Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v3.5.6
 
   1/*
   2 * Kprobes-based tracing events
   3 *
   4 * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5 *
   6 * This program is free software; you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License version 2 as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it will be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write to the Free Software
  17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18 */
 
  19
 
  20#include <linux/module.h>
  21#include <linux/uaccess.h>
 
 
  22
 
 
 
 
  23#include "trace_probe.h"
 
  24
  25#define KPROBE_EVENT_SYSTEM "kprobes"
 
 
  26
  27/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  28 * Kprobe event core functions
  29 */
  30
  31struct trace_probe {
  32	struct list_head	list;
  33	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
  34	unsigned long 		nhit;
  35	unsigned int		flags;	/* For TP_FLAG_* */
  36	const char		*symbol;	/* symbol name */
  37	struct ftrace_event_class	class;
  38	struct ftrace_event_call	call;
  39	ssize_t			size;		/* trace entry size */
  40	unsigned int		nr_args;
  41	struct probe_arg	args[];
  42};
  43
  44#define SIZEOF_TRACE_PROBE(n)			\
  45	(offsetof(struct trace_probe, args) +	\
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  46	(sizeof(struct probe_arg) * (n)))
  47
 
 
 
 
  48
  49static __kprobes int trace_probe_is_return(struct trace_probe *tp)
  50{
  51	return tp->rp.handler != NULL;
  52}
  53
  54static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
  55{
  56	return tp->symbol ? tp->symbol : "unknown";
  57}
  58
  59static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
  60{
  61	return tp->rp.kp.offset;
  62}
  63
  64static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
 
  65{
  66	return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
 
 
  67}
  68
  69static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
  70{
  71	return !!(tp->flags & TP_FLAG_REGISTERED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  72}
  73
  74static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
  75{
  76	return !!(kprobe_gone(&tp->rp.kp));
 
 
  77}
  78
  79static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
  80						struct module *mod)
  81{
  82	int len = strlen(mod->name);
  83	const char *name = trace_probe_symbol(tp);
  84	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  85}
  86
  87static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
  88{
  89	return !!strchr(trace_probe_symbol(tp), ':');
 
  90}
  91
  92static int register_probe_event(struct trace_probe *tp);
  93static void unregister_probe_event(struct trace_probe *tp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  94
  95static DEFINE_MUTEX(probe_lock);
  96static LIST_HEAD(probe_list);
  97
  98static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
  99static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 100				struct pt_regs *regs);
 101
 
 
 
 
 
 
 
 
 
 
 102/*
 103 * Allocate new trace_probe and initialize it (including kprobes).
 104 */
 105static struct trace_probe *alloc_trace_probe(const char *group,
 106					     const char *event,
 107					     void *addr,
 108					     const char *symbol,
 109					     unsigned long offs,
 
 110					     int nargs, bool is_return)
 111{
 112	struct trace_probe *tp;
 113	int ret = -ENOMEM;
 114
 115	tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 116	if (!tp)
 117		return ERR_PTR(ret);
 118
 
 
 
 
 119	if (symbol) {
 120		tp->symbol = kstrdup(symbol, GFP_KERNEL);
 121		if (!tp->symbol)
 122			goto error;
 123		tp->rp.kp.symbol_name = tp->symbol;
 124		tp->rp.kp.offset = offs;
 125	} else
 126		tp->rp.kp.addr = addr;
 127
 128	if (is_return)
 129		tp->rp.handler = kretprobe_dispatcher;
 130	else
 131		tp->rp.kp.pre_handler = kprobe_dispatcher;
 132
 133	if (!event || !is_good_name(event)) {
 134		ret = -EINVAL;
 135		goto error;
 136	}
 137
 138	tp->call.class = &tp->class;
 139	tp->call.name = kstrdup(event, GFP_KERNEL);
 140	if (!tp->call.name)
 141		goto error;
 142
 143	if (!group || !is_good_name(group)) {
 144		ret = -EINVAL;
 145		goto error;
 146	}
 147
 148	tp->class.system = kstrdup(group, GFP_KERNEL);
 149	if (!tp->class.system)
 150		goto error;
 151
 152	INIT_LIST_HEAD(&tp->list);
 153	return tp;
 154error:
 155	kfree(tp->call.name);
 156	kfree(tp->symbol);
 157	kfree(tp);
 158	return ERR_PTR(ret);
 159}
 160
 161static void free_trace_probe(struct trace_probe *tp)
 
 162{
 163	int i;
 
 
 
 
 
 
 
 
 164
 165	for (i = 0; i < tp->nr_args; i++)
 166		traceprobe_free_probe_arg(&tp->args[i]);
 
 167
 168	kfree(tp->call.class->system);
 169	kfree(tp->call.name);
 170	kfree(tp->symbol);
 171	kfree(tp);
 
 
 
 
 172}
 173
 174static struct trace_probe *find_trace_probe(const char *event,
 175					    const char *group)
 176{
 177	struct trace_probe *tp;
 
 178
 179	list_for_each_entry(tp, &probe_list, list)
 180		if (strcmp(tp->call.name, event) == 0 &&
 181		    strcmp(tp->call.class->system, group) == 0)
 182			return tp;
 183	return NULL;
 
 
 
 
 184}
 185
 186/* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
 187static int enable_trace_probe(struct trace_probe *tp, int flag)
 
 
 
 
 188{
 
 
 
 189	int ret = 0;
 190
 191	tp->flags |= flag;
 192	if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
 193	    !trace_probe_has_gone(tp)) {
 194		if (trace_probe_is_return(tp))
 195			ret = enable_kretprobe(&tp->rp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 196		else
 197			ret = enable_kprobe(&tp->rp.kp);
 198	}
 199
 200	return ret;
 201}
 202
 203/* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
 204static void disable_trace_probe(struct trace_probe *tp, int flag)
 
 
 
 
 205{
 206	tp->flags &= ~flag;
 207	if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
 208		if (trace_probe_is_return(tp))
 209			disable_kretprobe(&tp->rp);
 210		else
 211			disable_kprobe(&tp->rp.kp);
 212	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 213}
 
 
 
 214
 215/* Internal register function - just handle k*probes and flags */
 216static int __register_trace_probe(struct trace_probe *tp)
 217{
 218	int i, ret;
 219
 220	if (trace_probe_is_registered(tp))
 
 
 
 
 
 
 
 
 
 221		return -EINVAL;
 
 222
 223	for (i = 0; i < tp->nr_args; i++)
 224		traceprobe_update_arg(&tp->args[i]);
 
 
 
 225
 226	/* Set/clear disabled flag according to tp->flag */
 227	if (trace_probe_is_enabled(tp))
 228		tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 229	else
 230		tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 231
 232	if (trace_probe_is_return(tp))
 233		ret = register_kretprobe(&tp->rp);
 234	else
 235		ret = register_kprobe(&tp->rp.kp);
 236
 237	if (ret == 0)
 238		tp->flags |= TP_FLAG_REGISTERED;
 239	else {
 240		pr_warning("Could not insert probe at %s+%lu: %d\n",
 241			   trace_probe_symbol(tp), trace_probe_offset(tp), ret);
 242		if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
 243			pr_warning("This probe might be able to register after"
 244				   "target module is loaded. Continue.\n");
 245			ret = 0;
 246		} else if (ret == -EILSEQ) {
 247			pr_warning("Probing address(0x%p) is not an "
 248				   "instruction boundary.\n",
 249				   tp->rp.kp.addr);
 250			ret = -EINVAL;
 251		}
 252	}
 253
 254	return ret;
 255}
 256
 257/* Internal unregister function - just handle k*probes and flags */
 258static void __unregister_trace_probe(struct trace_probe *tp)
 259{
 260	if (trace_probe_is_registered(tp)) {
 261		if (trace_probe_is_return(tp))
 262			unregister_kretprobe(&tp->rp);
 263		else
 264			unregister_kprobe(&tp->rp.kp);
 265		tp->flags &= ~TP_FLAG_REGISTERED;
 266		/* Cleanup kprobe for reuse */
 267		if (tp->rp.kp.symbol_name)
 268			tp->rp.kp.addr = NULL;
 
 269	}
 270}
 271
 272/* Unregister a trace_probe and probe_event: call with locking probe_lock */
 273static int unregister_trace_probe(struct trace_probe *tp)
 274{
 
 
 
 
 275	/* Enabled event can not be unregistered */
 276	if (trace_probe_is_enabled(tp))
 
 
 
 
 277		return -EBUSY;
 278
 279	__unregister_trace_probe(tp);
 280	list_del(&tp->list);
 281	unregister_probe_event(tp);
 
 282
 283	return 0;
 284}
 285
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 286/* Register a trace_probe and probe_event */
 287static int register_trace_probe(struct trace_probe *tp)
 288{
 289	struct trace_probe *old_tp;
 290	int ret;
 291
 292	mutex_lock(&probe_lock);
 293
 294	/* Delete old (same name) event if exist */
 295	old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
 296	if (old_tp) {
 297		ret = unregister_trace_probe(old_tp);
 298		if (ret < 0)
 299			goto end;
 300		free_trace_probe(old_tp);
 
 
 
 
 301	}
 302
 303	/* Register new event */
 304	ret = register_probe_event(tp);
 305	if (ret) {
 306		pr_warning("Failed to register probe event(%d)\n", ret);
 307		goto end;
 308	}
 309
 310	/* Register k*probe */
 311	ret = __register_trace_probe(tp);
 
 
 
 
 
 312	if (ret < 0)
 313		unregister_probe_event(tp);
 314	else
 315		list_add_tail(&tp->list, &probe_list);
 316
 317end:
 318	mutex_unlock(&probe_lock);
 319	return ret;
 320}
 321
 322/* Module notifier call back, checking event on the module */
 323static int trace_probe_module_callback(struct notifier_block *nb,
 324				       unsigned long val, void *data)
 325{
 326	struct module *mod = data;
 327	struct trace_probe *tp;
 
 328	int ret;
 329
 330	if (val != MODULE_STATE_COMING)
 331		return NOTIFY_DONE;
 332
 333	/* Update probes on coming module */
 334	mutex_lock(&probe_lock);
 335	list_for_each_entry(tp, &probe_list, list) {
 336		if (trace_probe_within_module(tp, mod)) {
 337			/* Don't need to check busy - this should have gone. */
 338			__unregister_trace_probe(tp);
 339			ret = __register_trace_probe(tp);
 340			if (ret)
 341				pr_warning("Failed to re-register probe %s on"
 342					   "%s: %d\n",
 343					   tp->call.name, mod->name, ret);
 344		}
 345	}
 346	mutex_unlock(&probe_lock);
 347
 348	return NOTIFY_DONE;
 349}
 350
 351static struct notifier_block trace_probe_module_nb = {
 352	.notifier_call = trace_probe_module_callback,
 353	.priority = 1	/* Invoked after kprobe module callback */
 354};
 355
 356static int create_trace_probe(int argc, char **argv)
 
 
 
 
 
 
 
 
 357{
 358	/*
 359	 * Argument syntax:
 360	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 361	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 
 
 362	 * Fetch args:
 363	 *  $retval	: fetch return value
 364	 *  $stack	: fetch stack address
 365	 *  $stackN	: fetch Nth of stack (N:0-)
 
 366	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
 367	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 368	 *  %REG	: fetch register REG
 369	 * Dereferencing memory fetch:
 370	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 371	 * Alias name of args:
 372	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 373	 * Type of args:
 374	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 375	 */
 376	struct trace_probe *tp;
 377	int i, ret = 0;
 378	bool is_return = false, is_delete = false;
 379	char *symbol = NULL, *event = NULL, *group = NULL;
 380	char *arg;
 381	unsigned long offset = 0;
 
 382	void *addr = NULL;
 383	char buf[MAX_EVENT_NAME_LEN];
 
 384
 385	/* argc must be >= 1 */
 386	if (argv[0][0] == 'p')
 387		is_return = false;
 388	else if (argv[0][0] == 'r')
 389		is_return = true;
 390	else if (argv[0][0] == '-')
 391		is_delete = true;
 392	else {
 393		pr_info("Probe definition must be started with 'p', 'r' or"
 394			" '-'.\n");
 395		return -EINVAL;
 396	}
 397
 398	if (argv[0][1] == ':') {
 399		event = &argv[0][2];
 400		if (strchr(event, '/')) {
 401			group = event;
 402			event = strchr(group, '/') + 1;
 403			event[-1] = '\0';
 404			if (strlen(group) == 0) {
 405				pr_info("Group name is not specified\n");
 406				return -EINVAL;
 407			}
 
 
 408		}
 409		if (strlen(event) == 0) {
 410			pr_info("Event name is not specified\n");
 411			return -EINVAL;
 
 
 
 
 412		}
 413	}
 414	if (!group)
 415		group = KPROBE_EVENT_SYSTEM;
 416
 417	if (is_delete) {
 418		if (!event) {
 419			pr_info("Delete command needs an event name.\n");
 420			return -EINVAL;
 421		}
 422		mutex_lock(&probe_lock);
 423		tp = find_trace_probe(event, group);
 424		if (!tp) {
 425			mutex_unlock(&probe_lock);
 426			pr_info("Event %s/%s doesn't exist.\n", group, event);
 427			return -ENOENT;
 428		}
 429		/* delete an event */
 430		ret = unregister_trace_probe(tp);
 431		if (ret == 0)
 432			free_trace_probe(tp);
 433		mutex_unlock(&probe_lock);
 434		return ret;
 435	}
 436
 437	if (argc < 2) {
 438		pr_info("Probe point is not specified.\n");
 439		return -EINVAL;
 440	}
 441	if (isdigit(argv[1][0])) {
 442		if (is_return) {
 443			pr_info("Return probe point must be a symbol.\n");
 444			return -EINVAL;
 445		}
 446		/* an address specified */
 447		ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
 448		if (ret) {
 449			pr_info("Failed to parse address.\n");
 450			return ret;
 451		}
 452	} else {
 453		/* a symbol specified */
 454		symbol = argv[1];
 
 
 455		/* TODO: support .init module functions */
 456		ret = traceprobe_split_symbol_offset(symbol, &offset);
 457		if (ret) {
 458			pr_info("Failed to parse symbol.\n");
 459			return ret;
 460		}
 461		if (offset && is_return) {
 462			pr_info("Return probe must be used without offset.\n");
 463			return -EINVAL;
 
 
 464		}
 465	}
 466	argc -= 2; argv += 2;
 467
 468	/* setup a probe */
 469	if (!event) {
 
 
 
 
 
 470		/* Make a new event name */
 471		if (symbol)
 472			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 473				 is_return ? 'r' : 'p', symbol, offset);
 474		else
 475			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 476				 is_return ? 'r' : 'p', addr);
 
 477		event = buf;
 478	}
 479	tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
 480			       is_return);
 481	if (IS_ERR(tp)) {
 482		pr_info("Failed to allocate trace_probe.(%d)\n",
 483			(int)PTR_ERR(tp));
 484		return PTR_ERR(tp);
 
 
 
 485	}
 
 486
 487	/* parse arguments */
 488	ret = 0;
 489	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 490		/* Increment count for freeing args in error case */
 491		tp->nr_args++;
 492
 493		/* Parse argument name */
 494		arg = strchr(argv[i], '=');
 495		if (arg) {
 496			*arg++ = '\0';
 497			tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 498		} else {
 499			arg = argv[i];
 500			/* If argument name is omitted, set "argN" */
 501			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 502			tp->args[i].name = kstrdup(buf, GFP_KERNEL);
 503		}
 504
 505		if (!tp->args[i].name) {
 506			pr_info("Failed to allocate argument[%d] name.\n", i);
 507			ret = -ENOMEM;
 508			goto error;
 509		}
 510
 511		if (!is_good_name(tp->args[i].name)) {
 512			pr_info("Invalid argument[%d] name: %s\n",
 513				i, tp->args[i].name);
 514			ret = -EINVAL;
 515			goto error;
 516		}
 517
 518		if (traceprobe_conflict_field_name(tp->args[i].name,
 519							tp->args, i)) {
 520			pr_info("Argument[%d] name '%s' conflicts with "
 521				"another field.\n", i, argv[i]);
 522			ret = -EINVAL;
 523			goto error;
 524		}
 525
 526		/* Parse fetch argument */
 527		ret = traceprobe_parse_probe_arg(arg, &tp->size, &tp->args[i],
 528						is_return, true);
 529		if (ret) {
 530			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 531			goto error;
 532		}
 
 
 
 533	}
 534
 535	ret = register_trace_probe(tp);
 536	if (ret)
 537		goto error;
 538	return 0;
 539
 
 
 540error:
 541	free_trace_probe(tp);
 542	return ret;
 543}
 544
 545static int release_all_trace_probes(void)
 546{
 547	struct trace_probe *tp;
 548	int ret = 0;
 549
 550	mutex_lock(&probe_lock);
 551	/* Ensure no probe is in use. */
 552	list_for_each_entry(tp, &probe_list, list)
 553		if (trace_probe_is_enabled(tp)) {
 554			ret = -EBUSY;
 555			goto end;
 556		}
 557	/* TODO: Use batch unregistration */
 558	while (!list_empty(&probe_list)) {
 559		tp = list_entry(probe_list.next, struct trace_probe, list);
 560		unregister_trace_probe(tp);
 561		free_trace_probe(tp);
 562	}
 563
 564end:
 565	mutex_unlock(&probe_lock);
 566
 567	return ret;
 
 568}
 569
 570/* Probes listing interfaces */
 571static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 572{
 573	mutex_lock(&probe_lock);
 574	return seq_list_start(&probe_list, *pos);
 575}
 576
 577static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 578{
 579	return seq_list_next(v, &probe_list, pos);
 580}
 581
 582static void probes_seq_stop(struct seq_file *m, void *v)
 583{
 584	mutex_unlock(&probe_lock);
 585}
 586
 587static int probes_seq_show(struct seq_file *m, void *v)
 588{
 589	struct trace_probe *tp = v;
 590	int i;
 591
 592	seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
 593	seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
 594
 595	if (!tp->symbol)
 596		seq_printf(m, " 0x%p", tp->rp.kp.addr);
 597	else if (tp->rp.kp.offset)
 598		seq_printf(m, " %s+%u", trace_probe_symbol(tp),
 599			   tp->rp.kp.offset);
 
 600	else
 601		seq_printf(m, " %s", trace_probe_symbol(tp));
 602
 603	for (i = 0; i < tp->nr_args; i++)
 604		seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
 605	seq_printf(m, "\n");
 606
 607	return 0;
 608}
 609
 
 
 
 
 
 
 
 
 
 
 610static const struct seq_operations probes_seq_op = {
 611	.start  = probes_seq_start,
 612	.next   = probes_seq_next,
 613	.stop   = probes_seq_stop,
 614	.show   = probes_seq_show
 615};
 616
 617static int probes_open(struct inode *inode, struct file *file)
 618{
 619	int ret;
 620
 
 
 
 
 621	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 622		ret = release_all_trace_probes();
 623		if (ret < 0)
 624			return ret;
 625	}
 626
 627	return seq_open(file, &probes_seq_op);
 628}
 629
 630static ssize_t probes_write(struct file *file, const char __user *buffer,
 631			    size_t count, loff_t *ppos)
 632{
 633	return traceprobe_probes_write(file, buffer, count, ppos,
 634			create_trace_probe);
 635}
 636
 637static const struct file_operations kprobe_events_ops = {
 638	.owner          = THIS_MODULE,
 639	.open           = probes_open,
 640	.read           = seq_read,
 641	.llseek         = seq_lseek,
 642	.release        = seq_release,
 643	.write		= probes_write,
 644};
 645
 646/* Probes profiling interfaces */
 647static int probes_profile_seq_show(struct seq_file *m, void *v)
 648{
 649	struct trace_probe *tp = v;
 
 
 
 
 650
 651	seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
 652		   tp->rp.kp.nmissed);
 
 
 
 653
 654	return 0;
 655}
 656
 657static const struct seq_operations profile_seq_op = {
 658	.start  = probes_seq_start,
 659	.next   = probes_seq_next,
 660	.stop   = probes_seq_stop,
 661	.show   = probes_profile_seq_show
 662};
 663
 664static int profile_open(struct inode *inode, struct file *file)
 665{
 
 
 
 
 
 
 666	return seq_open(file, &profile_seq_op);
 667}
 668
 669static const struct file_operations kprobe_profile_ops = {
 670	.owner          = THIS_MODULE,
 671	.open           = profile_open,
 672	.read           = seq_read,
 673	.llseek         = seq_lseek,
 674	.release        = seq_release,
 675};
 676
 677/* Sum up total data length for dynamic arraies (strings) */
 678static __kprobes int __get_data_size(struct trace_probe *tp,
 679				     struct pt_regs *regs)
 680{
 681	int i, ret = 0;
 682	u32 len;
 683
 684	for (i = 0; i < tp->nr_args; i++)
 685		if (unlikely(tp->args[i].fetch_size.fn)) {
 686			call_fetch(&tp->args[i].fetch_size, regs, &len);
 687			ret += len;
 688		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 689
 690	return ret;
 691}
 692
 693/* Store the value of each argument */
 694static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
 695				       struct pt_regs *regs,
 696				       u8 *data, int maxlen)
 697{
 698	int i;
 699	u32 end = tp->size;
 700	u32 *dl;	/* Data (relative) location */
 701
 702	for (i = 0; i < tp->nr_args; i++) {
 703		if (unlikely(tp->args[i].fetch_size.fn)) {
 704			/*
 705			 * First, we set the relative location and
 706			 * maximum data length to *dl
 707			 */
 708			dl = (u32 *)(data + tp->args[i].offset);
 709			*dl = make_data_rloc(maxlen, end - tp->args[i].offset);
 710			/* Then try to fetch string or dynamic array data */
 711			call_fetch(&tp->args[i].fetch, regs, dl);
 712			/* Reduce maximum length */
 713			end += get_rloc_len(*dl);
 714			maxlen -= get_rloc_len(*dl);
 715			/* Trick here, convert data_rloc to data_loc */
 716			*dl = convert_rloc_to_loc(*dl,
 717				 ent_size + tp->args[i].offset);
 718		} else
 719			/* Just fetching data normally */
 720			call_fetch(&tp->args[i].fetch, regs,
 721				   data + tp->args[i].offset);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 722	}
 
 
 
 723}
 
 724
 725/* Kprobe handler */
 726static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 
 
 727{
 728	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 729	struct kprobe_trace_entry_head *entry;
 730	struct ring_buffer_event *event;
 731	struct ring_buffer *buffer;
 732	int size, dsize, pc;
 733	unsigned long irq_flags;
 734	struct ftrace_event_call *call = &tp->call;
 735
 736	tp->nhit++;
 
 
 
 737
 738	local_save_flags(irq_flags);
 739	pc = preempt_count();
 740
 741	dsize = __get_data_size(tp, regs);
 742	size = sizeof(*entry) + tp->size + dsize;
 743
 744	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 745						  size, irq_flags, pc);
 
 746	if (!event)
 747		return;
 748
 749	entry = ring_buffer_event_data(event);
 750	entry->ip = (unsigned long)kp->addr;
 751	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 
 
 
 
 752
 753	if (!filter_current_check_discard(buffer, call, entry, event))
 754		trace_nowake_buffer_unlock_commit_regs(buffer, event,
 755						       irq_flags, pc, regs);
 
 
 
 
 756}
 
 757
 758/* Kretprobe handler */
 759static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
 760					  struct pt_regs *regs)
 
 
 761{
 762	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
 763	struct kretprobe_trace_entry_head *entry;
 764	struct ring_buffer_event *event;
 765	struct ring_buffer *buffer;
 766	int size, pc, dsize;
 767	unsigned long irq_flags;
 768	struct ftrace_event_call *call = &tp->call;
 
 
 
 
 
 769
 770	local_save_flags(irq_flags);
 771	pc = preempt_count();
 772
 773	dsize = __get_data_size(tp, regs);
 774	size = sizeof(*entry) + tp->size + dsize;
 775
 776	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 777						  size, irq_flags, pc);
 
 778	if (!event)
 779		return;
 780
 781	entry = ring_buffer_event_data(event);
 782	entry->func = (unsigned long)tp->rp.kp.addr;
 783	entry->ret_ip = (unsigned long)ri->ret_addr;
 784	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 785
 786	if (!filter_current_check_discard(buffer, call, entry, event))
 787		trace_nowake_buffer_unlock_commit_regs(buffer, event,
 788						       irq_flags, pc, regs);
 789}
 790
 
 
 
 
 
 
 
 
 
 
 
 791/* Event entry printers */
 792enum print_line_t
 793print_kprobe_event(struct trace_iterator *iter, int flags,
 794		   struct trace_event *event)
 795{
 796	struct kprobe_trace_entry_head *field;
 797	struct trace_seq *s = &iter->seq;
 798	struct trace_probe *tp;
 799	u8 *data;
 800	int i;
 801
 802	field = (struct kprobe_trace_entry_head *)iter->ent;
 803	tp = container_of(event, struct trace_probe, call.event);
 
 
 
 804
 805	if (!trace_seq_printf(s, "%s: (", tp->call.name))
 806		goto partial;
 807
 808	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
 809		goto partial;
 810
 811	if (!trace_seq_puts(s, ")"))
 812		goto partial;
 813
 814	data = (u8 *)&field[1];
 815	for (i = 0; i < tp->nr_args; i++)
 816		if (!tp->args[i].type->print(s, tp->args[i].name,
 817					     data + tp->args[i].offset, field))
 818			goto partial;
 819
 820	if (!trace_seq_puts(s, "\n"))
 821		goto partial;
 
 822
 823	return TRACE_TYPE_HANDLED;
 824partial:
 825	return TRACE_TYPE_PARTIAL_LINE;
 826}
 827
 828enum print_line_t
 829print_kretprobe_event(struct trace_iterator *iter, int flags,
 830		      struct trace_event *event)
 831{
 832	struct kretprobe_trace_entry_head *field;
 833	struct trace_seq *s = &iter->seq;
 834	struct trace_probe *tp;
 835	u8 *data;
 836	int i;
 837
 838	field = (struct kretprobe_trace_entry_head *)iter->ent;
 839	tp = container_of(event, struct trace_probe, call.event);
 
 
 
 840
 841	if (!trace_seq_printf(s, "%s: (", tp->call.name))
 842		goto partial;
 843
 844	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
 845		goto partial;
 846
 847	if (!trace_seq_puts(s, " <- "))
 848		goto partial;
 849
 850	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
 851		goto partial;
 852
 853	if (!trace_seq_puts(s, ")"))
 854		goto partial;
 855
 856	data = (u8 *)&field[1];
 857	for (i = 0; i < tp->nr_args; i++)
 858		if (!tp->args[i].type->print(s, tp->args[i].name,
 859					     data + tp->args[i].offset, field))
 860			goto partial;
 861
 862	if (!trace_seq_puts(s, "\n"))
 863		goto partial;
 864
 865	return TRACE_TYPE_HANDLED;
 866partial:
 867	return TRACE_TYPE_PARTIAL_LINE;
 868}
 869
 870
 871static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
 872{
 873	int ret, i;
 874	struct kprobe_trace_entry_head field;
 875	struct trace_probe *tp = (struct trace_probe *)event_call->data;
 
 
 
 
 876
 877	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
 878	/* Set argument names as fields */
 879	for (i = 0; i < tp->nr_args; i++) {
 880		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
 881					 tp->args[i].name,
 882					 sizeof(field) + tp->args[i].offset,
 883					 tp->args[i].type->size,
 884					 tp->args[i].type->is_signed,
 885					 FILTER_OTHER);
 886		if (ret)
 887			return ret;
 888	}
 889	return 0;
 890}
 891
 892static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
 893{
 894	int ret, i;
 895	struct kretprobe_trace_entry_head field;
 896	struct trace_probe *tp = (struct trace_probe *)event_call->data;
 
 
 
 
 897
 898	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
 899	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
 900	/* Set argument names as fields */
 901	for (i = 0; i < tp->nr_args; i++) {
 902		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
 903					 tp->args[i].name,
 904					 sizeof(field) + tp->args[i].offset,
 905					 tp->args[i].type->size,
 906					 tp->args[i].type->is_signed,
 907					 FILTER_OTHER);
 908		if (ret)
 909			return ret;
 910	}
 911	return 0;
 912}
 913
 914static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
 915{
 916	int i;
 917	int pos = 0;
 918
 919	const char *fmt, *arg;
 920
 921	if (!trace_probe_is_return(tp)) {
 922		fmt = "(%lx)";
 923		arg = "REC->" FIELD_STRING_IP;
 924	} else {
 925		fmt = "(%lx <- %lx)";
 926		arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
 927	}
 928
 929	/* When len=0, we just calculate the needed length */
 930#define LEN_OR_ZERO (len ? len - pos : 0)
 931
 932	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 933
 934	for (i = 0; i < tp->nr_args; i++) {
 935		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
 936				tp->args[i].name, tp->args[i].type->fmt);
 937	}
 938
 939	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 940
 941	for (i = 0; i < tp->nr_args; i++) {
 942		if (strcmp(tp->args[i].type->name, "string") == 0)
 943			pos += snprintf(buf + pos, LEN_OR_ZERO,
 944					", __get_str(%s)",
 945					tp->args[i].name);
 946		else
 947			pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
 948					tp->args[i].name);
 949	}
 950
 951#undef LEN_OR_ZERO
 952
 953	/* return the length of print_fmt */
 954	return pos;
 955}
 956
 957static int set_print_fmt(struct trace_probe *tp)
 958{
 959	int len;
 960	char *print_fmt;
 961
 962	/* First: called with 0 length to calculate the needed length */
 963	len = __set_print_fmt(tp, NULL, 0);
 964	print_fmt = kmalloc(len + 1, GFP_KERNEL);
 965	if (!print_fmt)
 966		return -ENOMEM;
 967
 968	/* Second: actually write the @print_fmt */
 969	__set_print_fmt(tp, print_fmt, len + 1);
 970	tp->call.print_fmt = print_fmt;
 971
 972	return 0;
 973}
 974
 975#ifdef CONFIG_PERF_EVENTS
 976
 977/* Kprobe profile handler */
 978static __kprobes void kprobe_perf_func(struct kprobe *kp,
 979					 struct pt_regs *regs)
 980{
 981	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 982	struct ftrace_event_call *call = &tp->call;
 983	struct kprobe_trace_entry_head *entry;
 984	struct hlist_head *head;
 985	int size, __size, dsize;
 986	int rctx;
 987
 988	dsize = __get_data_size(tp, regs);
 989	__size = sizeof(*entry) + tp->size + dsize;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 990	size = ALIGN(__size + sizeof(u32), sizeof(u64));
 991	size -= sizeof(u32);
 992	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 993		     "profile buffer not large enough"))
 994		return;
 995
 996	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 997	if (!entry)
 998		return;
 999
1000	entry->ip = (unsigned long)kp->addr;
1001	memset(&entry[1], 0, dsize);
1002	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1003
1004	head = this_cpu_ptr(call->perf_events);
1005	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1006}
 
1007
1008/* Kretprobe profile handler */
1009static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1010					    struct pt_regs *regs)
 
1011{
1012	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1013	struct ftrace_event_call *call = &tp->call;
1014	struct kretprobe_trace_entry_head *entry;
1015	struct hlist_head *head;
1016	int size, __size, dsize;
1017	int rctx;
1018
1019	dsize = __get_data_size(tp, regs);
1020	__size = sizeof(*entry) + tp->size + dsize;
 
 
 
 
 
 
 
1021	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1022	size -= sizeof(u32);
1023	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1024		     "profile buffer not large enough"))
1025		return;
1026
1027	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1028	if (!entry)
1029		return;
1030
1031	entry->func = (unsigned long)tp->rp.kp.addr;
1032	entry->ret_ip = (unsigned long)ri->ret_addr;
1033	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 
 
 
 
 
 
 
 
 
 
 
 
1034
1035	head = this_cpu_ptr(call->perf_events);
1036	perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1037}
1038#endif	/* CONFIG_PERF_EVENTS */
1039
1040static __kprobes
1041int kprobe_register(struct ftrace_event_call *event,
1042		    enum trace_reg type, void *data)
 
 
 
 
 
1043{
1044	struct trace_probe *tp = (struct trace_probe *)event->data;
1045
1046	switch (type) {
1047	case TRACE_REG_REGISTER:
1048		return enable_trace_probe(tp, TP_FLAG_TRACE);
1049	case TRACE_REG_UNREGISTER:
1050		disable_trace_probe(tp, TP_FLAG_TRACE);
1051		return 0;
1052
1053#ifdef CONFIG_PERF_EVENTS
1054	case TRACE_REG_PERF_REGISTER:
1055		return enable_trace_probe(tp, TP_FLAG_PROFILE);
1056	case TRACE_REG_PERF_UNREGISTER:
1057		disable_trace_probe(tp, TP_FLAG_PROFILE);
1058		return 0;
1059	case TRACE_REG_PERF_OPEN:
1060	case TRACE_REG_PERF_CLOSE:
1061	case TRACE_REG_PERF_ADD:
1062	case TRACE_REG_PERF_DEL:
1063		return 0;
1064#endif
1065	}
1066	return 0;
1067}
1068
1069static __kprobes
1070int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1071{
1072	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 
 
 
1073
1074	if (tp->flags & TP_FLAG_TRACE)
1075		kprobe_trace_func(kp, regs);
1076#ifdef CONFIG_PERF_EVENTS
1077	if (tp->flags & TP_FLAG_PROFILE)
1078		kprobe_perf_func(kp, regs);
1079#endif
1080	return 0;	/* We don't tweek kernel, so just return 0 */
1081}
 
1082
1083static __kprobes
1084int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1085{
1086	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1087
1088	if (tp->flags & TP_FLAG_TRACE)
1089		kretprobe_trace_func(ri, regs);
 
 
1090#ifdef CONFIG_PERF_EVENTS
1091	if (tp->flags & TP_FLAG_PROFILE)
1092		kretprobe_perf_func(ri, regs);
1093#endif
1094	return 0;	/* We don't tweek kernel, so just return 0 */
1095}
 
1096
1097static struct trace_event_functions kretprobe_funcs = {
1098	.trace		= print_kretprobe_event
1099};
1100
1101static struct trace_event_functions kprobe_funcs = {
1102	.trace		= print_kprobe_event
1103};
1104
1105static int register_probe_event(struct trace_probe *tp)
1106{
1107	struct ftrace_event_call *call = &tp->call;
1108	int ret;
1109
1110	/* Initialize ftrace_event_call */
1111	INIT_LIST_HEAD(&call->class->fields);
1112	if (trace_probe_is_return(tp)) {
1113		call->event.funcs = &kretprobe_funcs;
1114		call->class->define_fields = kretprobe_event_define_fields;
1115	} else {
1116		call->event.funcs = &kprobe_funcs;
1117		call->class->define_fields = kprobe_event_define_fields;
1118	}
1119	if (set_print_fmt(tp) < 0)
1120		return -ENOMEM;
1121	ret = register_ftrace_event(&call->event);
1122	if (!ret) {
1123		kfree(call->print_fmt);
1124		return -ENODEV;
1125	}
1126	call->flags = 0;
1127	call->class->reg = kprobe_register;
1128	call->data = tp;
1129	ret = trace_add_event_call(call);
1130	if (ret) {
1131		pr_info("Failed to register kprobe event: %s\n", call->name);
1132		kfree(call->print_fmt);
1133		unregister_ftrace_event(&call->event);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1134	}
1135	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1136}
1137
1138static void unregister_probe_event(struct trace_probe *tp)
1139{
1140	/* tp->event is unregistered in trace_remove_event_call() */
1141	trace_remove_event_call(&tp->call);
1142	kfree(tp->call.print_fmt);
 
 
 
 
 
 
 
 
 
 
 
1143}
 
1144
1145/* Make a debugfs interface for controlling probe points */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1146static __init int init_kprobe_trace(void)
1147{
1148	struct dentry *d_tracer;
1149	struct dentry *entry;
 
 
 
 
 
1150
1151	if (register_module_notifier(&trace_probe_module_nb))
1152		return -EINVAL;
1153
1154	d_tracer = tracing_init_dentry();
1155	if (!d_tracer)
1156		return 0;
1157
1158	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1159				    NULL, &kprobe_events_ops);
1160
1161	/* Event list interface */
1162	if (!entry)
1163		pr_warning("Could not create debugfs "
1164			   "'kprobe_events' entry\n");
1165
1166	/* Profile interface */
1167	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1168				    NULL, &kprobe_profile_ops);
1169
1170	if (!entry)
1171		pr_warning("Could not create debugfs "
1172			   "'kprobe_profile' entry\n");
 
 
1173	return 0;
1174}
1175fs_initcall(init_kprobe_trace);
1176
1177
1178#ifdef CONFIG_FTRACE_STARTUP_TEST
1179
1180/*
1181 * The "__used" keeps gcc from removing the function symbol
1182 * from the kallsyms table.
1183 */
1184static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
1185					       int a4, int a5, int a6)
1186{
1187	return a1 + a2 + a3 + a4 + a5 + a6;
 
 
 
 
 
 
1188}
1189
 
 
 
 
1190static __init int kprobe_trace_self_tests_init(void)
1191{
1192	int ret, warn = 0;
1193	int (*target)(int, int, int, int, int, int);
1194	struct trace_probe *tp;
 
 
 
 
 
 
 
 
 
1195
1196	target = kprobe_trace_selftest_target;
1197
1198	pr_info("Testing kprobe tracing: ");
1199
1200	ret = traceprobe_command("p:testprobe kprobe_trace_selftest_target "
1201				  "$stack $stack0 +0($stack)",
1202				  create_trace_probe);
1203	if (WARN_ON_ONCE(ret)) {
1204		pr_warning("error on probing function entry.\n");
1205		warn++;
1206	} else {
1207		/* Enable trace point */
1208		tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1209		if (WARN_ON_ONCE(tp == NULL)) {
1210			pr_warning("error on getting new probe.\n");
1211			warn++;
1212		} else
1213			enable_trace_probe(tp, TP_FLAG_TRACE);
 
 
 
 
 
 
 
1214	}
1215
1216	ret = traceprobe_command("r:testprobe2 kprobe_trace_selftest_target "
1217				  "$retval", create_trace_probe);
1218	if (WARN_ON_ONCE(ret)) {
1219		pr_warning("error on probing function return.\n");
1220		warn++;
1221	} else {
1222		/* Enable trace point */
1223		tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1224		if (WARN_ON_ONCE(tp == NULL)) {
1225			pr_warning("error on getting new probe.\n");
1226			warn++;
1227		} else
1228			enable_trace_probe(tp, TP_FLAG_TRACE);
 
 
 
 
 
 
 
1229	}
1230
1231	if (warn)
1232		goto end;
1233
1234	ret = target(1, 2, 3, 4, 5, 6);
1235
 
 
 
 
 
 
 
 
1236	/* Disable trace points before removing it */
1237	tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
1238	if (WARN_ON_ONCE(tp == NULL)) {
1239		pr_warning("error on getting test probe.\n");
1240		warn++;
1241	} else
1242		disable_trace_probe(tp, TP_FLAG_TRACE);
 
 
 
 
 
 
 
 
 
 
 
 
1243
1244	tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
1245	if (WARN_ON_ONCE(tp == NULL)) {
1246		pr_warning("error on getting 2nd test probe.\n");
1247		warn++;
1248	} else
1249		disable_trace_probe(tp, TP_FLAG_TRACE);
 
 
 
1250
1251	ret = traceprobe_command("-:testprobe", create_trace_probe);
 
 
 
 
 
 
 
 
 
1252	if (WARN_ON_ONCE(ret)) {
1253		pr_warning("error on deleting a probe.\n");
1254		warn++;
1255	}
1256
1257	ret = traceprobe_command("-:testprobe2", create_trace_probe);
1258	if (WARN_ON_ONCE(ret)) {
1259		pr_warning("error on deleting a probe.\n");
1260		warn++;
1261	}
1262
1263end:
1264	release_all_trace_probes();
 
 
 
 
 
 
 
 
 
1265	if (warn)
1266		pr_cont("NG: Some tests are failed. Please check them.\n");
1267	else
1268		pr_cont("OK\n");
1269	return 0;
1270}
1271
1272late_initcall(kprobe_trace_self_tests_init);
1273
1274#endif
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