse_error;
 769		}
 770		if (event)
 771			len = event - &argv[0][1] - 1;
 772		else
 773			len = strlen(&argv[0][1]);
 774		if (len > MAX_EVENT_NAME_LEN - 1) {
 775			trace_probe_log_err(1, BAD_MAXACT);
 776			goto parse_error;
 777		}
 778		memcpy(buf, &argv[0][1], len);
 779		buf[len] = '\0';
 780		ret = kstrtouint(buf, 0, &maxactive);
 781		if (ret || !maxactive) {
 782			trace_probe_log_err(1, BAD_MAXACT);
 783			goto parse_error;
 
 
 784		}
 785		/* kretprobes instances are iterated over via a list. The
 786		 * maximum should stay reasonable.
 787		 */
 788		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 789			trace_probe_log_err(1, MAXACT_TOO_BIG);
 790			goto parse_error;
 791		}
 
 
 
 
 
 
 792	}
 793
 794	/* try to parse an address. if that fails, try to read the
 795	 * input as a symbol. */
 796	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 797		trace_probe_log_set_index(1);
 798		/* Check whether uprobe event specified */
 799		if (strchr(argv[1], '/') && strchr(argv[1], ':')) {
 800			ret = -ECANCELED;
 801			goto error;
 
 
 
 
 
 
 802		}
 
 803		/* a symbol specified */
 804		symbol = kstrdup(argv[1], GFP_KERNEL);
 805		if (!symbol)
 806			return -ENOMEM;
 807		/* TODO: support .init module functions */
 808		ret = traceprobe_split_symbol_offset(symbol, &offset);
 809		if (ret || offset < 0 || offset > UINT_MAX) {
 810			trace_probe_log_err(0, BAD_PROBE_ADDR);
 811			goto parse_error;
 812		}
 813		if (kprobe_on_func_entry(NULL, symbol, offset))
 814			flags |= TPARG_FL_FENTRY;
 815		if (offset && is_return && !(flags & TPARG_FL_FENTRY)) {
 816			trace_probe_log_err(0, BAD_RETPROBE);
 817			goto parse_error;
 818		}
 819	}
 
 820
 821	trace_probe_log_set_index(0);
 822	if (event) {
 823		ret = traceprobe_parse_event_name(&event, &group, buf,
 824						  event - argv[0]);
 825		if (ret)
 826			goto parse_error;
 827	} else {
 828		/* Make a new event name */
 829		if (symbol)
 830			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 831				 is_return ? 'r' : 'p', symbol, offset);
 832		else
 833			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 834				 is_return ? 'r' : 'p', addr);
 835		sanitize_event_name(buf);
 836		event = buf;
 837	}
 838
 839	/* setup a probe */
 840	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 841			       argc - 2, is_return);
 842	if (IS_ERR(tk)) {
 843		ret = PTR_ERR(tk);
 844		/* This must return -ENOMEM, else there is a bug */
 845		WARN_ON_ONCE(ret != -ENOMEM);
 846		goto out;	/* We know tk is not allocated */
 847	}
 848	argc -= 2; argv += 2;
 849
 850	/* parse arguments */
 
 851	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 852		tmp = kstrdup(argv[i], GFP_KERNEL);
 853		if (!tmp) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 854			ret = -ENOMEM;
 855			goto error;
 856		}
 857
 858		trace_probe_log_set_index(i + 2);
 859		ret = traceprobe_parse_probe_arg(&tk->tp, i, tmp, flags);
 860		kfree(tmp);
 861		if (ret)
 862			goto error;	/* This can be -ENOMEM */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 863	}
 864
 865	ret = traceprobe_set_print_fmt(&tk->tp, is_return);
 866	if (ret < 0)
 867		goto error;
 868
 869	ret = register_trace_kprobe(tk);
 870	if (ret) {
 871		trace_probe_log_set_index(1);
 872		if (ret == -EILSEQ)
 873			trace_probe_log_err(0, BAD_INSN_BNDRY);
 874		else if (ret == -ENOENT)
 875			trace_probe_log_err(0, BAD_PROBE_ADDR);
 876		else if (ret != -ENOMEM && ret != -EEXIST)
 877			trace_probe_log_err(0, FAIL_REG_PROBE);
 878		goto error;
 879	}
 880
 881out:
 882	trace_probe_log_clear();
 883	kfree(symbol);
 884	return ret;
 885
 886parse_error:
 887	ret = -EINVAL;
 888error:
 889	free_trace_kprobe(tk);
 890	goto out;
 891}
 892
 893static int create_or_delete_trace_kprobe(int argc, char **argv)
 894{
 895	int ret;
 896
 897	if (argv[0][0] == '-')
 898		return dyn_event_release(argc, argv, &trace_kprobe_ops);
 899
 900	ret = trace_kprobe_create(argc, (const char **)argv);
 901	return ret == -ECANCELED ? -EINVAL : ret;
 902}
 903
 904static int trace_kprobe_run_command(struct dynevent_cmd *cmd)
 905{
 906	return trace_run_command(cmd->seq.buffer, create_or_delete_trace_kprobe);
 907}
 908
 909/**
 910 * kprobe_event_cmd_init - Initialize a kprobe event command object
 911 * @cmd: A pointer to the dynevent_cmd struct representing the new event
 912 * @buf: A pointer to the buffer used to build the command
 913 * @maxlen: The length of the buffer passed in @buf
 914 *
 915 * Initialize a synthetic event command object.  Use this before
 916 * calling any of the other kprobe_event functions.
 917 */
 918void kprobe_event_cmd_init(struct dynevent_cmd *cmd, char *buf, int maxlen)
 919{
 920	dynevent_cmd_init(cmd, buf, maxlen, DYNEVENT_TYPE_KPROBE,
 921			  trace_kprobe_run_command);
 922}
 923EXPORT_SYMBOL_GPL(kprobe_event_cmd_init);
 924
 925/**
 926 * __kprobe_event_gen_cmd_start - Generate a kprobe event command from arg list
 927 * @cmd: A pointer to the dynevent_cmd struct representing the new event
 928 * @name: The name of the kprobe event
 929 * @loc: The location of the kprobe event
 930 * @kretprobe: Is this a return probe?
 931 * @args: Variable number of arg (pairs), one pair for each field
 932 *
 933 * NOTE: Users normally won't want to call this function directly, but
 934 * rather use the kprobe_event_gen_cmd_start() wrapper, which automatically
 935 * adds a NULL to the end of the arg list.  If this function is used
 936 * directly, make sure the last arg in the variable arg list is NULL.
 937 *
 938 * Generate a kprobe event command to be executed by
 939 * kprobe_event_gen_cmd_end().  This function can be used to generate the
 940 * complete command or only the first part of it; in the latter case,
 941 * kprobe_event_add_fields() can be used to add more fields following this.
 942 *
 943 * Unlikely the synth_event_gen_cmd_start(), @loc must be specified. This
 944 * returns -EINVAL if @loc == NULL.
 945 *
 946 * Return: 0 if successful, error otherwise.
 947 */
 948int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd, bool kretprobe,
 949				 const char *name, const char *loc, ...)
 950{
 951	char buf[MAX_EVENT_NAME_LEN];
 952	struct dynevent_arg arg;
 953	va_list args;
 954	int ret;
 955
 956	if (cmd->type != DYNEVENT_TYPE_KPROBE)
 957		return -EINVAL;
 958
 959	if (!loc)
 960		return -EINVAL;
 961
 962	if (kretprobe)
 963		snprintf(buf, MAX_EVENT_NAME_LEN, "r:kprobes/%s", name);
 964	else
 965		snprintf(buf, MAX_EVENT_NAME_LEN, "p:kprobes/%s", name);
 966
 967	ret = dynevent_str_add(cmd, buf);
 968	if (ret)
 969		return ret;
 970
 971	dynevent_arg_init(&arg, 0);
 972	arg.str = loc;
 973	ret = dynevent_arg_add(cmd, &arg, NULL);
 974	if (ret)
 975		return ret;
 976
 977	va_start(args, loc);
 978	for (;;) {
 979		const char *field;
 980
 981		field = va_arg(args, const char *);
 982		if (!field)
 983			break;
 984
 985		if (++cmd->n_fields > MAX_TRACE_ARGS) {
 986			ret = -EINVAL;
 987			break;
 988		}
 989
 990		arg.str = field;
 991		ret = dynevent_arg_add(cmd, &arg, NULL);
 
 992		if (ret)
 993			break;
 
 994	}
 995	va_end(args);
 
 
 996
 997	return ret;
 998}
 999EXPORT_SYMBOL_GPL(__kprobe_event_gen_cmd_start);
1000
1001/**
1002 * __kprobe_event_add_fields - Add probe fields to a kprobe command from arg list
1003 * @cmd: A pointer to the dynevent_cmd struct representing the new event
1004 * @args: Variable number of arg (pairs), one pair for each field
1005 *
1006 * NOTE: Users normally won't want to call this function directly, but
1007 * rather use the kprobe_event_add_fields() wrapper, which
1008 * automatically adds a NULL to the end of the arg list.  If this
1009 * function is used directly, make sure the last arg in the variable
1010 * arg list is NULL.
1011 *
1012 * Add probe fields to an existing kprobe command using a variable
1013 * list of args.  Fields are added in the same order they're listed.
1014 *
1015 * Return: 0 if successful, error otherwise.
1016 */
1017int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...)
1018{
1019	struct dynevent_arg arg;
1020	va_list args;
1021	int ret = 0;
1022
1023	if (cmd->type != DYNEVENT_TYPE_KPROBE)
1024		return -EINVAL;
1025
1026	dynevent_arg_init(&arg, 0);
1027
1028	va_start(args, cmd);
1029	for (;;) {
1030		const char *field;
1031
1032		field = va_arg(args, const char *);
1033		if (!field)
1034			break;
1035
1036		if (++cmd->n_fields > MAX_TRACE_ARGS) {
1037			ret = -EINVAL;
1038			break;
1039		}
1040
1041		arg.str = field;
1042		ret = dynevent_arg_add(cmd, &arg, NULL);
1043		if (ret)
1044			break;
1045	}
1046	va_end(args);
1047
1048	return ret;
1049}
1050EXPORT_SYMBOL_GPL(__kprobe_event_add_fields);
1051
1052/**
1053 * kprobe_event_delete - Delete a kprobe event
1054 * @name: The name of the kprobe event to delete
1055 *
1056 * Delete a kprobe event with the give @name from kernel code rather
1057 * than directly from the command line.
1058 *
1059 * Return: 0 if successful, error otherwise.
1060 */
1061int kprobe_event_delete(const char *name)
1062{
1063	char buf[MAX_EVENT_NAME_LEN];
1064
1065	snprintf(buf, MAX_EVENT_NAME_LEN, "-:%s", name);
1066
1067	return trace_run_command(buf, create_or_delete_trace_kprobe);
1068}
1069EXPORT_SYMBOL_GPL(kprobe_event_delete);
1070
1071static int trace_kprobe_release(struct dyn_event *ev)
1072{
1073	struct trace_kprobe *tk = to_trace_kprobe(ev);
1074	int ret = unregister_trace_kprobe(tk);
1075
1076	if (!ret)
1077		free_trace_kprobe(tk);
1078	return ret;
1079}
1080
1081static int trace_kprobe_show(struct seq_file *m, struct dyn_event *ev)
1082{
1083	struct trace_kprobe *tk = to_trace_kprobe(ev);
1084	int i;
1085
1086	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
1087	if (trace_kprobe_is_return(tk) && tk->rp.maxactive)
1088		seq_printf(m, "%d", tk->rp.maxactive);
1089	seq_printf(m, ":%s/%s", trace_probe_group_name(&tk->tp),
1090				trace_probe_name(&tk->tp));
1091
1092	if (!tk->symbol)
1093		seq_printf(m, " 0x%p", tk->rp.kp.addr);
1094	else if (tk->rp.kp.offset)
1095		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
1096			   tk->rp.kp.offset);
1097	else
1098		seq_printf(m, " %s", trace_kprobe_symbol(tk));
1099
1100	for (i = 0; i < tk->tp.nr_args; i++)
1101		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
1102	seq_putc(m, '\n');
1103
1104	return 0;
1105}
1106
1107static int probes_seq_show(struct seq_file *m, void *v)
1108{
1109	struct dyn_event *ev = v;
1110
1111	if (!is_trace_kprobe(ev))
1112		return 0;
1113
1114	return trace_kprobe_show(m, ev);
1115}
1116
1117static const struct seq_operations probes_seq_op = {
1118	.start  = dyn_event_seq_start,
1119	.next   = dyn_event_seq_next,
1120	.stop   = dyn_event_seq_stop,
1121	.show   = probes_seq_show
1122};
1123
1124static int probes_open(struct inode *inode, struct file *file)
1125{
1126	int ret;
1127
1128	ret = security_locked_down(LOCKDOWN_TRACEFS);
1129	if (ret)
1130		return ret;
1131
1132	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
1133		ret = dyn_events_release_all(&trace_kprobe_ops);
1134		if (ret < 0)
1135			return ret;
1136	}
1137
1138	return seq_open(file, &probes_seq_op);
1139}
1140
1141static ssize_t probes_write(struct file *file, const char __user *buffer,
1142			    size_t count, loff_t *ppos)
1143{
1144	return trace_parse_run_command(file, buffer, count, ppos,
1145				       create_or_delete_trace_kprobe);
1146}
1147
1148static const struct file_operations kprobe_events_ops = {
1149	.owner          = THIS_MODULE,
1150	.open           = probes_open,
1151	.read           = seq_read,
1152	.llseek         = seq_lseek,
1153	.release        = seq_release,
1154	.write		= probes_write,
1155};
1156
1157/* Probes profiling interfaces */
1158static int probes_profile_seq_show(struct seq_file *m, void *v)
1159{
1160	struct dyn_event *ev = v;
1161	struct trace_kprobe *tk;
 
1162
1163	if (!is_trace_kprobe(ev))
1164		return 0;
1165
1166	tk = to_trace_kprobe(ev);
1167	seq_printf(m, "  %-44s %15lu %15lu\n",
1168		   trace_probe_name(&tk->tp),
1169		   trace_kprobe_nhit(tk),
1170		   tk->rp.kp.nmissed);
1171
1172	return 0;
1173}
1174
1175static const struct seq_operations profile_seq_op = {
1176	.start  = dyn_event_seq_start,
1177	.next   = dyn_event_seq_next,
1178	.stop   = dyn_event_seq_stop,
1179	.show   = probes_profile_seq_show
1180};
1181
1182static int profile_open(struct inode *inode, struct file *file)
1183{
1184	int ret;
1185
1186	ret = security_locked_down(LOCKDOWN_TRACEFS);
1187	if (ret)
1188		return ret;
1189
1190	return seq_open(file, &profile_seq_op);
1191}
1192
1193static const struct file_operations kprobe_profile_ops = {
1194	.owner          = THIS_MODULE,
1195	.open           = profile_open,
1196	.read           = seq_read,
1197	.llseek         = seq_lseek,
1198	.release        = seq_release,
1199};
1200
1201/* Kprobe specific fetch functions */
1202
1203/* Return the length of string -- including null terminal byte */
1204static nokprobe_inline int
1205fetch_store_strlen_user(unsigned long addr)
1206{
1207	const void __user *uaddr =  (__force const void __user *)addr;
1208
1209	return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
1210}
1211
1212/* Return the length of string -- including null terminal byte */
1213static nokprobe_inline int
1214fetch_store_strlen(unsigned long addr)
1215{
1216	int ret, len = 0;
1217	u8 c;
1218
1219#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1220	if (addr < TASK_SIZE)
1221		return fetch_store_strlen_user(addr);
1222#endif
1223
1224	do {
1225		ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
1226		len++;
1227	} while (c && ret == 0 && len < MAX_STRING_SIZE);
1228
1229	return (ret < 0) ? ret : len;
1230}
1231
1232/*
1233 * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
1234 * with max length and relative data location.
1235 */
1236static nokprobe_inline int
1237fetch_store_string_user(unsigned long addr, void *dest, void *base)
1238{
1239	const void __user *uaddr =  (__force const void __user *)addr;
1240	int maxlen = get_loc_len(*(u32 *)dest);
1241	void *__dest;
1242	long ret;
1243
1244	if (unlikely(!maxlen))
1245		return -ENOMEM;
1246
1247	__dest = get_loc_data(dest, base);
1248
1249	ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
1250	if (ret >= 0)
1251		*(u32 *)dest = make_data_loc(ret, __dest - base);
1252
1253	return ret;
1254}
1255
1256/*
1257 * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
1258 * length and relative data location.
1259 */
1260static nokprobe_inline int
1261fetch_store_string(unsigned long addr, void *dest, void *base)
1262{
1263	int maxlen = get_loc_len(*(u32 *)dest);
1264	void *__dest;
1265	long ret;
1266
1267#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1268	if ((unsigned long)addr < TASK_SIZE)
1269		return fetch_store_string_user(addr, dest, base);
1270#endif
1271
1272	if (unlikely(!maxlen))
1273		return -ENOMEM;
1274
1275	__dest = get_loc_data(dest, base);
1276
1277	/*
1278	 * Try to get string again, since the string can be changed while
1279	 * probing.
1280	 */
1281	ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
1282	if (ret >= 0)
1283		*(u32 *)dest = make_data_loc(ret, __dest - base);
1284
1285	return ret;
1286}
1287
1288static nokprobe_inline int
1289probe_mem_read_user(void *dest, void *src, size_t size)
1290{
1291	const void __user *uaddr =  (__force const void __user *)src;
1292
1293	return copy_from_user_nofault(dest, uaddr, size);
1294}
1295
1296static nokprobe_inline int
1297probe_mem_read(void *dest, void *src, size_t size)
1298{
1299#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1300	if ((unsigned long)src < TASK_SIZE)
1301		return probe_mem_read_user(dest, src, size);
1302#endif
1303	return copy_from_kernel_nofault(dest, src, size);
1304}
1305
1306/* Note that we don't verify it, since the code does not come from user space */
1307static int
1308process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
1309		   void *base)
1310{
1311	unsigned long val;
1312
1313retry:
1314	/* 1st stage: get value from context */
1315	switch (code->op) {
1316	case FETCH_OP_REG:
1317		val = regs_get_register(regs, code->param);
1318		break;
1319	case FETCH_OP_STACK:
1320		val = regs_get_kernel_stack_nth(regs, code->param);
1321		break;
1322	case FETCH_OP_STACKP:
1323		val = kernel_stack_pointer(regs);
1324		break;
1325	case FETCH_OP_RETVAL:
1326		val = regs_return_value(regs);
1327		break;
1328	case FETCH_OP_IMM:
1329		val = code->immediate;
1330		break;
1331	case FETCH_OP_COMM:
1332		val = (unsigned long)current->comm;
1333		break;
1334	case FETCH_OP_DATA:
1335		val = (unsigned long)code->data;
1336		break;
1337#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
1338	case FETCH_OP_ARG:
1339		val = regs_get_kernel_argument(regs, code->param);
1340		break;
1341#endif
1342	case FETCH_NOP_SYMBOL:	/* Ignore a place holder */
1343		code++;
1344		goto retry;
1345	default:
1346		return -EILSEQ;
1347	}
1348	code++;
1349
1350	return process_fetch_insn_bottom(code, val, dest, base);
1351}
1352NOKPROBE_SYMBOL(process_fetch_insn)
1353
1354/* Kprobe handler */
1355static nokprobe_inline void
1356__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1357		    struct trace_event_file *trace_file)
1358{
1359	struct kprobe_trace_entry_head *entry;
1360	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1361	struct trace_event_buffer fbuffer;
1362	int dsize;
 
 
1363
1364	WARN_ON(call != trace_file->event_call);
1365
1366	if (trace_trigger_soft_disabled(trace_file))
1367		return;
1368
1369	local_save_flags(fbuffer.flags);
1370	fbuffer.pc = preempt_count();
1371	fbuffer.trace_file = trace_file;
1372
1373	dsize = __get_data_size(&tk->tp, regs);
 
1374
1375	fbuffer.event =
1376		trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1377					call->event.type,
1378					sizeof(*entry) + tk->tp.size + dsize,
1379					fbuffer.flags, fbuffer.pc);
1380	if (!fbuffer.event)
1381		return;
1382
1383	fbuffer.regs = regs;
1384	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1385	entry->ip = (unsigned long)tk->rp.kp.addr;
1386	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1387
1388	trace_event_buffer_commit(&fbuffer);
 
1389}
1390
1391static void
1392kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1393{
1394	struct event_file_link *link;
1395
1396	trace_probe_for_each_link_rcu(link, &tk->tp)
1397		__kprobe_trace_func(tk, regs, link->file);
1398}
1399NOKPROBE_SYMBOL(kprobe_trace_func);
1400
1401/* Kretprobe handler */
1402static nokprobe_inline void
1403__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1404		       struct pt_regs *regs,
1405		       struct trace_event_file *trace_file)
1406{
1407	struct kretprobe_trace_entry_head *entry;
1408	struct trace_event_buffer fbuffer;
1409	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
1410	int dsize;
 
 
1411
1412	WARN_ON(call != trace_file->event_call);
1413
1414	if (trace_trigger_soft_disabled(trace_file))
1415		return;
1416
1417	local_save_flags(fbuffer.flags);
1418	fbuffer.pc = preempt_count();
1419	fbuffer.trace_file = trace_file;
1420
1421	dsize = __get_data_size(&tk->tp, regs);
1422	fbuffer.event =
1423		trace_event_buffer_lock_reserve(&fbuffer.buffer, trace_file,
1424					call->event.type,
1425					sizeof(*entry) + tk->tp.size + dsize,
1426					fbuffer.flags, fbuffer.pc);
1427	if (!fbuffer.event)
1428		return;
1429
1430	fbuffer.regs = regs;
1431	entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
1432	entry->func = (unsigned long)tk->rp.kp.addr;
1433	entry->ret_ip = (unsigned long)ri->ret_addr;
1434	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1435
1436	trace_event_buffer_commit(&fbuffer);
 
1437}
1438
1439static void
1440kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1441		     struct pt_regs *regs)
1442{
1443	struct event_file_link *link;
1444
1445	trace_probe_for_each_link_rcu(link, &tk->tp)
1446		__kretprobe_trace_func(tk, ri, regs, link->file);
1447}
1448NOKPROBE_SYMBOL(kretprobe_trace_func);
1449
1450/* Event entry printers */
1451static enum print_line_t
1452print_kprobe_event(struct trace_iterator *iter, int flags,
1453		   struct trace_event *event)
1454{
1455	struct kprobe_trace_entry_head *field;
1456	struct trace_seq *s = &iter->seq;
1457	struct trace_probe *tp;
 
 
1458
1459	field = (struct kprobe_trace_entry_head *)iter->ent;
1460	tp = trace_probe_primary_from_call(
1461		container_of(event, struct trace_event_call, event));
1462	if (WARN_ON_ONCE(!tp))
1463		goto out;
1464
1465	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1466
1467	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1468		goto out;
1469
1470	trace_seq_putc(s, ')');
1471
1472	if (print_probe_args(s, tp->args, tp->nr_args,
1473			     (u8 *)&field[1], field) < 0)
1474		goto out;
 
 
1475
1476	trace_seq_putc(s, '\n');
1477 out:
1478	return trace_handle_return(s);
1479}
1480
1481static enum print_line_t
1482print_kretprobe_event(struct trace_iterator *iter, int flags,
1483		      struct trace_event *event)
1484{
1485	struct kretprobe_trace_entry_head *field;
1486	struct trace_seq *s = &iter->seq;
1487	struct trace_probe *tp;
 
 
1488
1489	field = (struct kretprobe_trace_entry_head *)iter->ent;
1490	tp = trace_probe_primary_from_call(
1491		container_of(event, struct trace_event_call, event));
1492	if (WARN_ON_ONCE(!tp))
1493		goto out;
1494
1495	trace_seq_printf(s, "%s: (", trace_probe_name(tp));
1496
1497	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1498		goto out;
1499
1500	trace_seq_puts(s, " <- ");
1501
1502	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1503		goto out;
1504
1505	trace_seq_putc(s, ')');
1506
1507	if (print_probe_args(s, tp->args, tp->nr_args,
1508			     (u8 *)&field[1], field) < 0)
1509		goto out;
 
 
1510
1511	trace_seq_putc(s, '\n');
1512
1513 out:
1514	return trace_handle_return(s);
1515}
1516
1517
1518static int kprobe_event_define_fields(struct trace_event_call *event_call)
1519{
1520	int ret;
1521	struct kprobe_trace_entry_head field;
1522	struct trace_probe *tp;
1523
1524	tp = trace_probe_primary_from_call(event_call);
1525	if (WARN_ON_ONCE(!tp))
1526		return -ENOENT;
1527
1528	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
 
 
 
1529
1530	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
 
 
 
 
 
 
 
 
 
1531}
1532
1533static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1534{
1535	int ret;
1536	struct kretprobe_trace_entry_head field;
1537	struct trace_probe *tp;
1538
1539	tp = trace_probe_primary_from_call(event_call);
1540	if (WARN_ON_ONCE(!tp))
1541		return -ENOENT;
1542
1543	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1544	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
 
 
 
1545
1546	return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
 
 
 
 
 
 
 
 
 
1547}
1548
1549#ifdef CONFIG_PERF_EVENTS
1550
1551/* Kprobe profile handler */
1552static int
1553kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1554{
1555	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
 
1556	struct kprobe_trace_entry_head *entry;
1557	struct hlist_head *head;
1558	int size, __size, dsize;
1559	int rctx;
1560
1561	if (bpf_prog_array_valid(call)) {
1562		unsigned long orig_ip = instruction_pointer(regs);
1563		int ret;
1564
1565		ret = trace_call_bpf(call, regs);
1566
1567		/*
1568		 * We need to check and see if we modified the pc of the
1569		 * pt_regs, and if so return 1 so that we don't do the
1570		 * single stepping.
1571		 */
1572		if (orig_ip != instruction_pointer(regs))
1573			return 1;
1574		if (!ret)
1575			return 0;
1576	}
1577
1578	head = this_cpu_ptr(call->perf_events);
1579	if (hlist_empty(head))
1580		return 0;
1581
1582	dsize = __get_data_size(&tk->tp, regs);
1583	__size = sizeof(*entry) + tk->tp.size + dsize;
1584	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1585	size -= sizeof(u32);
1586
1587	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1588	if (!entry)
1589		return 0;
1590
1591	entry->ip = (unsigned long)tk->rp.kp.addr;
1592	memset(&entry[1], 0, dsize);
1593	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1594	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1595			      head, NULL);
1596	return 0;
1597}
1598NOKPROBE_SYMBOL(kprobe_perf_func);
1599
1600/* Kretprobe profile handler */
1601static void
1602kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1603		    struct pt_regs *regs)
1604{
1605	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
 
1606	struct kretprobe_trace_entry_head *entry;
1607	struct hlist_head *head;
1608	int size, __size, dsize;
1609	int rctx;
1610
1611	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1612		return;
1613
1614	head = this_cpu_ptr(call->perf_events);
1615	if (hlist_empty(head))
1616		return;
1617
1618	dsize = __get_data_size(&tk->tp, regs);
1619	__size = sizeof(*entry) + tk->tp.size + dsize;
1620	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1621	size -= sizeof(u32);
1622
1623	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1624	if (!entry)
1625		return;
1626
1627	entry->func = (unsigned long)tk->rp.kp.addr;
1628	entry->ret_ip = (unsigned long)ri->ret_addr;
1629	store_trace_args(&entry[1], &tk->tp, regs, sizeof(*entry), dsize);
1630	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1631			      head, NULL);
1632}
1633NOKPROBE_SYMBOL(kretprobe_perf_func);
1634
1635int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1636			const char **symbol, u64 *probe_offset,
1637			u64 *probe_addr, bool perf_type_tracepoint)
1638{
1639	const char *pevent = trace_event_name(event->tp_event);
1640	const char *group = event->tp_event->class->system;
1641	struct trace_kprobe *tk;
1642
1643	if (perf_type_tracepoint)
1644		tk = find_trace_kprobe(pevent, group);
1645	else
1646		tk = trace_kprobe_primary_from_call(event->tp_event);
1647	if (!tk)
1648		return -EINVAL;
1649
1650	*fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1651					      : BPF_FD_TYPE_KPROBE;
1652	if (tk->symbol) {
1653		*symbol = tk->symbol;
1654		*probe_offset = tk->rp.kp.offset;
1655		*probe_addr = 0;
1656	} else {
1657		*symbol = NULL;
1658		*probe_offset = 0;
1659		*probe_addr = (unsigned long)tk->rp.kp.addr;
1660	}
1661	return 0;
1662}
1663#endif	/* CONFIG_PERF_EVENTS */
1664
1665/*
1666 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1667 *
1668 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1669 * lockless, but we can't race with this __init function.
1670 */
1671static int kprobe_register(struct trace_event_call *event,
1672			   enum trace_reg type, void *data)
1673{
 
1674	struct trace_event_file *file = data;
1675
1676	switch (type) {
1677	case TRACE_REG_REGISTER:
1678		return enable_trace_kprobe(event, file);
1679	case TRACE_REG_UNREGISTER:
1680		return disable_trace_kprobe(event, file);
1681
1682#ifdef CONFIG_PERF_EVENTS
1683	case TRACE_REG_PERF_REGISTER:
1684		return enable_trace_kprobe(event, NULL);
1685	case TRACE_REG_PERF_UNREGISTER:
1686		return disable_trace_kprobe(event, NULL);
1687	case TRACE_REG_PERF_OPEN:
1688	case TRACE_REG_PERF_CLOSE:
1689	case TRACE_REG_PERF_ADD:
1690	case TRACE_REG_PERF_DEL:
1691		return 0;
1692#endif
1693	}
1694	return 0;
1695}
1696
1697static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1698{
1699	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1700	int ret = 0;
1701
1702	raw_cpu_inc(*tk->nhit);
1703
1704	if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1705		kprobe_trace_func(tk, regs);
1706#ifdef CONFIG_PERF_EVENTS
1707	if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1708		ret = kprobe_perf_func(tk, regs);
1709#endif
1710	return ret;
1711}
1712NOKPROBE_SYMBOL(kprobe_dispatcher);
1713
1714static int
1715kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1716{
1717	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1718
1719	raw_cpu_inc(*tk->nhit);
1720
1721	if (trace_probe_test_flag(&tk->tp, TP_FLAG_TRACE))
1722		kretprobe_trace_func(tk, ri, regs);
1723#ifdef CONFIG_PERF_EVENTS
1724	if (trace_probe_test_flag(&tk->tp, TP_FLAG_PROFILE))
1725		kretprobe_perf_func(tk, ri, regs);
1726#endif
1727	return 0;	/* We don't tweek kernel, so just return 0 */
1728}
1729NOKPROBE_SYMBOL(kretprobe_dispatcher);
1730
1731static struct trace_event_functions kretprobe_funcs = {
1732	.trace		= print_kretprobe_event
1733};
1734
1735static struct trace_event_functions kprobe_funcs = {
1736	.trace		= print_kprobe_event
1737};
1738
1739static struct trace_event_fields kretprobe_fields_array[] = {
1740	{ .type = TRACE_FUNCTION_TYPE,
1741	  .define_fields = kretprobe_event_define_fields },
1742	{}
1743};
1744
1745static struct trace_event_fields kprobe_fields_array[] = {
1746	{ .type = TRACE_FUNCTION_TYPE,
1747	  .define_fields = kprobe_event_define_fields },
1748	{}
1749};
1750
1751static inline void init_trace_event_call(struct trace_kprobe *tk)
1752{
1753	struct trace_event_call *call = trace_probe_event_call(&tk->tp);
 
1754
 
 
1755	if (trace_kprobe_is_return(tk)) {
1756		call->event.funcs = &kretprobe_funcs;
1757		call->class->fields_array = kretprobe_fields_array;
1758	} else {
1759		call->event.funcs = &kprobe_funcs;
1760		call->class->fields_array = kprobe_fields_array;
 
 
 
 
 
 
 
1761	}
1762
1763	call->flags = TRACE_EVENT_FL_KPROBE;
1764	call->class->reg = kprobe_register;
1765}
1766
1767static int register_kprobe_event(struct trace_kprobe *tk)
1768{
1769	init_trace_event_call(tk);
1770
1771	return trace_probe_register_event_call(&tk->tp);
 
 
1772}
1773
1774static int unregister_kprobe_event(struct trace_kprobe *tk)
1775{
1776	return trace_probe_unregister_event_call(&tk->tp);
1777}
1778
1779#ifdef CONFIG_PERF_EVENTS
1780/* create a trace_kprobe, but don't add it to global lists */
1781struct trace_event_call *
1782create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1783			  bool is_return)
1784{
1785	struct trace_kprobe *tk;
1786	int ret;
1787	char *event;
1788
1789	/*
1790	 * local trace_kprobes are not added to dyn_event, so they are never
1791	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1792	 * duplicated name here.
1793	 */
1794	event = func ? func : "DUMMY_EVENT";
1795
1796	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1797				offs, 0 /* maxactive */, 0 /* nargs */,
1798				is_return);
1799
1800	if (IS_ERR(tk)) {
1801		pr_info("Failed to allocate trace_probe.(%d)\n",
1802			(int)PTR_ERR(tk));
1803		return ERR_CAST(tk);
1804	}
1805
1806	init_trace_event_call(tk);
1807
1808	if (traceprobe_set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1809		ret = -ENOMEM;
1810		goto error;
1811	}
1812
1813	ret = __register_trace_kprobe(tk);
1814	if (ret < 0)
1815		goto error;
1816
1817	return trace_probe_event_call(&tk->tp);
1818error:
1819	free_trace_kprobe(tk);
1820	return ERR_PTR(ret);
1821}
1822
1823void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1824{
1825	struct trace_kprobe *tk;
1826
1827	tk = trace_kprobe_primary_from_call(event_call);
1828	if (unlikely(!tk))
1829		return;
1830
1831	if (trace_probe_is_enabled(&tk->tp)) {
1832		WARN_ON(1);
1833		return;
1834	}
1835
1836	__unregister_trace_kprobe(tk);
1837
1838	free_trace_kprobe(tk);
1839}
1840#endif /* CONFIG_PERF_EVENTS */
1841
1842static __init void enable_boot_kprobe_events(void)
1843{
1844	struct trace_array *tr = top_trace_array();
1845	struct trace_event_file *file;
1846	struct trace_kprobe *tk;
1847	struct dyn_event *pos;
1848
1849	mutex_lock(&event_mutex);
1850	for_each_trace_kprobe(tk, pos) {
1851		list_for_each_entry(file, &tr->events, list)
1852			if (file->event_call == trace_probe_event_call(&tk->tp))
1853				trace_event_enable_disable(file, 1, 0);
1854	}
1855	mutex_unlock(&event_mutex);
1856}
1857
1858static __init void setup_boot_kprobe_events(void)
1859{
1860	char *p, *cmd = kprobe_boot_events_buf;
1861	int ret;
1862
1863	strreplace(kprobe_boot_events_buf, ',', ' ');
1864
1865	while (cmd && *cmd != '\0') {
1866		p = strchr(cmd, ';');
1867		if (p)
1868			*p++ = '\0';
1869
1870		ret = trace_run_command(cmd, create_or_delete_trace_kprobe);
1871		if (ret)
1872			pr_warn("Failed to add event(%d): %s\n", ret, cmd);
1873		else
1874			kprobe_boot_events_enabled = true;
1875
1876		cmd = p;
1877	}
1878
1879	enable_boot_kprobe_events();
1880}
1881
1882/*
1883 * Register dynevent at subsys_initcall. This allows kernel to setup kprobe
1884 * events in fs_initcall without tracefs.
1885 */
1886static __init int init_kprobe_trace_early(void)
1887{
1888	int ret;
1889
1890	ret = dyn_event_register(&trace_kprobe_ops);
1891	if (ret)
1892		return ret;
1893
1894	if (register_module_notifier(&trace_kprobe_module_nb))
1895		return -EINVAL;
1896
1897	return 0;
1898}
1899subsys_initcall(init_kprobe_trace_early);
1900
1901/* Make a tracefs interface for controlling probe points */
1902static __init int init_kprobe_trace(void)
1903{
1904	struct dentry *d_tracer;
1905	struct dentry *entry;
1906
 
 
 
1907	d_tracer = tracing_init_dentry();
1908	if (IS_ERR(d_tracer))
1909		return 0;
1910
1911	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1912				    NULL, &kprobe_events_ops);
1913
1914	/* Event list interface */
1915	if (!entry)
1916		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1917
1918	/* Profile interface */
1919	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1920				    NULL, &kprobe_profile_ops);
1921
1922	if (!entry)
1923		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1924
1925	setup_boot_kprobe_events();
1926
1927	return 0;
1928}
1929fs_initcall(init_kprobe_trace);
1930
1931
1932#ifdef CONFIG_FTRACE_STARTUP_TEST
1933static __init struct trace_event_file *
 
 
 
 
 
 
 
 
 
 
 
1934find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1935{
1936	struct trace_event_file *file;
1937
1938	list_for_each_entry(file, &tr->events, list)
1939		if (file->event_call == trace_probe_event_call(&tk->tp))
1940			return file;
1941
1942	return NULL;
1943}
1944
1945/*
1946 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1947 * stage, we can do this lockless.
1948 */
1949static __init int kprobe_trace_self_tests_init(void)
1950{
1951	int ret, warn = 0;
1952	int (*target)(int, int, int, int, int, int);
1953	struct trace_kprobe *tk;
1954	struct trace_event_file *file;
1955
1956	if (tracing_is_disabled())
1957		return -ENODEV;
1958
1959	if (kprobe_boot_events_enabled) {
1960		pr_info("Skipping kprobe tests due to kprobe_event on cmdline\n");
1961		return 0;
1962	}
1963
1964	target = kprobe_trace_selftest_target;
1965
1966	pr_info("Testing kprobe tracing: ");
1967
1968	ret = trace_run_command("p:testprobe kprobe_trace_selftest_target $stack $stack0 +0($stack)",
1969				create_or_delete_trace_kprobe);
 
1970	if (WARN_ON_ONCE(ret)) {
1971		pr_warn("error on probing function entry.\n");
1972		warn++;
1973	} else {
1974		/* Enable trace point */
1975		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1976		if (WARN_ON_ONCE(tk == NULL)) {
1977			pr_warn("error on getting new probe.\n");
1978			warn++;
1979		} else {
1980			file = find_trace_probe_file(tk, top_trace_array());
1981			if (WARN_ON_ONCE(file == NULL)) {
1982				pr_warn("error on getting probe file.\n");
1983				warn++;
1984			} else
1985				enable_trace_kprobe(
1986					trace_probe_event_call(&tk->tp), file);
1987		}
1988	}
1989
1990	ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target $retval",
1991				create_or_delete_trace_kprobe);
1992	if (WARN_ON_ONCE(ret)) {
1993		pr_warn("error on probing function return.\n");
1994		warn++;
1995	} else {
1996		/* Enable trace point */
1997		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1998		if (WARN_ON_ONCE(tk == NULL)) {
1999			pr_warn("error on getting 2nd new probe.\n");
2000			warn++;
2001		} else {
2002			file = find_trace_probe_file(tk, top_trace_array());
2003			if (WARN_ON_ONCE(file == NULL)) {
2004				pr_warn("error on getting probe file.\n");
2005				warn++;
2006			} else
2007				enable_trace_kprobe(
2008					trace_probe_event_call(&tk->tp), file);
2009		}
2010	}
2011
2012	if (warn)
2013		goto end;
2014
2015	ret = target(1, 2, 3, 4, 5, 6);
2016
2017	/*
2018	 * Not expecting an error here, the check is only to prevent the
2019	 * optimizer from removing the call to target() as otherwise there
2020	 * are no side-effects and the call is never performed.
2021	 */
2022	if (ret != 21)
2023		warn++;
2024
2025	/* Disable trace points before removing it */
2026	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
2027	if (WARN_ON_ONCE(tk == NULL)) {
2028		pr_warn("error on getting test probe.\n");
2029		warn++;
2030	} else {
2031		if (trace_kprobe_nhit(tk) != 1) {
2032			pr_warn("incorrect number of testprobe hits\n");
2033			warn++;
2034		}
2035
2036		file = find_trace_probe_file(tk, top_trace_array());
2037		if (WARN_ON_ONCE(file == NULL)) {
2038			pr_warn("error on getting probe file.\n");
2039			warn++;
2040		} else
2041			disable_trace_kprobe(
2042				trace_probe_event_call(&tk->tp), file);
2043	}
2044
2045	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
2046	if (WARN_ON_ONCE(tk == NULL)) {
2047		pr_warn("error on getting 2nd test probe.\n");
2048		warn++;
2049	} else {
2050		if (trace_kprobe_nhit(tk) != 1) {
2051			pr_warn("incorrect number of testprobe2 hits\n");
2052			warn++;
2053		}
2054
2055		file = find_trace_probe_file(tk, top_trace_array());
2056		if (WARN_ON_ONCE(file == NULL)) {
2057			pr_warn("error on getting probe file.\n");
2058			warn++;
2059		} else
2060			disable_trace_kprobe(
2061				trace_probe_event_call(&tk->tp), file);
2062	}
2063
2064	ret = trace_run_command("-:testprobe", create_or_delete_trace_kprobe);
2065	if (WARN_ON_ONCE(ret)) {
2066		pr_warn("error on deleting a probe.\n");
2067		warn++;
2068	}
2069
2070	ret = trace_run_command("-:testprobe2", create_or_delete_trace_kprobe);
2071	if (WARN_ON_ONCE(ret)) {
2072		pr_warn("error on deleting a probe.\n");
2073		warn++;
2074	}
2075
2076end:
2077	ret = dyn_events_release_all(&trace_kprobe_ops);
2078	if (WARN_ON_ONCE(ret)) {
2079		pr_warn("error on cleaning up probes.\n");
2080		warn++;
2081	}
2082	/*
2083	 * Wait for the optimizer work to finish. Otherwise it might fiddle
2084	 * with probes in already freed __init text.
2085	 */
2086	wait_for_kprobe_optimizer();
2087	if (warn)
2088		pr_cont("NG: Some tests are failed. Please check them.\n");
2089	else
2090		pr_cont("OK\n");
2091	return 0;
2092}
2093
2094late_initcall(kprobe_trace_self_tests_init);
2095
2096#endif