Linux Audio

Check our new training course

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