Linux Audio

Check our new training course

Loading...
v4.17
   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#define pr_fmt(fmt)	"trace_kprobe: " fmt
  20
  21#include <linux/module.h>
  22#include <linux/uaccess.h>
  23#include <linux/rculist.h>
  24#include <linux/error-injection.h>
  25
  26#include "trace_probe.h"
  27
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  28#define KPROBE_EVENT_SYSTEM "kprobes"
  29#define KRETPROBE_MAXACTIVE_MAX 4096
  30
  31/**
  32 * Kprobe event core functions
  33 */
  34struct trace_kprobe {
  35	struct list_head	list;
  36	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
  37	unsigned long __percpu *nhit;
  38	const char		*symbol;	/* symbol name */
  39	struct trace_probe	tp;
 
 
 
 
 
  40};
  41
  42#define SIZEOF_TRACE_KPROBE(n)				\
  43	(offsetof(struct trace_kprobe, tp.args) +	\
  44	(sizeof(struct probe_arg) * (n)))
  45
  46static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  47{
  48	return tk->rp.handler != NULL;
  49}
  50
  51static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  52{
  53	return tk->symbol ? tk->symbol : "unknown";
  54}
  55
  56static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  57{
  58	return tk->rp.kp.offset;
  59}
  60
  61static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  62{
  63	return !!(kprobe_gone(&tk->rp.kp));
  64}
 
 
 
 
  65
  66static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  67						 struct module *mod)
  68{
  69	int len = strlen(mod->name);
  70	const char *name = trace_kprobe_symbol(tk);
  71	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  72}
  73
  74static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  75{
  76	return !!strchr(trace_kprobe_symbol(tk), ':');
  77}
  78
  79static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
 
  80{
  81	unsigned long nhit = 0;
  82	int cpu;
  83
  84	for_each_possible_cpu(cpu)
  85		nhit += *per_cpu_ptr(tk->nhit, cpu);
  86
  87	return nhit;
  88}
  89
  90bool trace_kprobe_on_func_entry(struct trace_event_call *call)
  91{
  92	struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 
 
 
  93
  94	return kprobe_on_func_entry(tk->rp.kp.addr,
  95			tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
  96			tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
  97}
  98
  99bool trace_kprobe_error_injectable(struct trace_event_call *call)
 
 
 
 100{
 101	struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 102	unsigned long addr;
 103
 104	if (tk->symbol) {
 105		addr = (unsigned long)
 106			kallsyms_lookup_name(trace_kprobe_symbol(tk));
 107		addr += tk->rp.kp.offset;
 108	} else {
 109		addr = (unsigned long)tk->rp.kp.addr;
 110	}
 111	return within_error_injection_list(addr);
 112}
 
 113
 114static int register_kprobe_event(struct trace_kprobe *tk);
 115static int unregister_kprobe_event(struct trace_kprobe *tk);
 116
 117static DEFINE_MUTEX(probe_lock);
 118static LIST_HEAD(probe_list);
 119
 120static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 121static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 122				struct pt_regs *regs);
 123
 124/* Memory fetching by symbol */
 125struct symbol_cache {
 126	char		*symbol;
 127	long		offset;
 128	unsigned long	addr;
 129};
 130
 131unsigned long update_symbol_cache(struct symbol_cache *sc)
 
 132{
 133	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 134
 135	if (sc->addr)
 136		sc->addr += sc->offset;
 137
 138	return sc->addr;
 139}
 140
 141void free_symbol_cache(struct symbol_cache *sc)
 142{
 143	kfree(sc->symbol);
 144	kfree(sc);
 145}
 146
 147struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 148{
 149	struct symbol_cache *sc;
 150
 151	if (!sym || strlen(sym) == 0)
 152		return NULL;
 153
 154	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 155	if (!sc)
 156		return NULL;
 157
 158	sc->symbol = kstrdup(sym, GFP_KERNEL);
 159	if (!sc->symbol) {
 160		kfree(sc);
 161		return NULL;
 162	}
 163	sc->offset = offset;
 164	update_symbol_cache(sc);
 165
 166	return sc;
 167}
 168
 
 169/*
 170 * Kprobes-specific fetch functions
 
 171 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 172#define DEFINE_FETCH_stack(type)					\
 173static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,		\
 174					  void *offset, void *dest)	\
 175{									\
 176	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
 177				(unsigned int)((unsigned long)offset));	\
 178}									\
 179NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 180
 181DEFINE_BASIC_FETCH_FUNCS(stack)
 182/* No string on the stack entry */
 183#define fetch_stack_string	NULL
 184#define fetch_stack_string_size	NULL
 
 
 
 
 
 
 
 
 
 
 
 185
 186#define DEFINE_FETCH_memory(type)					\
 187static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,		\
 188					  void *addr, void *dest)	\
 189{									\
 190	type retval;							\
 191	if (probe_kernel_address(addr, retval))				\
 192		*(type *)dest = 0;					\
 193	else								\
 194		*(type *)dest = retval;					\
 195}									\
 196NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 197
 198DEFINE_BASIC_FETCH_FUNCS(memory)
 199/*
 200 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 201 * length and relative data location.
 202 */
 203static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 204					    void *addr, void *dest)
 205{
 
 206	int maxlen = get_rloc_len(*(u32 *)dest);
 207	u8 *dst = get_rloc_data(dest);
 208	long ret;
 209
 210	if (!maxlen)
 211		return;
 212
 213	/*
 214	 * Try to get string again, since the string can be changed while
 215	 * probing.
 216	 */
 217	ret = strncpy_from_unsafe(dst, addr, maxlen);
 
 
 
 
 
 
 
 218
 219	if (ret < 0) {	/* Failed to fetch string */
 220		dst[0] = '\0';
 221		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 222	} else {
 223		*(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 224	}
 225}
 226NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 227
 228/* Return the length of string -- including null terminal byte */
 229static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 230						 void *addr, void *dest)
 231{
 232	mm_segment_t old_fs;
 233	int ret, len = 0;
 234	u8 c;
 
 235
 236	old_fs = get_fs();
 237	set_fs(KERNEL_DS);
 238	pagefault_disable();
 239
 240	do {
 241		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 242		len++;
 243	} while (c && ret == 0 && len < MAX_STRING_SIZE);
 244
 245	pagefault_enable();
 246	set_fs(old_fs);
 247
 248	if (ret < 0)	/* Failed to check the length */
 249		*(u32 *)dest = 0;
 250	else
 251		*(u32 *)dest = len;
 252}
 253NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 254
 255#define DEFINE_FETCH_symbol(type)					\
 256void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 
 257{									\
 258	struct symbol_cache *sc = data;					\
 259	if (sc->addr)							\
 260		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
 261	else								\
 262		*(type *)dest = 0;					\
 263}									\
 264NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 265
 266DEFINE_BASIC_FETCH_FUNCS(symbol)
 267DEFINE_FETCH_symbol(string)
 268DEFINE_FETCH_symbol(string_size)
 269
 270/* kprobes don't support file_offset fetch methods */
 271#define fetch_file_offset_u8		NULL
 272#define fetch_file_offset_u16		NULL
 273#define fetch_file_offset_u32		NULL
 274#define fetch_file_offset_u64		NULL
 275#define fetch_file_offset_string	NULL
 276#define fetch_file_offset_string_size	NULL
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 277
 278/* Fetch type information table */
 279static const struct fetch_type kprobes_fetch_type_table[] = {
 
 
 
 
 
 
 
 
 
 280	/* Special types */
 281	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 282					sizeof(u32), 1, "__data_loc char[]"),
 283	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 284					string_size, sizeof(u32), 0, "u32"),
 285	/* Basic types */
 286	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 287	ASSIGN_FETCH_TYPE(u16, u16, 0),
 288	ASSIGN_FETCH_TYPE(u32, u32, 0),
 289	ASSIGN_FETCH_TYPE(u64, u64, 0),
 290	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 291	ASSIGN_FETCH_TYPE(s16, u16, 1),
 292	ASSIGN_FETCH_TYPE(s32, u32, 1),
 293	ASSIGN_FETCH_TYPE(s64, u64, 1),
 294	ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 295	ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 296	ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 297	ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 
 298
 299	ASSIGN_FETCH_TYPE_END
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 300};
 301
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 302/*
 303 * Allocate new trace_probe and initialize it (including kprobes).
 304 */
 305static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 306					     const char *event,
 307					     void *addr,
 308					     const char *symbol,
 309					     unsigned long offs,
 310					     int maxactive,
 311					     int nargs, bool is_return)
 312{
 313	struct trace_kprobe *tk;
 314	int ret = -ENOMEM;
 315
 316	tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 317	if (!tk)
 318		return ERR_PTR(ret);
 319
 320	tk->nhit = alloc_percpu(unsigned long);
 321	if (!tk->nhit)
 322		goto error;
 323
 324	if (symbol) {
 325		tk->symbol = kstrdup(symbol, GFP_KERNEL);
 326		if (!tk->symbol)
 327			goto error;
 328		tk->rp.kp.symbol_name = tk->symbol;
 329		tk->rp.kp.offset = offs;
 330	} else
 331		tk->rp.kp.addr = addr;
 332
 333	if (is_return)
 334		tk->rp.handler = kretprobe_dispatcher;
 335	else
 336		tk->rp.kp.pre_handler = kprobe_dispatcher;
 337
 338	tk->rp.maxactive = maxactive;
 339
 340	if (!event || !is_good_name(event)) {
 341		ret = -EINVAL;
 342		goto error;
 343	}
 344
 345	tk->tp.call.class = &tk->tp.class;
 346	tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 347	if (!tk->tp.call.name)
 348		goto error;
 349
 350	if (!group || !is_good_name(group)) {
 351		ret = -EINVAL;
 352		goto error;
 353	}
 354
 355	tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 356	if (!tk->tp.class.system)
 357		goto error;
 358
 359	INIT_LIST_HEAD(&tk->list);
 360	INIT_LIST_HEAD(&tk->tp.files);
 361	return tk;
 362error:
 363	kfree(tk->tp.call.name);
 364	kfree(tk->symbol);
 365	free_percpu(tk->nhit);
 366	kfree(tk);
 367	return ERR_PTR(ret);
 368}
 369
 370static void free_trace_kprobe(struct trace_kprobe *tk)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 371{
 372	int i;
 373
 374	for (i = 0; i < tk->tp.nr_args; i++)
 375		traceprobe_free_probe_arg(&tk->tp.args[i]);
 376
 377	kfree(tk->tp.call.class->system);
 378	kfree(tk->tp.call.name);
 379	kfree(tk->symbol);
 380	free_percpu(tk->nhit);
 381	kfree(tk);
 382}
 383
 384static struct trace_kprobe *find_trace_kprobe(const char *event,
 385					      const char *group)
 386{
 387	struct trace_kprobe *tk;
 388
 389	list_for_each_entry(tk, &probe_list, list)
 390		if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 391		    strcmp(tk->tp.call.class->system, group) == 0)
 392			return tk;
 393	return NULL;
 394}
 395
 396/*
 397 * Enable trace_probe
 398 * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 399 */
 400static int
 401enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 402{
 403	int ret = 0;
 404
 405	if (file) {
 406		struct event_file_link *link;
 407
 408		link = kmalloc(sizeof(*link), GFP_KERNEL);
 409		if (!link) {
 410			ret = -ENOMEM;
 411			goto out;
 412		}
 413
 414		link->file = file;
 415		list_add_tail_rcu(&link->list, &tk->tp.files);
 416
 417		tk->tp.flags |= TP_FLAG_TRACE;
 418	} else
 419		tk->tp.flags |= TP_FLAG_PROFILE;
 420
 421	if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 422		if (trace_kprobe_is_return(tk))
 423			ret = enable_kretprobe(&tk->rp);
 424		else
 425			ret = enable_kprobe(&tk->rp.kp);
 426	}
 427 out:
 428	return ret;
 429}
 430
 431/*
 432 * Disable trace_probe
 433 * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 434 */
 435static int
 436disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 437{
 438	struct event_file_link *link = NULL;
 439	int wait = 0;
 440	int ret = 0;
 441
 442	if (file) {
 443		link = find_event_file_link(&tk->tp, file);
 444		if (!link) {
 445			ret = -EINVAL;
 446			goto out;
 447		}
 448
 449		list_del_rcu(&link->list);
 450		wait = 1;
 451		if (!list_empty(&tk->tp.files))
 452			goto out;
 453
 454		tk->tp.flags &= ~TP_FLAG_TRACE;
 455	} else
 456		tk->tp.flags &= ~TP_FLAG_PROFILE;
 457
 458	if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 459		if (trace_kprobe_is_return(tk))
 460			disable_kretprobe(&tk->rp);
 461		else
 462			disable_kprobe(&tk->rp.kp);
 463		wait = 1;
 464	}
 465
 466	/*
 467	 * if tk is not added to any list, it must be a local trace_kprobe
 468	 * created with perf_event_open. We don't need to wait for these
 469	 * trace_kprobes
 470	 */
 471	if (list_empty(&tk->list))
 472		wait = 0;
 473 out:
 474	if (wait) {
 475		/*
 476		 * Synchronize with kprobe_trace_func/kretprobe_trace_func
 477		 * to ensure disabled (all running handlers are finished).
 478		 * This is not only for kfree(), but also the caller,
 479		 * trace_remove_event_call() supposes it for releasing
 480		 * event_call related objects, which will be accessed in
 481		 * the kprobe_trace_func/kretprobe_trace_func.
 482		 */
 483		synchronize_sched();
 484		kfree(link);	/* Ignored if link == NULL */
 485	}
 486
 487	return ret;
 488}
 489
 490/* Internal register function - just handle k*probes and flags */
 491static int __register_trace_kprobe(struct trace_kprobe *tk)
 492{
 493	int i, ret;
 494
 495	if (trace_probe_is_registered(&tk->tp))
 496		return -EINVAL;
 497
 498	for (i = 0; i < tk->tp.nr_args; i++)
 499		traceprobe_update_arg(&tk->tp.args[i]);
 500
 501	/* Set/clear disabled flag according to tp->flag */
 502	if (trace_probe_is_enabled(&tk->tp))
 503		tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 504	else
 505		tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 506
 507	if (trace_kprobe_is_return(tk))
 508		ret = register_kretprobe(&tk->rp);
 509	else
 510		ret = register_kprobe(&tk->rp.kp);
 511
 512	if (ret == 0)
 513		tk->tp.flags |= TP_FLAG_REGISTERED;
 514	else {
 515		if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 516			pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 
 
 
 517			ret = 0;
 518		} else if (ret == -EILSEQ) {
 519			pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 520				tk->rp.kp.addr);
 
 521			ret = -EINVAL;
 522		}
 523	}
 524
 525	return ret;
 526}
 527
 528/* Internal unregister function - just handle k*probes and flags */
 529static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 530{
 531	if (trace_probe_is_registered(&tk->tp)) {
 532		if (trace_kprobe_is_return(tk))
 533			unregister_kretprobe(&tk->rp);
 534		else
 535			unregister_kprobe(&tk->rp.kp);
 536		tk->tp.flags &= ~TP_FLAG_REGISTERED;
 537		/* Cleanup kprobe for reuse */
 538		if (tk->rp.kp.symbol_name)
 539			tk->rp.kp.addr = NULL;
 540	}
 541}
 542
 543/* Unregister a trace_probe and probe_event: call with locking probe_lock */
 544static int unregister_trace_kprobe(struct trace_kprobe *tk)
 545{
 546	/* Enabled event can not be unregistered */
 547	if (trace_probe_is_enabled(&tk->tp))
 548		return -EBUSY;
 549
 550	/* Will fail if probe is being used by ftrace or perf */
 551	if (unregister_kprobe_event(tk))
 552		return -EBUSY;
 553
 554	__unregister_trace_kprobe(tk);
 555	list_del(&tk->list);
 556
 557	return 0;
 558}
 559
 560/* Register a trace_probe and probe_event */
 561static int register_trace_kprobe(struct trace_kprobe *tk)
 562{
 563	struct trace_kprobe *old_tk;
 564	int ret;
 565
 566	mutex_lock(&probe_lock);
 567
 568	/* Delete old (same name) event if exist */
 569	old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 570			tk->tp.call.class->system);
 571	if (old_tk) {
 572		ret = unregister_trace_kprobe(old_tk);
 573		if (ret < 0)
 574			goto end;
 575		free_trace_kprobe(old_tk);
 576	}
 577
 578	/* Register new event */
 579	ret = register_kprobe_event(tk);
 580	if (ret) {
 581		pr_warn("Failed to register probe event(%d)\n", ret);
 582		goto end;
 583	}
 584
 585	/* Register k*probe */
 586	ret = __register_trace_kprobe(tk);
 587	if (ret < 0)
 588		unregister_kprobe_event(tk);
 589	else
 590		list_add_tail(&tk->list, &probe_list);
 591
 592end:
 593	mutex_unlock(&probe_lock);
 594	return ret;
 595}
 596
 597/* Module notifier call back, checking event on the module */
 598static int trace_kprobe_module_callback(struct notifier_block *nb,
 599				       unsigned long val, void *data)
 600{
 601	struct module *mod = data;
 602	struct trace_kprobe *tk;
 603	int ret;
 604
 605	if (val != MODULE_STATE_COMING)
 606		return NOTIFY_DONE;
 607
 608	/* Update probes on coming module */
 609	mutex_lock(&probe_lock);
 610	list_for_each_entry(tk, &probe_list, list) {
 611		if (trace_kprobe_within_module(tk, mod)) {
 612			/* Don't need to check busy - this should have gone. */
 613			__unregister_trace_kprobe(tk);
 614			ret = __register_trace_kprobe(tk);
 615			if (ret)
 616				pr_warn("Failed to re-register probe %s on %s: %d\n",
 617					trace_event_name(&tk->tp.call),
 618					mod->name, ret);
 619		}
 620	}
 621	mutex_unlock(&probe_lock);
 622
 623	return NOTIFY_DONE;
 624}
 625
 626static struct notifier_block trace_kprobe_module_nb = {
 627	.notifier_call = trace_kprobe_module_callback,
 628	.priority = 1	/* Invoked after kprobe module callback */
 629};
 630
 631/* Convert certain expected symbols into '_' when generating event names */
 632static inline void sanitize_event_name(char *name)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 633{
 634	while (*name++ != '\0')
 635		if (*name == ':' || *name == '.')
 636			*name = '_';
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 637}
 638
 639static int create_trace_kprobe(int argc, char **argv)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 640{
 641	/*
 642	 * Argument syntax:
 643	 *  - Add kprobe:
 644	 *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 645	 *  - Add kretprobe:
 646	 *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 647	 * Fetch args:
 648	 *  $retval	: fetch return value
 649	 *  $stack	: fetch stack address
 650	 *  $stackN	: fetch Nth of stack (N:0-)
 651	 *  $comm       : fetch current task comm
 652	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
 653	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 654	 *  %REG	: fetch register REG
 655	 * Dereferencing memory fetch:
 656	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 657	 * Alias name of args:
 658	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 659	 * Type of args:
 660	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 661	 */
 662	struct trace_kprobe *tk;
 663	int i, ret = 0;
 664	bool is_return = false, is_delete = false;
 665	char *symbol = NULL, *event = NULL, *group = NULL;
 666	int maxactive = 0;
 667	char *arg;
 668	long offset = 0;
 669	void *addr = NULL;
 670	char buf[MAX_EVENT_NAME_LEN];
 671
 672	/* argc must be >= 1 */
 673	if (argv[0][0] == 'p')
 674		is_return = false;
 675	else if (argv[0][0] == 'r')
 676		is_return = true;
 677	else if (argv[0][0] == '-')
 678		is_delete = true;
 679	else {
 680		pr_info("Probe definition must be started with 'p', 'r' or"
 681			" '-'.\n");
 682		return -EINVAL;
 683	}
 684
 685	event = strchr(&argv[0][1], ':');
 686	if (event) {
 687		event[0] = '\0';
 688		event++;
 689	}
 690	if (is_return && isdigit(argv[0][1])) {
 691		ret = kstrtouint(&argv[0][1], 0, &maxactive);
 692		if (ret) {
 693			pr_info("Failed to parse maxactive.\n");
 694			return ret;
 695		}
 696		/* kretprobes instances are iterated over via a list. The
 697		 * maximum should stay reasonable.
 698		 */
 699		if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 700			pr_info("Maxactive is too big (%d > %d).\n",
 701				maxactive, KRETPROBE_MAXACTIVE_MAX);
 702			return -E2BIG;
 703		}
 704	}
 705
 706	if (event) {
 707		if (strchr(event, '/')) {
 708			group = event;
 709			event = strchr(group, '/') + 1;
 710			event[-1] = '\0';
 711			if (strlen(group) == 0) {
 712				pr_info("Group name is not specified\n");
 713				return -EINVAL;
 714			}
 715		}
 716		if (strlen(event) == 0) {
 717			pr_info("Event name is not specified\n");
 718			return -EINVAL;
 719		}
 720	}
 721	if (!group)
 722		group = KPROBE_EVENT_SYSTEM;
 723
 724	if (is_delete) {
 725		if (!event) {
 726			pr_info("Delete command needs an event name.\n");
 727			return -EINVAL;
 728		}
 729		mutex_lock(&probe_lock);
 730		tk = find_trace_kprobe(event, group);
 731		if (!tk) {
 732			mutex_unlock(&probe_lock);
 733			pr_info("Event %s/%s doesn't exist.\n", group, event);
 734			return -ENOENT;
 735		}
 736		/* delete an event */
 737		ret = unregister_trace_kprobe(tk);
 738		if (ret == 0)
 739			free_trace_kprobe(tk);
 740		mutex_unlock(&probe_lock);
 741		return ret;
 742	}
 743
 744	if (argc < 2) {
 745		pr_info("Probe point is not specified.\n");
 746		return -EINVAL;
 747	}
 748
 749	/* try to parse an address. if that fails, try to read the
 750	 * input as a symbol. */
 751	if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 
 
 
 
 
 
 
 
 752		/* a symbol specified */
 753		symbol = argv[1];
 754		/* TODO: support .init module functions */
 755		ret = traceprobe_split_symbol_offset(symbol, &offset);
 756		if (ret || offset < 0 || offset > UINT_MAX) {
 757			pr_info("Failed to parse either an address or a symbol.\n");
 758			return ret;
 759		}
 760		if (offset && is_return &&
 761		    !kprobe_on_func_entry(NULL, symbol, offset)) {
 762			pr_info("Given offset is not valid for return probe.\n");
 763			return -EINVAL;
 764		}
 765	}
 766	argc -= 2; argv += 2;
 767
 768	/* setup a probe */
 769	if (!event) {
 770		/* Make a new event name */
 771		if (symbol)
 772			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 773				 is_return ? 'r' : 'p', symbol, offset);
 774		else
 775			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 776				 is_return ? 'r' : 'p', addr);
 777		sanitize_event_name(buf);
 778		event = buf;
 779	}
 780	tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 781			       argc, is_return);
 782	if (IS_ERR(tk)) {
 783		pr_info("Failed to allocate trace_probe.(%d)\n",
 784			(int)PTR_ERR(tk));
 785		return PTR_ERR(tk);
 786	}
 787
 788	/* parse arguments */
 789	ret = 0;
 790	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 791		struct probe_arg *parg = &tk->tp.args[i];
 792
 793		/* Increment count for freeing args in error case */
 794		tk->tp.nr_args++;
 795
 796		/* Parse argument name */
 797		arg = strchr(argv[i], '=');
 798		if (arg) {
 799			*arg++ = '\0';
 800			parg->name = kstrdup(argv[i], GFP_KERNEL);
 801		} else {
 802			arg = argv[i];
 803			/* If argument name is omitted, set "argN" */
 804			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 805			parg->name = kstrdup(buf, GFP_KERNEL);
 806		}
 807
 808		if (!parg->name) {
 809			pr_info("Failed to allocate argument[%d] name.\n", i);
 810			ret = -ENOMEM;
 811			goto error;
 812		}
 813
 814		if (!is_good_name(parg->name)) {
 815			pr_info("Invalid argument[%d] name: %s\n",
 816				i, parg->name);
 817			ret = -EINVAL;
 818			goto error;
 819		}
 820
 821		if (traceprobe_conflict_field_name(parg->name,
 822							tk->tp.args, i)) {
 823			pr_info("Argument[%d] name '%s' conflicts with "
 824				"another field.\n", i, argv[i]);
 825			ret = -EINVAL;
 826			goto error;
 827		}
 828
 829		/* Parse fetch argument */
 830		ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 831						is_return, true,
 832						kprobes_fetch_type_table);
 833		if (ret) {
 834			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 835			goto error;
 836		}
 837	}
 838
 839	ret = register_trace_kprobe(tk);
 840	if (ret)
 841		goto error;
 842	return 0;
 843
 844error:
 845	free_trace_kprobe(tk);
 846	return ret;
 847}
 848
 849static int release_all_trace_kprobes(void)
 850{
 851	struct trace_kprobe *tk;
 852	int ret = 0;
 853
 854	mutex_lock(&probe_lock);
 855	/* Ensure no probe is in use. */
 856	list_for_each_entry(tk, &probe_list, list)
 857		if (trace_probe_is_enabled(&tk->tp)) {
 858			ret = -EBUSY;
 859			goto end;
 860		}
 861	/* TODO: Use batch unregistration */
 862	while (!list_empty(&probe_list)) {
 863		tk = list_entry(probe_list.next, struct trace_kprobe, list);
 864		ret = unregister_trace_kprobe(tk);
 865		if (ret)
 866			goto end;
 867		free_trace_kprobe(tk);
 868	}
 869
 870end:
 871	mutex_unlock(&probe_lock);
 872
 873	return ret;
 874}
 875
 876/* Probes listing interfaces */
 877static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 878{
 879	mutex_lock(&probe_lock);
 880	return seq_list_start(&probe_list, *pos);
 881}
 882
 883static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 884{
 885	return seq_list_next(v, &probe_list, pos);
 886}
 887
 888static void probes_seq_stop(struct seq_file *m, void *v)
 889{
 890	mutex_unlock(&probe_lock);
 891}
 892
 893static int probes_seq_show(struct seq_file *m, void *v)
 894{
 895	struct trace_kprobe *tk = v;
 896	int i;
 897
 898	seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 899	seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 900			trace_event_name(&tk->tp.call));
 901
 902	if (!tk->symbol)
 903		seq_printf(m, " 0x%p", tk->rp.kp.addr);
 904	else if (tk->rp.kp.offset)
 905		seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 906			   tk->rp.kp.offset);
 907	else
 908		seq_printf(m, " %s", trace_kprobe_symbol(tk));
 909
 910	for (i = 0; i < tk->tp.nr_args; i++)
 911		seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 912	seq_putc(m, '\n');
 913
 914	return 0;
 915}
 916
 917static const struct seq_operations probes_seq_op = {
 918	.start  = probes_seq_start,
 919	.next   = probes_seq_next,
 920	.stop   = probes_seq_stop,
 921	.show   = probes_seq_show
 922};
 923
 924static int probes_open(struct inode *inode, struct file *file)
 925{
 926	int ret;
 927
 928	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 929		ret = release_all_trace_kprobes();
 930		if (ret < 0)
 931			return ret;
 932	}
 933
 934	return seq_open(file, &probes_seq_op);
 935}
 936
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 937static ssize_t probes_write(struct file *file, const char __user *buffer,
 938			    size_t count, loff_t *ppos)
 939{
 940	return trace_parse_run_command(file, buffer, count, ppos,
 941				       create_trace_kprobe);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 942}
 943
 944static const struct file_operations kprobe_events_ops = {
 945	.owner          = THIS_MODULE,
 946	.open           = probes_open,
 947	.read           = seq_read,
 948	.llseek         = seq_lseek,
 949	.release        = seq_release,
 950	.write		= probes_write,
 951};
 952
 953/* Probes profiling interfaces */
 954static int probes_profile_seq_show(struct seq_file *m, void *v)
 955{
 956	struct trace_kprobe *tk = v;
 957
 958	seq_printf(m, "  %-44s %15lu %15lu\n",
 959		   trace_event_name(&tk->tp.call),
 960		   trace_kprobe_nhit(tk),
 961		   tk->rp.kp.nmissed);
 962
 963	return 0;
 964}
 965
 966static const struct seq_operations profile_seq_op = {
 967	.start  = probes_seq_start,
 968	.next   = probes_seq_next,
 969	.stop   = probes_seq_stop,
 970	.show   = probes_profile_seq_show
 971};
 972
 973static int profile_open(struct inode *inode, struct file *file)
 974{
 975	return seq_open(file, &profile_seq_op);
 976}
 977
 978static const struct file_operations kprobe_profile_ops = {
 979	.owner          = THIS_MODULE,
 980	.open           = profile_open,
 981	.read           = seq_read,
 982	.llseek         = seq_lseek,
 983	.release        = seq_release,
 984};
 985
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 986/* Kprobe handler */
 987static nokprobe_inline void
 988__kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
 989		    struct trace_event_file *trace_file)
 990{
 
 991	struct kprobe_trace_entry_head *entry;
 992	struct ring_buffer_event *event;
 993	struct ring_buffer *buffer;
 994	int size, dsize, pc;
 995	unsigned long irq_flags;
 996	struct trace_event_call *call = &tk->tp.call;
 997
 998	WARN_ON(call != trace_file->event_call);
 999
1000	if (trace_trigger_soft_disabled(trace_file))
1001		return;
1002
1003	local_save_flags(irq_flags);
1004	pc = preempt_count();
1005
1006	dsize = __get_data_size(&tk->tp, regs);
1007	size = sizeof(*entry) + tk->tp.size + dsize;
1008
1009	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1010						call->event.type,
1011						size, irq_flags, pc);
1012	if (!event)
1013		return;
1014
1015	entry = ring_buffer_event_data(event);
1016	entry->ip = (unsigned long)tk->rp.kp.addr;
1017	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1018
1019	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1020					 entry, irq_flags, pc, regs);
1021}
1022
1023static void
1024kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1025{
1026	struct event_file_link *link;
1027
1028	list_for_each_entry_rcu(link, &tk->tp.files, list)
1029		__kprobe_trace_func(tk, regs, link->file);
1030}
1031NOKPROBE_SYMBOL(kprobe_trace_func);
1032
1033/* Kretprobe handler */
1034static nokprobe_inline void
1035__kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1036		       struct pt_regs *regs,
1037		       struct trace_event_file *trace_file)
1038{
 
1039	struct kretprobe_trace_entry_head *entry;
1040	struct ring_buffer_event *event;
1041	struct ring_buffer *buffer;
1042	int size, pc, dsize;
1043	unsigned long irq_flags;
1044	struct trace_event_call *call = &tk->tp.call;
1045
1046	WARN_ON(call != trace_file->event_call);
1047
1048	if (trace_trigger_soft_disabled(trace_file))
1049		return;
1050
1051	local_save_flags(irq_flags);
1052	pc = preempt_count();
1053
1054	dsize = __get_data_size(&tk->tp, regs);
1055	size = sizeof(*entry) + tk->tp.size + dsize;
1056
1057	event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1058						call->event.type,
1059						size, irq_flags, pc);
1060	if (!event)
1061		return;
1062
1063	entry = ring_buffer_event_data(event);
1064	entry->func = (unsigned long)tk->rp.kp.addr;
1065	entry->ret_ip = (unsigned long)ri->ret_addr;
1066	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1067
1068	event_trigger_unlock_commit_regs(trace_file, buffer, event,
1069					 entry, irq_flags, pc, regs);
1070}
1071
1072static void
1073kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1074		     struct pt_regs *regs)
1075{
1076	struct event_file_link *link;
1077
1078	list_for_each_entry_rcu(link, &tk->tp.files, list)
1079		__kretprobe_trace_func(tk, ri, regs, link->file);
 
1080}
1081NOKPROBE_SYMBOL(kretprobe_trace_func);
1082
1083/* Event entry printers */
1084static enum print_line_t
1085print_kprobe_event(struct trace_iterator *iter, int flags,
1086		   struct trace_event *event)
1087{
1088	struct kprobe_trace_entry_head *field;
1089	struct trace_seq *s = &iter->seq;
1090	struct trace_probe *tp;
1091	u8 *data;
1092	int i;
1093
1094	field = (struct kprobe_trace_entry_head *)iter->ent;
1095	tp = container_of(event, struct trace_probe, call.event);
1096
1097	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
 
1098
1099	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1100		goto out;
1101
1102	trace_seq_putc(s, ')');
 
1103
1104	data = (u8 *)&field[1];
1105	for (i = 0; i < tp->nr_args; i++)
1106		if (!tp->args[i].type->print(s, tp->args[i].name,
1107					     data + tp->args[i].offset, field))
1108			goto out;
 
 
 
1109
1110	trace_seq_putc(s, '\n');
1111 out:
1112	return trace_handle_return(s);
1113}
1114
1115static enum print_line_t
1116print_kretprobe_event(struct trace_iterator *iter, int flags,
1117		      struct trace_event *event)
1118{
1119	struct kretprobe_trace_entry_head *field;
1120	struct trace_seq *s = &iter->seq;
1121	struct trace_probe *tp;
1122	u8 *data;
1123	int i;
1124
1125	field = (struct kretprobe_trace_entry_head *)iter->ent;
1126	tp = container_of(event, struct trace_probe, call.event);
1127
1128	trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
 
1129
1130	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1131		goto out;
1132
1133	trace_seq_puts(s, " <- ");
 
1134
1135	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1136		goto out;
1137
1138	trace_seq_putc(s, ')');
 
1139
1140	data = (u8 *)&field[1];
1141	for (i = 0; i < tp->nr_args; i++)
1142		if (!tp->args[i].type->print(s, tp->args[i].name,
1143					     data + tp->args[i].offset, field))
1144			goto out;
1145
1146	trace_seq_putc(s, '\n');
1147
1148 out:
1149	return trace_handle_return(s);
1150}
1151
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1152
1153static int kprobe_event_define_fields(struct trace_event_call *event_call)
1154{
1155	int ret, i;
1156	struct kprobe_trace_entry_head field;
1157	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1158
1159	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1160	/* Set argument names as fields */
1161	for (i = 0; i < tk->tp.nr_args; i++) {
1162		struct probe_arg *parg = &tk->tp.args[i];
1163
1164		ret = trace_define_field(event_call, parg->type->fmttype,
1165					 parg->name,
1166					 sizeof(field) + parg->offset,
1167					 parg->type->size,
1168					 parg->type->is_signed,
1169					 FILTER_OTHER);
1170		if (ret)
1171			return ret;
1172	}
1173	return 0;
1174}
1175
1176static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1177{
1178	int ret, i;
1179	struct kretprobe_trace_entry_head field;
1180	struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1181
1182	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1183	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1184	/* Set argument names as fields */
1185	for (i = 0; i < tk->tp.nr_args; i++) {
1186		struct probe_arg *parg = &tk->tp.args[i];
1187
1188		ret = trace_define_field(event_call, parg->type->fmttype,
1189					 parg->name,
1190					 sizeof(field) + parg->offset,
1191					 parg->type->size,
1192					 parg->type->is_signed,
1193					 FILTER_OTHER);
1194		if (ret)
1195			return ret;
1196	}
1197	return 0;
1198}
1199
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1200#ifdef CONFIG_PERF_EVENTS
1201
1202/* Kprobe profile handler */
1203static int
1204kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1205{
1206	struct trace_event_call *call = &tk->tp.call;
 
1207	struct kprobe_trace_entry_head *entry;
1208	struct hlist_head *head;
1209	int size, __size, dsize;
1210	int rctx;
1211
1212	if (bpf_prog_array_valid(call)) {
1213		unsigned long orig_ip = instruction_pointer(regs);
1214		int ret;
1215
1216		ret = trace_call_bpf(call, regs);
1217
1218		/*
1219		 * We need to check and see if we modified the pc of the
1220		 * pt_regs, and if so clear the kprobe and return 1 so that we
1221		 * don't do the single stepping.
1222		 * The ftrace kprobe handler leaves it up to us to re-enable
1223		 * preemption here before returning if we've modified the ip.
1224		 */
1225		if (orig_ip != instruction_pointer(regs)) {
1226			reset_current_kprobe();
1227			preempt_enable_no_resched();
1228			return 1;
1229		}
1230		if (!ret)
1231			return 0;
1232	}
1233
1234	head = this_cpu_ptr(call->perf_events);
1235	if (hlist_empty(head))
1236		return 0;
1237
1238	dsize = __get_data_size(&tk->tp, regs);
1239	__size = sizeof(*entry) + tk->tp.size + dsize;
1240	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1241	size -= sizeof(u32);
 
 
 
1242
1243	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1244	if (!entry)
1245		return 0;
1246
1247	entry->ip = (unsigned long)tk->rp.kp.addr;
1248	memset(&entry[1], 0, dsize);
1249	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1250	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1251			      head, NULL);
1252	return 0;
1253}
1254NOKPROBE_SYMBOL(kprobe_perf_func);
1255
1256/* Kretprobe profile handler */
1257static void
1258kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1259		    struct pt_regs *regs)
1260{
1261	struct trace_event_call *call = &tk->tp.call;
 
1262	struct kretprobe_trace_entry_head *entry;
1263	struct hlist_head *head;
1264	int size, __size, dsize;
1265	int rctx;
1266
1267	if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1268		return;
1269
1270	head = this_cpu_ptr(call->perf_events);
1271	if (hlist_empty(head))
1272		return;
1273
1274	dsize = __get_data_size(&tk->tp, regs);
1275	__size = sizeof(*entry) + tk->tp.size + dsize;
1276	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1277	size -= sizeof(u32);
 
 
 
1278
1279	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1280	if (!entry)
1281		return;
1282
1283	entry->func = (unsigned long)tk->rp.kp.addr;
1284	entry->ret_ip = (unsigned long)ri->ret_addr;
1285	store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1286	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1287			      head, NULL);
 
1288}
1289NOKPROBE_SYMBOL(kretprobe_perf_func);
1290#endif	/* CONFIG_PERF_EVENTS */
1291
1292/*
1293 * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1294 *
1295 * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1296 * lockless, but we can't race with this __init function.
1297 */
1298static int kprobe_register(struct trace_event_call *event,
1299			   enum trace_reg type, void *data)
1300{
1301	struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1302	struct trace_event_file *file = data;
1303
1304	switch (type) {
1305	case TRACE_REG_REGISTER:
1306		return enable_trace_kprobe(tk, file);
1307	case TRACE_REG_UNREGISTER:
1308		return disable_trace_kprobe(tk, file);
 
1309
1310#ifdef CONFIG_PERF_EVENTS
1311	case TRACE_REG_PERF_REGISTER:
1312		return enable_trace_kprobe(tk, NULL);
1313	case TRACE_REG_PERF_UNREGISTER:
1314		return disable_trace_kprobe(tk, NULL);
1315	case TRACE_REG_PERF_OPEN:
1316	case TRACE_REG_PERF_CLOSE:
1317	case TRACE_REG_PERF_ADD:
1318	case TRACE_REG_PERF_DEL:
1319		return 0;
1320#endif
1321	}
1322	return 0;
1323}
1324
1325static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
 
1326{
1327	struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1328	int ret = 0;
1329
1330	raw_cpu_inc(*tk->nhit);
1331
1332	if (tk->tp.flags & TP_FLAG_TRACE)
1333		kprobe_trace_func(tk, regs);
1334#ifdef CONFIG_PERF_EVENTS
1335	if (tk->tp.flags & TP_FLAG_PROFILE)
1336		ret = kprobe_perf_func(tk, regs);
1337#endif
1338	return ret;
1339}
1340NOKPROBE_SYMBOL(kprobe_dispatcher);
1341
1342static int
1343kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1344{
1345	struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1346
1347	raw_cpu_inc(*tk->nhit);
1348
1349	if (tk->tp.flags & TP_FLAG_TRACE)
1350		kretprobe_trace_func(tk, ri, regs);
1351#ifdef CONFIG_PERF_EVENTS
1352	if (tk->tp.flags & TP_FLAG_PROFILE)
1353		kretprobe_perf_func(tk, ri, regs);
1354#endif
1355	return 0;	/* We don't tweek kernel, so just return 0 */
1356}
1357NOKPROBE_SYMBOL(kretprobe_dispatcher);
1358
1359static struct trace_event_functions kretprobe_funcs = {
1360	.trace		= print_kretprobe_event
1361};
1362
1363static struct trace_event_functions kprobe_funcs = {
1364	.trace		= print_kprobe_event
1365};
1366
1367static inline void init_trace_event_call(struct trace_kprobe *tk,
1368					 struct trace_event_call *call)
1369{
 
 
 
 
1370	INIT_LIST_HEAD(&call->class->fields);
1371	if (trace_kprobe_is_return(tk)) {
1372		call->event.funcs = &kretprobe_funcs;
1373		call->class->define_fields = kretprobe_event_define_fields;
1374	} else {
1375		call->event.funcs = &kprobe_funcs;
1376		call->class->define_fields = kprobe_event_define_fields;
1377	}
1378
1379	call->flags = TRACE_EVENT_FL_KPROBE;
1380	call->class->reg = kprobe_register;
1381	call->data = tk;
1382}
1383
1384static int register_kprobe_event(struct trace_kprobe *tk)
1385{
1386	struct trace_event_call *call = &tk->tp.call;
1387	int ret = 0;
1388
1389	init_trace_event_call(tk, call);
1390
1391	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1392		return -ENOMEM;
1393	ret = register_trace_event(&call->event);
1394	if (!ret) {
1395		kfree(call->print_fmt);
1396		return -ENODEV;
1397	}
 
 
 
1398	ret = trace_add_event_call(call);
1399	if (ret) {
1400		pr_info("Failed to register kprobe event: %s\n",
1401			trace_event_name(call));
1402		kfree(call->print_fmt);
1403		unregister_trace_event(&call->event);
1404	}
1405	return ret;
1406}
1407
1408static int unregister_kprobe_event(struct trace_kprobe *tk)
1409{
1410	int ret;
1411
1412	/* tp->event is unregistered in trace_remove_event_call() */
1413	ret = trace_remove_event_call(&tk->tp.call);
1414	if (!ret)
1415		kfree(tk->tp.call.print_fmt);
1416	return ret;
1417}
1418
1419#ifdef CONFIG_PERF_EVENTS
1420/* create a trace_kprobe, but don't add it to global lists */
1421struct trace_event_call *
1422create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1423			  bool is_return)
1424{
1425	struct trace_kprobe *tk;
1426	int ret;
1427	char *event;
1428
1429	/*
1430	 * local trace_kprobes are not added to probe_list, so they are never
1431	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1432	 * duplicated name here.
1433	 */
1434	event = func ? func : "DUMMY_EVENT";
1435
1436	tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1437				offs, 0 /* maxactive */, 0 /* nargs */,
1438				is_return);
1439
1440	if (IS_ERR(tk)) {
1441		pr_info("Failed to allocate trace_probe.(%d)\n",
1442			(int)PTR_ERR(tk));
1443		return ERR_CAST(tk);
1444	}
1445
1446	init_trace_event_call(tk, &tk->tp.call);
1447
1448	if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1449		ret = -ENOMEM;
1450		goto error;
1451	}
1452
1453	ret = __register_trace_kprobe(tk);
1454	if (ret < 0)
1455		goto error;
1456
1457	return &tk->tp.call;
1458error:
1459	free_trace_kprobe(tk);
1460	return ERR_PTR(ret);
1461}
1462
1463void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1464{
1465	struct trace_kprobe *tk;
1466
1467	tk = container_of(event_call, struct trace_kprobe, tp.call);
1468
1469	if (trace_probe_is_enabled(&tk->tp)) {
1470		WARN_ON(1);
1471		return;
1472	}
1473
1474	__unregister_trace_kprobe(tk);
1475	free_trace_kprobe(tk);
1476}
1477#endif /* CONFIG_PERF_EVENTS */
1478
1479/* Make a tracefs interface for controlling probe points */
1480static __init int init_kprobe_trace(void)
1481{
1482	struct dentry *d_tracer;
1483	struct dentry *entry;
1484
1485	if (register_module_notifier(&trace_kprobe_module_nb))
1486		return -EINVAL;
1487
1488	d_tracer = tracing_init_dentry();
1489	if (IS_ERR(d_tracer))
1490		return 0;
1491
1492	entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1493				    NULL, &kprobe_events_ops);
1494
1495	/* Event list interface */
1496	if (!entry)
1497		pr_warn("Could not create tracefs 'kprobe_events' entry\n");
 
1498
1499	/* Profile interface */
1500	entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1501				    NULL, &kprobe_profile_ops);
1502
1503	if (!entry)
1504		pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
 
1505	return 0;
1506}
1507fs_initcall(init_kprobe_trace);
1508
1509
1510#ifdef CONFIG_FTRACE_STARTUP_TEST
 
1511/*
1512 * The "__used" keeps gcc from removing the function symbol
1513 * from the kallsyms table. 'noinline' makes sure that there
1514 * isn't an inlined version used by the test method below
1515 */
1516static __used __init noinline int
1517kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1518{
1519	return a1 + a2 + a3 + a4 + a5 + a6;
1520}
1521
1522static __init struct trace_event_file *
1523find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1524{
1525	struct trace_event_file *file;
1526
1527	list_for_each_entry(file, &tr->events, list)
1528		if (file->event_call == &tk->tp.call)
1529			return file;
1530
1531	return NULL;
1532}
1533
1534/*
1535 * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1536 * stage, we can do this lockless.
1537 */
1538static __init int kprobe_trace_self_tests_init(void)
1539{
1540	int ret, warn = 0;
1541	int (*target)(int, int, int, int, int, int);
1542	struct trace_kprobe *tk;
1543	struct trace_event_file *file;
1544
1545	if (tracing_is_disabled())
1546		return -ENODEV;
1547
1548	target = kprobe_trace_selftest_target;
1549
1550	pr_info("Testing kprobe tracing: ");
1551
1552	ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1553				"$stack $stack0 +0($stack)",
1554				create_trace_kprobe);
1555	if (WARN_ON_ONCE(ret)) {
1556		pr_warn("error on probing function entry.\n");
1557		warn++;
1558	} else {
1559		/* Enable trace point */
1560		tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1561		if (WARN_ON_ONCE(tk == NULL)) {
1562			pr_warn("error on getting new probe.\n");
1563			warn++;
1564		} else {
1565			file = find_trace_probe_file(tk, top_trace_array());
1566			if (WARN_ON_ONCE(file == NULL)) {
1567				pr_warn("error on getting probe file.\n");
1568				warn++;
1569			} else
1570				enable_trace_kprobe(tk, file);
1571		}
1572	}
1573
1574	ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1575				"$retval", create_trace_kprobe);
1576	if (WARN_ON_ONCE(ret)) {
1577		pr_warn("error on probing function return.\n");
1578		warn++;
1579	} else {
1580		/* Enable trace point */
1581		tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1582		if (WARN_ON_ONCE(tk == NULL)) {
1583			pr_warn("error on getting 2nd new probe.\n");
1584			warn++;
1585		} else {
1586			file = find_trace_probe_file(tk, top_trace_array());
1587			if (WARN_ON_ONCE(file == NULL)) {
1588				pr_warn("error on getting probe file.\n");
1589				warn++;
1590			} else
1591				enable_trace_kprobe(tk, file);
1592		}
1593	}
1594
1595	if (warn)
1596		goto end;
1597
1598	ret = target(1, 2, 3, 4, 5, 6);
1599
1600	/*
1601	 * Not expecting an error here, the check is only to prevent the
1602	 * optimizer from removing the call to target() as otherwise there
1603	 * are no side-effects and the call is never performed.
1604	 */
1605	if (ret != 21)
1606		warn++;
1607
1608	/* Disable trace points before removing it */
1609	tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1610	if (WARN_ON_ONCE(tk == NULL)) {
1611		pr_warn("error on getting test probe.\n");
1612		warn++;
1613	} else {
1614		if (trace_kprobe_nhit(tk) != 1) {
1615			pr_warn("incorrect number of testprobe hits\n");
1616			warn++;
1617		}
1618
1619		file = find_trace_probe_file(tk, top_trace_array());
1620		if (WARN_ON_ONCE(file == NULL)) {
1621			pr_warn("error on getting probe file.\n");
1622			warn++;
1623		} else
1624			disable_trace_kprobe(tk, file);
1625	}
1626
1627	tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1628	if (WARN_ON_ONCE(tk == NULL)) {
1629		pr_warn("error on getting 2nd test probe.\n");
1630		warn++;
1631	} else {
1632		if (trace_kprobe_nhit(tk) != 1) {
1633			pr_warn("incorrect number of testprobe2 hits\n");
1634			warn++;
1635		}
1636
1637		file = find_trace_probe_file(tk, top_trace_array());
1638		if (WARN_ON_ONCE(file == NULL)) {
1639			pr_warn("error on getting probe file.\n");
1640			warn++;
1641		} else
1642			disable_trace_kprobe(tk, file);
1643	}
1644
1645	ret = trace_run_command("-:testprobe", create_trace_kprobe);
1646	if (WARN_ON_ONCE(ret)) {
1647		pr_warn("error on deleting a probe.\n");
1648		warn++;
1649	}
1650
1651	ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1652	if (WARN_ON_ONCE(ret)) {
1653		pr_warn("error on deleting a probe.\n");
1654		warn++;
1655	}
1656
1657end:
1658	release_all_trace_kprobes();
1659	/*
1660	 * Wait for the optimizer work to finish. Otherwise it might fiddle
1661	 * with probes in already freed __init text.
1662	 */
1663	wait_for_kprobe_optimizer();
1664	if (warn)
1665		pr_cont("NG: Some tests are failed. Please check them.\n");
1666	else
1667		pr_cont("OK\n");
1668	return 0;
1669}
1670
1671late_initcall(kprobe_trace_self_tests_init);
1672
1673#endif
v3.1
   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#include <linux/kprobes.h>
  23#include <linux/seq_file.h>
  24#include <linux/slab.h>
  25#include <linux/smp.h>
  26#include <linux/debugfs.h>
  27#include <linux/types.h>
  28#include <linux/string.h>
  29#include <linux/ctype.h>
  30#include <linux/ptrace.h>
  31#include <linux/perf_event.h>
  32#include <linux/stringify.h>
  33#include <linux/limits.h>
  34#include <asm/bitsperlong.h>
  35
  36#include "trace.h"
  37#include "trace_output.h"
  38
  39#define MAX_TRACE_ARGS 128
  40#define MAX_ARGSTR_LEN 63
  41#define MAX_EVENT_NAME_LEN 64
  42#define MAX_STRING_SIZE PATH_MAX
  43#define KPROBE_EVENT_SYSTEM "kprobes"
 
  44
  45/* Reserved field names */
  46#define FIELD_STRING_IP "__probe_ip"
  47#define FIELD_STRING_RETIP "__probe_ret_ip"
  48#define FIELD_STRING_FUNC "__probe_func"
  49
  50const char *reserved_field_names[] = {
  51	"common_type",
  52	"common_flags",
  53	"common_preempt_count",
  54	"common_pid",
  55	"common_tgid",
  56	FIELD_STRING_IP,
  57	FIELD_STRING_RETIP,
  58	FIELD_STRING_FUNC,
  59};
  60
  61/* Printing function type */
  62typedef int (*print_type_func_t)(struct trace_seq *, const char *, void *,
  63				 void *);
  64#define PRINT_TYPE_FUNC_NAME(type)	print_type_##type
  65#define PRINT_TYPE_FMT_NAME(type)	print_type_format_##type
  66
  67/* Printing  in basic type function template */
  68#define DEFINE_BASIC_PRINT_TYPE_FUNC(type, fmt, cast)			\
  69static __kprobes int PRINT_TYPE_FUNC_NAME(type)(struct trace_seq *s,	\
  70						const char *name,	\
  71						void *data, void *ent)\
  72{									\
  73	return trace_seq_printf(s, " %s=" fmt, name, (cast)*(type *)data);\
  74}									\
  75static const char PRINT_TYPE_FMT_NAME(type)[] = fmt;
 
 
 
  76
  77DEFINE_BASIC_PRINT_TYPE_FUNC(u8, "%x", unsigned int)
  78DEFINE_BASIC_PRINT_TYPE_FUNC(u16, "%x", unsigned int)
  79DEFINE_BASIC_PRINT_TYPE_FUNC(u32, "%lx", unsigned long)
  80DEFINE_BASIC_PRINT_TYPE_FUNC(u64, "%llx", unsigned long long)
  81DEFINE_BASIC_PRINT_TYPE_FUNC(s8, "%d", int)
  82DEFINE_BASIC_PRINT_TYPE_FUNC(s16, "%d", int)
  83DEFINE_BASIC_PRINT_TYPE_FUNC(s32, "%ld", long)
  84DEFINE_BASIC_PRINT_TYPE_FUNC(s64, "%lld", long long)
  85
  86/* data_rloc: data relative location, compatible with u32 */
  87#define make_data_rloc(len, roffs)	\
  88	(((u32)(len) << 16) | ((u32)(roffs) & 0xffff))
  89#define get_rloc_len(dl)	((u32)(dl) >> 16)
  90#define get_rloc_offs(dl)	((u32)(dl) & 0xffff)
 
 
  91
  92static inline void *get_rloc_data(u32 *dl)
  93{
  94	return (u8 *)dl + get_rloc_offs(*dl);
  95}
  96
  97/* For data_loc conversion */
  98static inline void *get_loc_data(u32 *dl, void *ent)
  99{
 100	return (u8 *)ent + get_rloc_offs(*dl);
 
 
 
 
 
 
 101}
 102
 103/*
 104 * Convert data_rloc to data_loc:
 105 *  data_rloc stores the offset from data_rloc itself, but data_loc
 106 *  stores the offset from event entry.
 107 */
 108#define convert_rloc_to_loc(dl, offs)	((u32)(dl) + (offs))
 109
 110/* For defining macros, define string/string_size types */
 111typedef u32 string;
 112typedef u32 string_size;
 
 113
 114/* Print type function for string type */
 115static __kprobes int PRINT_TYPE_FUNC_NAME(string)(struct trace_seq *s,
 116						  const char *name,
 117						  void *data, void *ent)
 118{
 119	int len = *(u32 *)data >> 16;
 
 120
 121	if (!len)
 122		return trace_seq_printf(s, " %s=(fault)", name);
 123	else
 124		return trace_seq_printf(s, " %s=\"%s\"", name,
 125					(const char *)get_loc_data(data, ent));
 
 
 
 126}
 127static const char PRINT_TYPE_FMT_NAME(string)[] = "\\\"%s\\\"";
 128
 129/* Data fetch function type */
 130typedef	void (*fetch_func_t)(struct pt_regs *, void *, void *);
 131
 132struct fetch_param {
 133	fetch_func_t	fn;
 134	void *data;
 
 
 
 
 
 
 
 
 
 135};
 136
 137static __kprobes void call_fetch(struct fetch_param *fprm,
 138				 struct pt_regs *regs, void *dest)
 139{
 140	return fprm->fn(regs, fprm->data, dest);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 141}
 142
 143#define FETCH_FUNC_NAME(method, type)	fetch_##method##_##type
 144/*
 145 * Define macro for basic types - we don't need to define s* types, because
 146 * we have to care only about bitwidth at recording time.
 147 */
 148#define DEFINE_BASIC_FETCH_FUNCS(method) \
 149DEFINE_FETCH_##method(u8)		\
 150DEFINE_FETCH_##method(u16)		\
 151DEFINE_FETCH_##method(u32)		\
 152DEFINE_FETCH_##method(u64)
 153
 154#define CHECK_FETCH_FUNCS(method, fn)			\
 155	(((FETCH_FUNC_NAME(method, u8) == fn) ||	\
 156	  (FETCH_FUNC_NAME(method, u16) == fn) ||	\
 157	  (FETCH_FUNC_NAME(method, u32) == fn) ||	\
 158	  (FETCH_FUNC_NAME(method, u64) == fn) ||	\
 159	  (FETCH_FUNC_NAME(method, string) == fn) ||	\
 160	  (FETCH_FUNC_NAME(method, string_size) == fn)) \
 161	 && (fn != NULL))
 162
 163/* Data fetch function templates */
 164#define DEFINE_FETCH_reg(type)						\
 165static __kprobes void FETCH_FUNC_NAME(reg, type)(struct pt_regs *regs,	\
 166					void *offset, void *dest)	\
 167{									\
 168	*(type *)dest = (type)regs_get_register(regs,			\
 169				(unsigned int)((unsigned long)offset));	\
 170}
 171DEFINE_BASIC_FETCH_FUNCS(reg)
 172/* No string on the register */
 173#define fetch_reg_string NULL
 174#define fetch_reg_string_size NULL
 175
 176#define DEFINE_FETCH_stack(type)					\
 177static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
 178					  void *offset, void *dest)	\
 179{									\
 180	*(type *)dest = (type)regs_get_kernel_stack_nth(regs,		\
 181				(unsigned int)((unsigned long)offset));	\
 182}
 
 
 183DEFINE_BASIC_FETCH_FUNCS(stack)
 184/* No string on the stack entry */
 185#define fetch_stack_string NULL
 186#define fetch_stack_string_size NULL
 187
 188#define DEFINE_FETCH_retval(type)					\
 189static __kprobes void FETCH_FUNC_NAME(retval, type)(struct pt_regs *regs,\
 190					  void *dummy, void *dest)	\
 191{									\
 192	*(type *)dest = (type)regs_return_value(regs);			\
 193}
 194DEFINE_BASIC_FETCH_FUNCS(retval)
 195/* No string on the retval */
 196#define fetch_retval_string NULL
 197#define fetch_retval_string_size NULL
 198
 199#define DEFINE_FETCH_memory(type)					\
 200static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
 201					  void *addr, void *dest)	\
 202{									\
 203	type retval;							\
 204	if (probe_kernel_address(addr, retval))				\
 205		*(type *)dest = 0;					\
 206	else								\
 207		*(type *)dest = retval;					\
 208}
 
 
 209DEFINE_BASIC_FETCH_FUNCS(memory)
 210/*
 211 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 212 * length and relative data location.
 213 */
 214static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 215						      void *addr, void *dest)
 216{
 217	long ret;
 218	int maxlen = get_rloc_len(*(u32 *)dest);
 219	u8 *dst = get_rloc_data(dest);
 220	u8 *src = addr;
 221	mm_segment_t old_fs = get_fs();
 222	if (!maxlen)
 223		return;
 
 224	/*
 225	 * Try to get string again, since the string can be changed while
 226	 * probing.
 227	 */
 228	set_fs(KERNEL_DS);
 229	pagefault_disable();
 230	do
 231		ret = __copy_from_user_inatomic(dst++, src++, 1);
 232	while (dst[-1] && ret == 0 && src - (u8 *)addr < maxlen);
 233	dst[-1] = '\0';
 234	pagefault_enable();
 235	set_fs(old_fs);
 236
 237	if (ret < 0) {	/* Failed to fetch string */
 238		((u8 *)get_rloc_data(dest))[0] = '\0';
 239		*(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 240	} else
 241		*(u32 *)dest = make_data_rloc(src - (u8 *)addr,
 242					      get_rloc_offs(*(u32 *)dest));
 243}
 
 
 244/* Return the length of string -- including null terminal byte */
 245static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 246							void *addr, void *dest)
 247{
 
 248	int ret, len = 0;
 249	u8 c;
 250	mm_segment_t old_fs = get_fs();
 251
 
 252	set_fs(KERNEL_DS);
 253	pagefault_disable();
 
 254	do {
 255		ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 256		len++;
 257	} while (c && ret == 0 && len < MAX_STRING_SIZE);
 
 258	pagefault_enable();
 259	set_fs(old_fs);
 260
 261	if (ret < 0)	/* Failed to check the length */
 262		*(u32 *)dest = 0;
 263	else
 264		*(u32 *)dest = len;
 265}
 266
 267/* Memory fetching by symbol */
 268struct symbol_cache {
 269	char *symbol;
 270	long offset;
 271	unsigned long addr;
 272};
 273
 274static unsigned long update_symbol_cache(struct symbol_cache *sc)
 275{
 276	sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 277	if (sc->addr)
 278		sc->addr += sc->offset;
 279	return sc->addr;
 280}
 281
 282static void free_symbol_cache(struct symbol_cache *sc)
 283{
 284	kfree(sc->symbol);
 285	kfree(sc);
 286}
 287
 288static struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 289{
 290	struct symbol_cache *sc;
 291
 292	if (!sym || strlen(sym) == 0)
 293		return NULL;
 294	sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 295	if (!sc)
 296		return NULL;
 297
 298	sc->symbol = kstrdup(sym, GFP_KERNEL);
 299	if (!sc->symbol) {
 300		kfree(sc);
 301		return NULL;
 302	}
 303	sc->offset = offset;
 304
 305	update_symbol_cache(sc);
 306	return sc;
 307}
 308
 309#define DEFINE_FETCH_symbol(type)					\
 310static __kprobes void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs,\
 311					  void *data, void *dest)	\
 312{									\
 313	struct symbol_cache *sc = data;					\
 314	if (sc->addr)							\
 315		fetch_memory_##type(regs, (void *)sc->addr, dest);	\
 316	else								\
 317		*(type *)dest = 0;					\
 318}
 
 
 319DEFINE_BASIC_FETCH_FUNCS(symbol)
 320DEFINE_FETCH_symbol(string)
 321DEFINE_FETCH_symbol(string_size)
 322
 323/* Dereference memory access function */
 324struct deref_fetch_param {
 325	struct fetch_param orig;
 326	long offset;
 327};
 328
 329#define DEFINE_FETCH_deref(type)					\
 330static __kprobes void FETCH_FUNC_NAME(deref, type)(struct pt_regs *regs,\
 331					    void *data, void *dest)	\
 332{									\
 333	struct deref_fetch_param *dprm = data;				\
 334	unsigned long addr;						\
 335	call_fetch(&dprm->orig, regs, &addr);				\
 336	if (addr) {							\
 337		addr += dprm->offset;					\
 338		fetch_memory_##type(regs, (void *)addr, dest);		\
 339	} else								\
 340		*(type *)dest = 0;					\
 341}
 342DEFINE_BASIC_FETCH_FUNCS(deref)
 343DEFINE_FETCH_deref(string)
 344DEFINE_FETCH_deref(string_size)
 345
 346static __kprobes void update_deref_fetch_param(struct deref_fetch_param *data)
 347{
 348	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
 349		update_deref_fetch_param(data->orig.data);
 350	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
 351		update_symbol_cache(data->orig.data);
 352}
 353
 354static __kprobes void free_deref_fetch_param(struct deref_fetch_param *data)
 355{
 356	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
 357		free_deref_fetch_param(data->orig.data);
 358	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
 359		free_symbol_cache(data->orig.data);
 360	kfree(data);
 361}
 362
 363/* Bitfield fetch function */
 364struct bitfield_fetch_param {
 365	struct fetch_param orig;
 366	unsigned char hi_shift;
 367	unsigned char low_shift;
 368};
 369
 370#define DEFINE_FETCH_bitfield(type)					\
 371static __kprobes void FETCH_FUNC_NAME(bitfield, type)(struct pt_regs *regs,\
 372					    void *data, void *dest)	\
 373{									\
 374	struct bitfield_fetch_param *bprm = data;			\
 375	type buf = 0;							\
 376	call_fetch(&bprm->orig, regs, &buf);				\
 377	if (buf) {							\
 378		buf <<= bprm->hi_shift;					\
 379		buf >>= bprm->low_shift;				\
 380	}								\
 381	*(type *)dest = buf;						\
 382}
 383DEFINE_BASIC_FETCH_FUNCS(bitfield)
 384#define fetch_bitfield_string NULL
 385#define fetch_bitfield_string_size NULL
 386
 387static __kprobes void
 388update_bitfield_fetch_param(struct bitfield_fetch_param *data)
 389{
 390	/*
 391	 * Don't check the bitfield itself, because this must be the
 392	 * last fetch function.
 393	 */
 394	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
 395		update_deref_fetch_param(data->orig.data);
 396	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
 397		update_symbol_cache(data->orig.data);
 398}
 399
 400static __kprobes void
 401free_bitfield_fetch_param(struct bitfield_fetch_param *data)
 402{
 403	/*
 404	 * Don't check the bitfield itself, because this must be the
 405	 * last fetch function.
 406	 */
 407	if (CHECK_FETCH_FUNCS(deref, data->orig.fn))
 408		free_deref_fetch_param(data->orig.data);
 409	else if (CHECK_FETCH_FUNCS(symbol, data->orig.fn))
 410		free_symbol_cache(data->orig.data);
 411	kfree(data);
 412}
 413
 414/* Default (unsigned long) fetch type */
 415#define __DEFAULT_FETCH_TYPE(t) u##t
 416#define _DEFAULT_FETCH_TYPE(t) __DEFAULT_FETCH_TYPE(t)
 417#define DEFAULT_FETCH_TYPE _DEFAULT_FETCH_TYPE(BITS_PER_LONG)
 418#define DEFAULT_FETCH_TYPE_STR __stringify(DEFAULT_FETCH_TYPE)
 419
 420/* Fetch types */
 421enum {
 422	FETCH_MTD_reg = 0,
 423	FETCH_MTD_stack,
 424	FETCH_MTD_retval,
 425	FETCH_MTD_memory,
 426	FETCH_MTD_symbol,
 427	FETCH_MTD_deref,
 428	FETCH_MTD_bitfield,
 429	FETCH_MTD_END,
 430};
 431
 432#define ASSIGN_FETCH_FUNC(method, type)	\
 433	[FETCH_MTD_##method] = FETCH_FUNC_NAME(method, type)
 434
 435#define __ASSIGN_FETCH_TYPE(_name, ptype, ftype, _size, sign, _fmttype)	\
 436	{.name = _name,				\
 437	 .size = _size,					\
 438	 .is_signed = sign,				\
 439	 .print = PRINT_TYPE_FUNC_NAME(ptype),		\
 440	 .fmt = PRINT_TYPE_FMT_NAME(ptype),		\
 441	 .fmttype = _fmttype,				\
 442	 .fetch = {					\
 443ASSIGN_FETCH_FUNC(reg, ftype),				\
 444ASSIGN_FETCH_FUNC(stack, ftype),			\
 445ASSIGN_FETCH_FUNC(retval, ftype),			\
 446ASSIGN_FETCH_FUNC(memory, ftype),			\
 447ASSIGN_FETCH_FUNC(symbol, ftype),			\
 448ASSIGN_FETCH_FUNC(deref, ftype),			\
 449ASSIGN_FETCH_FUNC(bitfield, ftype),			\
 450	  }						\
 451	}
 452
 453#define ASSIGN_FETCH_TYPE(ptype, ftype, sign)			\
 454	__ASSIGN_FETCH_TYPE(#ptype, ptype, ftype, sizeof(ftype), sign, #ptype)
 455
 456#define FETCH_TYPE_STRING 0
 457#define FETCH_TYPE_STRSIZE 1
 458
 459/* Fetch type information table */
 460static const struct fetch_type {
 461	const char	*name;		/* Name of type */
 462	size_t		size;		/* Byte size of type */
 463	int		is_signed;	/* Signed flag */
 464	print_type_func_t	print;	/* Print functions */
 465	const char	*fmt;		/* Fromat string */
 466	const char	*fmttype;	/* Name in format file */
 467	/* Fetch functions */
 468	fetch_func_t	fetch[FETCH_MTD_END];
 469} fetch_type_table[] = {
 470	/* Special types */
 471	[FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 472					sizeof(u32), 1, "__data_loc char[]"),
 473	[FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 474					string_size, sizeof(u32), 0, "u32"),
 475	/* Basic types */
 476	ASSIGN_FETCH_TYPE(u8,  u8,  0),
 477	ASSIGN_FETCH_TYPE(u16, u16, 0),
 478	ASSIGN_FETCH_TYPE(u32, u32, 0),
 479	ASSIGN_FETCH_TYPE(u64, u64, 0),
 480	ASSIGN_FETCH_TYPE(s8,  u8,  1),
 481	ASSIGN_FETCH_TYPE(s16, u16, 1),
 482	ASSIGN_FETCH_TYPE(s32, u32, 1),
 483	ASSIGN_FETCH_TYPE(s64, u64, 1),
 484};
 485
 486static const struct fetch_type *find_fetch_type(const char *type)
 487{
 488	int i;
 489
 490	if (!type)
 491		type = DEFAULT_FETCH_TYPE_STR;
 492
 493	/* Special case: bitfield */
 494	if (*type == 'b') {
 495		unsigned long bs;
 496		type = strchr(type, '/');
 497		if (!type)
 498			goto fail;
 499		type++;
 500		if (strict_strtoul(type, 0, &bs))
 501			goto fail;
 502		switch (bs) {
 503		case 8:
 504			return find_fetch_type("u8");
 505		case 16:
 506			return find_fetch_type("u16");
 507		case 32:
 508			return find_fetch_type("u32");
 509		case 64:
 510			return find_fetch_type("u64");
 511		default:
 512			goto fail;
 513		}
 514	}
 515
 516	for (i = 0; i < ARRAY_SIZE(fetch_type_table); i++)
 517		if (strcmp(type, fetch_type_table[i].name) == 0)
 518			return &fetch_type_table[i];
 519fail:
 520	return NULL;
 521}
 522
 523/* Special function : only accept unsigned long */
 524static __kprobes void fetch_stack_address(struct pt_regs *regs,
 525					  void *dummy, void *dest)
 526{
 527	*(unsigned long *)dest = kernel_stack_pointer(regs);
 528}
 529
 530static fetch_func_t get_fetch_size_function(const struct fetch_type *type,
 531					    fetch_func_t orig_fn)
 532{
 533	int i;
 534
 535	if (type != &fetch_type_table[FETCH_TYPE_STRING])
 536		return NULL;	/* Only string type needs size function */
 537	for (i = 0; i < FETCH_MTD_END; i++)
 538		if (type->fetch[i] == orig_fn)
 539			return fetch_type_table[FETCH_TYPE_STRSIZE].fetch[i];
 540
 541	WARN_ON(1);	/* This should not happen */
 542	return NULL;
 543}
 544
 545/**
 546 * Kprobe event core functions
 547 */
 548
 549struct probe_arg {
 550	struct fetch_param	fetch;
 551	struct fetch_param	fetch_size;
 552	unsigned int		offset;	/* Offset from argument entry */
 553	const char		*name;	/* Name of this argument */
 554	const char		*comm;	/* Command of this argument */
 555	const struct fetch_type	*type;	/* Type of this argument */
 556};
 557
 558/* Flags for trace_probe */
 559#define TP_FLAG_TRACE	1
 560#define TP_FLAG_PROFILE	2
 561#define TP_FLAG_REGISTERED 4
 562
 563struct trace_probe {
 564	struct list_head	list;
 565	struct kretprobe	rp;	/* Use rp.kp for kprobe use */
 566	unsigned long 		nhit;
 567	unsigned int		flags;	/* For TP_FLAG_* */
 568	const char		*symbol;	/* symbol name */
 569	struct ftrace_event_class	class;
 570	struct ftrace_event_call	call;
 571	ssize_t			size;		/* trace entry size */
 572	unsigned int		nr_args;
 573	struct probe_arg	args[];
 574};
 575
 576#define SIZEOF_TRACE_PROBE(n)			\
 577	(offsetof(struct trace_probe, args) +	\
 578	(sizeof(struct probe_arg) * (n)))
 579
 580
 581static __kprobes int trace_probe_is_return(struct trace_probe *tp)
 582{
 583	return tp->rp.handler != NULL;
 584}
 585
 586static __kprobes const char *trace_probe_symbol(struct trace_probe *tp)
 587{
 588	return tp->symbol ? tp->symbol : "unknown";
 589}
 590
 591static __kprobes unsigned long trace_probe_offset(struct trace_probe *tp)
 592{
 593	return tp->rp.kp.offset;
 594}
 595
 596static __kprobes bool trace_probe_is_enabled(struct trace_probe *tp)
 597{
 598	return !!(tp->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE));
 599}
 600
 601static __kprobes bool trace_probe_is_registered(struct trace_probe *tp)
 602{
 603	return !!(tp->flags & TP_FLAG_REGISTERED);
 604}
 605
 606static __kprobes bool trace_probe_has_gone(struct trace_probe *tp)
 607{
 608	return !!(kprobe_gone(&tp->rp.kp));
 609}
 610
 611static __kprobes bool trace_probe_within_module(struct trace_probe *tp,
 612						struct module *mod)
 613{
 614	int len = strlen(mod->name);
 615	const char *name = trace_probe_symbol(tp);
 616	return strncmp(mod->name, name, len) == 0 && name[len] == ':';
 617}
 618
 619static __kprobes bool trace_probe_is_on_module(struct trace_probe *tp)
 620{
 621	return !!strchr(trace_probe_symbol(tp), ':');
 622}
 623
 624static int register_probe_event(struct trace_probe *tp);
 625static void unregister_probe_event(struct trace_probe *tp);
 626
 627static DEFINE_MUTEX(probe_lock);
 628static LIST_HEAD(probe_list);
 629
 630static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 631static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 632				struct pt_regs *regs);
 633
 634/* Check the name is good for event/group/fields */
 635static int is_good_name(const char *name)
 636{
 637	if (!isalpha(*name) && *name != '_')
 638		return 0;
 639	while (*++name != '\0') {
 640		if (!isalpha(*name) && !isdigit(*name) && *name != '_')
 641			return 0;
 642	}
 643	return 1;
 644}
 645
 646/*
 647 * Allocate new trace_probe and initialize it (including kprobes).
 648 */
 649static struct trace_probe *alloc_trace_probe(const char *group,
 650					     const char *event,
 651					     void *addr,
 652					     const char *symbol,
 653					     unsigned long offs,
 654					     int nargs, int is_return)
 
 655{
 656	struct trace_probe *tp;
 657	int ret = -ENOMEM;
 658
 659	tp = kzalloc(SIZEOF_TRACE_PROBE(nargs), GFP_KERNEL);
 660	if (!tp)
 661		return ERR_PTR(ret);
 662
 
 
 
 
 663	if (symbol) {
 664		tp->symbol = kstrdup(symbol, GFP_KERNEL);
 665		if (!tp->symbol)
 666			goto error;
 667		tp->rp.kp.symbol_name = tp->symbol;
 668		tp->rp.kp.offset = offs;
 669	} else
 670		tp->rp.kp.addr = addr;
 671
 672	if (is_return)
 673		tp->rp.handler = kretprobe_dispatcher;
 674	else
 675		tp->rp.kp.pre_handler = kprobe_dispatcher;
 
 
 676
 677	if (!event || !is_good_name(event)) {
 678		ret = -EINVAL;
 679		goto error;
 680	}
 681
 682	tp->call.class = &tp->class;
 683	tp->call.name = kstrdup(event, GFP_KERNEL);
 684	if (!tp->call.name)
 685		goto error;
 686
 687	if (!group || !is_good_name(group)) {
 688		ret = -EINVAL;
 689		goto error;
 690	}
 691
 692	tp->class.system = kstrdup(group, GFP_KERNEL);
 693	if (!tp->class.system)
 694		goto error;
 695
 696	INIT_LIST_HEAD(&tp->list);
 697	return tp;
 
 698error:
 699	kfree(tp->call.name);
 700	kfree(tp->symbol);
 701	kfree(tp);
 
 702	return ERR_PTR(ret);
 703}
 704
 705static void update_probe_arg(struct probe_arg *arg)
 706{
 707	if (CHECK_FETCH_FUNCS(bitfield, arg->fetch.fn))
 708		update_bitfield_fetch_param(arg->fetch.data);
 709	else if (CHECK_FETCH_FUNCS(deref, arg->fetch.fn))
 710		update_deref_fetch_param(arg->fetch.data);
 711	else if (CHECK_FETCH_FUNCS(symbol, arg->fetch.fn))
 712		update_symbol_cache(arg->fetch.data);
 713}
 714
 715static void free_probe_arg(struct probe_arg *arg)
 716{
 717	if (CHECK_FETCH_FUNCS(bitfield, arg->fetch.fn))
 718		free_bitfield_fetch_param(arg->fetch.data);
 719	else if (CHECK_FETCH_FUNCS(deref, arg->fetch.fn))
 720		free_deref_fetch_param(arg->fetch.data);
 721	else if (CHECK_FETCH_FUNCS(symbol, arg->fetch.fn))
 722		free_symbol_cache(arg->fetch.data);
 723	kfree(arg->name);
 724	kfree(arg->comm);
 725}
 726
 727static void free_trace_probe(struct trace_probe *tp)
 728{
 729	int i;
 730
 731	for (i = 0; i < tp->nr_args; i++)
 732		free_probe_arg(&tp->args[i]);
 733
 734	kfree(tp->call.class->system);
 735	kfree(tp->call.name);
 736	kfree(tp->symbol);
 737	kfree(tp);
 
 738}
 739
 740static struct trace_probe *find_trace_probe(const char *event,
 741					    const char *group)
 742{
 743	struct trace_probe *tp;
 744
 745	list_for_each_entry(tp, &probe_list, list)
 746		if (strcmp(tp->call.name, event) == 0 &&
 747		    strcmp(tp->call.class->system, group) == 0)
 748			return tp;
 749	return NULL;
 750}
 751
 752/* Enable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
 753static int enable_trace_probe(struct trace_probe *tp, int flag)
 
 
 
 
 754{
 755	int ret = 0;
 756
 757	tp->flags |= flag;
 758	if (trace_probe_is_enabled(tp) && trace_probe_is_registered(tp) &&
 759	    !trace_probe_has_gone(tp)) {
 760		if (trace_probe_is_return(tp))
 761			ret = enable_kretprobe(&tp->rp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 762		else
 763			ret = enable_kprobe(&tp->rp.kp);
 764	}
 765
 766	return ret;
 767}
 768
 769/* Disable trace_probe - @flag must be TP_FLAG_TRACE or TP_FLAG_PROFILE */
 770static void disable_trace_probe(struct trace_probe *tp, int flag)
 
 
 
 
 771{
 772	tp->flags &= ~flag;
 773	if (!trace_probe_is_enabled(tp) && trace_probe_is_registered(tp)) {
 774		if (trace_probe_is_return(tp))
 775			disable_kretprobe(&tp->rp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 776		else
 777			disable_kprobe(&tp->rp.kp);
 
 778	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 779}
 780
 781/* Internal register function - just handle k*probes and flags */
 782static int __register_trace_probe(struct trace_probe *tp)
 783{
 784	int i, ret;
 785
 786	if (trace_probe_is_registered(tp))
 787		return -EINVAL;
 788
 789	for (i = 0; i < tp->nr_args; i++)
 790		update_probe_arg(&tp->args[i]);
 791
 792	/* Set/clear disabled flag according to tp->flag */
 793	if (trace_probe_is_enabled(tp))
 794		tp->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 795	else
 796		tp->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 797
 798	if (trace_probe_is_return(tp))
 799		ret = register_kretprobe(&tp->rp);
 800	else
 801		ret = register_kprobe(&tp->rp.kp);
 802
 803	if (ret == 0)
 804		tp->flags |= TP_FLAG_REGISTERED;
 805	else {
 806		pr_warning("Could not insert probe at %s+%lu: %d\n",
 807			   trace_probe_symbol(tp), trace_probe_offset(tp), ret);
 808		if (ret == -ENOENT && trace_probe_is_on_module(tp)) {
 809			pr_warning("This probe might be able to register after"
 810				   "target module is loaded. Continue.\n");
 811			ret = 0;
 812		} else if (ret == -EILSEQ) {
 813			pr_warning("Probing address(0x%p) is not an "
 814				   "instruction boundary.\n",
 815				   tp->rp.kp.addr);
 816			ret = -EINVAL;
 817		}
 818	}
 819
 820	return ret;
 821}
 822
 823/* Internal unregister function - just handle k*probes and flags */
 824static void __unregister_trace_probe(struct trace_probe *tp)
 825{
 826	if (trace_probe_is_registered(tp)) {
 827		if (trace_probe_is_return(tp))
 828			unregister_kretprobe(&tp->rp);
 829		else
 830			unregister_kprobe(&tp->rp.kp);
 831		tp->flags &= ~TP_FLAG_REGISTERED;
 832		/* Cleanup kprobe for reuse */
 833		if (tp->rp.kp.symbol_name)
 834			tp->rp.kp.addr = NULL;
 835	}
 836}
 837
 838/* Unregister a trace_probe and probe_event: call with locking probe_lock */
 839static void unregister_trace_probe(struct trace_probe *tp)
 840{
 841	__unregister_trace_probe(tp);
 842	list_del(&tp->list);
 843	unregister_probe_event(tp);
 
 
 
 
 
 
 
 
 
 844}
 845
 846/* Register a trace_probe and probe_event */
 847static int register_trace_probe(struct trace_probe *tp)
 848{
 849	struct trace_probe *old_tp;
 850	int ret;
 851
 852	mutex_lock(&probe_lock);
 853
 854	/* Delete old (same name) event if exist */
 855	old_tp = find_trace_probe(tp->call.name, tp->call.class->system);
 856	if (old_tp) {
 857		unregister_trace_probe(old_tp);
 858		free_trace_probe(old_tp);
 
 
 
 859	}
 860
 861	/* Register new event */
 862	ret = register_probe_event(tp);
 863	if (ret) {
 864		pr_warning("Failed to register probe event(%d)\n", ret);
 865		goto end;
 866	}
 867
 868	/* Register k*probe */
 869	ret = __register_trace_probe(tp);
 870	if (ret < 0)
 871		unregister_probe_event(tp);
 872	else
 873		list_add_tail(&tp->list, &probe_list);
 874
 875end:
 876	mutex_unlock(&probe_lock);
 877	return ret;
 878}
 879
 880/* Module notifier call back, checking event on the module */
 881static int trace_probe_module_callback(struct notifier_block *nb,
 882				       unsigned long val, void *data)
 883{
 884	struct module *mod = data;
 885	struct trace_probe *tp;
 886	int ret;
 887
 888	if (val != MODULE_STATE_COMING)
 889		return NOTIFY_DONE;
 890
 891	/* Update probes on coming module */
 892	mutex_lock(&probe_lock);
 893	list_for_each_entry(tp, &probe_list, list) {
 894		if (trace_probe_within_module(tp, mod)) {
 895			__unregister_trace_probe(tp);
 896			ret = __register_trace_probe(tp);
 
 897			if (ret)
 898				pr_warning("Failed to re-register probe %s on"
 899					   "%s: %d\n",
 900					   tp->call.name, mod->name, ret);
 901		}
 902	}
 903	mutex_unlock(&probe_lock);
 904
 905	return NOTIFY_DONE;
 906}
 907
 908static struct notifier_block trace_probe_module_nb = {
 909	.notifier_call = trace_probe_module_callback,
 910	.priority = 1	/* Invoked after kprobe module callback */
 911};
 912
 913/* Split symbol and offset. */
 914static int split_symbol_offset(char *symbol, unsigned long *offset)
 915{
 916	char *tmp;
 917	int ret;
 918
 919	if (!offset)
 920		return -EINVAL;
 921
 922	tmp = strchr(symbol, '+');
 923	if (tmp) {
 924		/* skip sign because strict_strtol doesn't accept '+' */
 925		ret = strict_strtoul(tmp + 1, 0, offset);
 926		if (ret)
 927			return ret;
 928		*tmp = '\0';
 929	} else
 930		*offset = 0;
 931	return 0;
 932}
 933
 934#define PARAM_MAX_ARGS 16
 935#define PARAM_MAX_STACK (THREAD_SIZE / sizeof(unsigned long))
 936
 937static int parse_probe_vars(char *arg, const struct fetch_type *t,
 938			    struct fetch_param *f, int is_return)
 939{
 940	int ret = 0;
 941	unsigned long param;
 942
 943	if (strcmp(arg, "retval") == 0) {
 944		if (is_return)
 945			f->fn = t->fetch[FETCH_MTD_retval];
 946		else
 947			ret = -EINVAL;
 948	} else if (strncmp(arg, "stack", 5) == 0) {
 949		if (arg[5] == '\0') {
 950			if (strcmp(t->name, DEFAULT_FETCH_TYPE_STR) == 0)
 951				f->fn = fetch_stack_address;
 952			else
 953				ret = -EINVAL;
 954		} else if (isdigit(arg[5])) {
 955			ret = strict_strtoul(arg + 5, 10, &param);
 956			if (ret || param > PARAM_MAX_STACK)
 957				ret = -EINVAL;
 958			else {
 959				f->fn = t->fetch[FETCH_MTD_stack];
 960				f->data = (void *)param;
 961			}
 962		} else
 963			ret = -EINVAL;
 964	} else
 965		ret = -EINVAL;
 966	return ret;
 967}
 968
 969/* Recursive argument parser */
 970static int __parse_probe_arg(char *arg, const struct fetch_type *t,
 971			     struct fetch_param *f, int is_return)
 972{
 973	int ret = 0;
 974	unsigned long param;
 975	long offset;
 976	char *tmp;
 977
 978	switch (arg[0]) {
 979	case '$':
 980		ret = parse_probe_vars(arg + 1, t, f, is_return);
 981		break;
 982	case '%':	/* named register */
 983		ret = regs_query_register_offset(arg + 1);
 984		if (ret >= 0) {
 985			f->fn = t->fetch[FETCH_MTD_reg];
 986			f->data = (void *)(unsigned long)ret;
 987			ret = 0;
 988		}
 989		break;
 990	case '@':	/* memory or symbol */
 991		if (isdigit(arg[1])) {
 992			ret = strict_strtoul(arg + 1, 0, &param);
 993			if (ret)
 994				break;
 995			f->fn = t->fetch[FETCH_MTD_memory];
 996			f->data = (void *)param;
 997		} else {
 998			ret = split_symbol_offset(arg + 1, &offset);
 999			if (ret)
1000				break;
1001			f->data = alloc_symbol_cache(arg + 1, offset);
1002			if (f->data)
1003				f->fn = t->fetch[FETCH_MTD_symbol];
1004		}
1005		break;
1006	case '+':	/* deref memory */
1007		arg++;	/* Skip '+', because strict_strtol() rejects it. */
1008	case '-':
1009		tmp = strchr(arg, '(');
1010		if (!tmp)
1011			break;
1012		*tmp = '\0';
1013		ret = strict_strtol(arg, 0, &offset);
1014		if (ret)
1015			break;
1016		arg = tmp + 1;
1017		tmp = strrchr(arg, ')');
1018		if (tmp) {
1019			struct deref_fetch_param *dprm;
1020			const struct fetch_type *t2 = find_fetch_type(NULL);
1021			*tmp = '\0';
1022			dprm = kzalloc(sizeof(struct deref_fetch_param),
1023				       GFP_KERNEL);
1024			if (!dprm)
1025				return -ENOMEM;
1026			dprm->offset = offset;
1027			ret = __parse_probe_arg(arg, t2, &dprm->orig,
1028						is_return);
1029			if (ret)
1030				kfree(dprm);
1031			else {
1032				f->fn = t->fetch[FETCH_MTD_deref];
1033				f->data = (void *)dprm;
1034			}
1035		}
1036		break;
1037	}
1038	if (!ret && !f->fn) {	/* Parsed, but do not find fetch method */
1039		pr_info("%s type has no corresponding fetch method.\n",
1040			t->name);
1041		ret = -EINVAL;
1042	}
1043	return ret;
1044}
1045
1046#define BYTES_TO_BITS(nb)	((BITS_PER_LONG * (nb)) / sizeof(long))
1047
1048/* Bitfield type needs to be parsed into a fetch function */
1049static int __parse_bitfield_probe_arg(const char *bf,
1050				      const struct fetch_type *t,
1051				      struct fetch_param *f)
1052{
1053	struct bitfield_fetch_param *bprm;
1054	unsigned long bw, bo;
1055	char *tail;
1056
1057	if (*bf != 'b')
1058		return 0;
1059
1060	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1061	if (!bprm)
1062		return -ENOMEM;
1063	bprm->orig = *f;
1064	f->fn = t->fetch[FETCH_MTD_bitfield];
1065	f->data = (void *)bprm;
1066
1067	bw = simple_strtoul(bf + 1, &tail, 0);	/* Use simple one */
1068	if (bw == 0 || *tail != '@')
1069		return -EINVAL;
1070
1071	bf = tail + 1;
1072	bo = simple_strtoul(bf, &tail, 0);
1073	if (tail == bf || *tail != '/')
1074		return -EINVAL;
1075
1076	bprm->hi_shift = BYTES_TO_BITS(t->size) - (bw + bo);
1077	bprm->low_shift = bprm->hi_shift + bo;
1078	return (BYTES_TO_BITS(t->size) < (bw + bo)) ? -EINVAL : 0;
1079}
1080
1081/* String length checking wrapper */
1082static int parse_probe_arg(char *arg, struct trace_probe *tp,
1083			   struct probe_arg *parg, int is_return)
1084{
1085	const char *t;
1086	int ret;
1087
1088	if (strlen(arg) > MAX_ARGSTR_LEN) {
1089		pr_info("Argument is too long.: %s\n",  arg);
1090		return -ENOSPC;
1091	}
1092	parg->comm = kstrdup(arg, GFP_KERNEL);
1093	if (!parg->comm) {
1094		pr_info("Failed to allocate memory for command '%s'.\n", arg);
1095		return -ENOMEM;
1096	}
1097	t = strchr(parg->comm, ':');
1098	if (t) {
1099		arg[t - parg->comm] = '\0';
1100		t++;
1101	}
1102	parg->type = find_fetch_type(t);
1103	if (!parg->type) {
1104		pr_info("Unsupported type: %s\n", t);
1105		return -EINVAL;
1106	}
1107	parg->offset = tp->size;
1108	tp->size += parg->type->size;
1109	ret = __parse_probe_arg(arg, parg->type, &parg->fetch, is_return);
1110	if (ret >= 0 && t != NULL)
1111		ret = __parse_bitfield_probe_arg(t, parg->type, &parg->fetch);
1112	if (ret >= 0) {
1113		parg->fetch_size.fn = get_fetch_size_function(parg->type,
1114							      parg->fetch.fn);
1115		parg->fetch_size.data = parg->fetch.data;
1116	}
1117	return ret;
1118}
1119
1120/* Return 1 if name is reserved or already used by another argument */
1121static int conflict_field_name(const char *name,
1122			       struct probe_arg *args, int narg)
1123{
1124	int i;
1125	for (i = 0; i < ARRAY_SIZE(reserved_field_names); i++)
1126		if (strcmp(reserved_field_names[i], name) == 0)
1127			return 1;
1128	for (i = 0; i < narg; i++)
1129		if (strcmp(args[i].name, name) == 0)
1130			return 1;
1131	return 0;
1132}
1133
1134static int create_trace_probe(int argc, char **argv)
1135{
1136	/*
1137	 * Argument syntax:
1138	 *  - Add kprobe: p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
1139	 *  - Add kretprobe: r[:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 
 
1140	 * Fetch args:
1141	 *  $retval	: fetch return value
1142	 *  $stack	: fetch stack address
1143	 *  $stackN	: fetch Nth of stack (N:0-)
 
1144	 *  @ADDR	: fetch memory at ADDR (ADDR should be in kernel)
1145	 *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
1146	 *  %REG	: fetch register REG
1147	 * Dereferencing memory fetch:
1148	 *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
1149	 * Alias name of args:
1150	 *  NAME=FETCHARG : set NAME as alias of FETCHARG.
1151	 * Type of args:
1152	 *  FETCHARG:TYPE : use TYPE instead of unsigned long.
1153	 */
1154	struct trace_probe *tp;
1155	int i, ret = 0;
1156	int is_return = 0, is_delete = 0;
1157	char *symbol = NULL, *event = NULL, *group = NULL;
 
1158	char *arg;
1159	unsigned long offset = 0;
1160	void *addr = NULL;
1161	char buf[MAX_EVENT_NAME_LEN];
1162
1163	/* argc must be >= 1 */
1164	if (argv[0][0] == 'p')
1165		is_return = 0;
1166	else if (argv[0][0] == 'r')
1167		is_return = 1;
1168	else if (argv[0][0] == '-')
1169		is_delete = 1;
1170	else {
1171		pr_info("Probe definition must be started with 'p', 'r' or"
1172			" '-'.\n");
1173		return -EINVAL;
1174	}
1175
1176	if (argv[0][1] == ':') {
1177		event = &argv[0][2];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1178		if (strchr(event, '/')) {
1179			group = event;
1180			event = strchr(group, '/') + 1;
1181			event[-1] = '\0';
1182			if (strlen(group) == 0) {
1183				pr_info("Group name is not specified\n");
1184				return -EINVAL;
1185			}
1186		}
1187		if (strlen(event) == 0) {
1188			pr_info("Event name is not specified\n");
1189			return -EINVAL;
1190		}
1191	}
1192	if (!group)
1193		group = KPROBE_EVENT_SYSTEM;
1194
1195	if (is_delete) {
1196		if (!event) {
1197			pr_info("Delete command needs an event name.\n");
1198			return -EINVAL;
1199		}
1200		mutex_lock(&probe_lock);
1201		tp = find_trace_probe(event, group);
1202		if (!tp) {
1203			mutex_unlock(&probe_lock);
1204			pr_info("Event %s/%s doesn't exist.\n", group, event);
1205			return -ENOENT;
1206		}
1207		/* delete an event */
1208		unregister_trace_probe(tp);
1209		free_trace_probe(tp);
 
1210		mutex_unlock(&probe_lock);
1211		return 0;
1212	}
1213
1214	if (argc < 2) {
1215		pr_info("Probe point is not specified.\n");
1216		return -EINVAL;
1217	}
1218	if (isdigit(argv[1][0])) {
1219		if (is_return) {
1220			pr_info("Return probe point must be a symbol.\n");
1221			return -EINVAL;
1222		}
1223		/* an address specified */
1224		ret = strict_strtoul(&argv[1][0], 0, (unsigned long *)&addr);
1225		if (ret) {
1226			pr_info("Failed to parse address.\n");
1227			return ret;
1228		}
1229	} else {
1230		/* a symbol specified */
1231		symbol = argv[1];
1232		/* TODO: support .init module functions */
1233		ret = split_symbol_offset(symbol, &offset);
1234		if (ret) {
1235			pr_info("Failed to parse symbol.\n");
1236			return ret;
1237		}
1238		if (offset && is_return) {
1239			pr_info("Return probe must be used without offset.\n");
 
1240			return -EINVAL;
1241		}
1242	}
1243	argc -= 2; argv += 2;
1244
1245	/* setup a probe */
1246	if (!event) {
1247		/* Make a new event name */
1248		if (symbol)
1249			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
1250				 is_return ? 'r' : 'p', symbol, offset);
1251		else
1252			snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
1253				 is_return ? 'r' : 'p', addr);
 
1254		event = buf;
1255	}
1256	tp = alloc_trace_probe(group, event, addr, symbol, offset, argc,
1257			       is_return);
1258	if (IS_ERR(tp)) {
1259		pr_info("Failed to allocate trace_probe.(%d)\n",
1260			(int)PTR_ERR(tp));
1261		return PTR_ERR(tp);
1262	}
1263
1264	/* parse arguments */
1265	ret = 0;
1266	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 
 
1267		/* Increment count for freeing args in error case */
1268		tp->nr_args++;
1269
1270		/* Parse argument name */
1271		arg = strchr(argv[i], '=');
1272		if (arg) {
1273			*arg++ = '\0';
1274			tp->args[i].name = kstrdup(argv[i], GFP_KERNEL);
1275		} else {
1276			arg = argv[i];
1277			/* If argument name is omitted, set "argN" */
1278			snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
1279			tp->args[i].name = kstrdup(buf, GFP_KERNEL);
1280		}
1281
1282		if (!tp->args[i].name) {
1283			pr_info("Failed to allocate argument[%d] name.\n", i);
1284			ret = -ENOMEM;
1285			goto error;
1286		}
1287
1288		if (!is_good_name(tp->args[i].name)) {
1289			pr_info("Invalid argument[%d] name: %s\n",
1290				i, tp->args[i].name);
1291			ret = -EINVAL;
1292			goto error;
1293		}
1294
1295		if (conflict_field_name(tp->args[i].name, tp->args, i)) {
 
1296			pr_info("Argument[%d] name '%s' conflicts with "
1297				"another field.\n", i, argv[i]);
1298			ret = -EINVAL;
1299			goto error;
1300		}
1301
1302		/* Parse fetch argument */
1303		ret = parse_probe_arg(arg, tp, &tp->args[i], is_return);
 
 
1304		if (ret) {
1305			pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
1306			goto error;
1307		}
1308	}
1309
1310	ret = register_trace_probe(tp);
1311	if (ret)
1312		goto error;
1313	return 0;
1314
1315error:
1316	free_trace_probe(tp);
1317	return ret;
1318}
1319
1320static void release_all_trace_probes(void)
1321{
1322	struct trace_probe *tp;
 
1323
1324	mutex_lock(&probe_lock);
 
 
 
 
 
 
1325	/* TODO: Use batch unregistration */
1326	while (!list_empty(&probe_list)) {
1327		tp = list_entry(probe_list.next, struct trace_probe, list);
1328		unregister_trace_probe(tp);
1329		free_trace_probe(tp);
 
 
1330	}
 
 
1331	mutex_unlock(&probe_lock);
 
 
1332}
1333
1334/* Probes listing interfaces */
1335static void *probes_seq_start(struct seq_file *m, loff_t *pos)
1336{
1337	mutex_lock(&probe_lock);
1338	return seq_list_start(&probe_list, *pos);
1339}
1340
1341static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
1342{
1343	return seq_list_next(v, &probe_list, pos);
1344}
1345
1346static void probes_seq_stop(struct seq_file *m, void *v)
1347{
1348	mutex_unlock(&probe_lock);
1349}
1350
1351static int probes_seq_show(struct seq_file *m, void *v)
1352{
1353	struct trace_probe *tp = v;
1354	int i;
1355
1356	seq_printf(m, "%c", trace_probe_is_return(tp) ? 'r' : 'p');
1357	seq_printf(m, ":%s/%s", tp->call.class->system, tp->call.name);
1358
1359	if (!tp->symbol)
1360		seq_printf(m, " 0x%p", tp->rp.kp.addr);
1361	else if (tp->rp.kp.offset)
1362		seq_printf(m, " %s+%u", trace_probe_symbol(tp),
1363			   tp->rp.kp.offset);
 
1364	else
1365		seq_printf(m, " %s", trace_probe_symbol(tp));
1366
1367	for (i = 0; i < tp->nr_args; i++)
1368		seq_printf(m, " %s=%s", tp->args[i].name, tp->args[i].comm);
1369	seq_printf(m, "\n");
1370
1371	return 0;
1372}
1373
1374static const struct seq_operations probes_seq_op = {
1375	.start  = probes_seq_start,
1376	.next   = probes_seq_next,
1377	.stop   = probes_seq_stop,
1378	.show   = probes_seq_show
1379};
1380
1381static int probes_open(struct inode *inode, struct file *file)
1382{
1383	if ((file->f_mode & FMODE_WRITE) &&
1384	    (file->f_flags & O_TRUNC))
1385		release_all_trace_probes();
 
 
 
 
1386
1387	return seq_open(file, &probes_seq_op);
1388}
1389
1390static int command_trace_probe(const char *buf)
1391{
1392	char **argv;
1393	int argc = 0, ret = 0;
1394
1395	argv = argv_split(GFP_KERNEL, buf, &argc);
1396	if (!argv)
1397		return -ENOMEM;
1398
1399	if (argc)
1400		ret = create_trace_probe(argc, argv);
1401
1402	argv_free(argv);
1403	return ret;
1404}
1405
1406#define WRITE_BUFSIZE 4096
1407
1408static ssize_t probes_write(struct file *file, const char __user *buffer,
1409			    size_t count, loff_t *ppos)
1410{
1411	char *kbuf, *tmp;
1412	int ret;
1413	size_t done;
1414	size_t size;
1415
1416	kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
1417	if (!kbuf)
1418		return -ENOMEM;
1419
1420	ret = done = 0;
1421	while (done < count) {
1422		size = count - done;
1423		if (size >= WRITE_BUFSIZE)
1424			size = WRITE_BUFSIZE - 1;
1425		if (copy_from_user(kbuf, buffer + done, size)) {
1426			ret = -EFAULT;
1427			goto out;
1428		}
1429		kbuf[size] = '\0';
1430		tmp = strchr(kbuf, '\n');
1431		if (tmp) {
1432			*tmp = '\0';
1433			size = tmp - kbuf + 1;
1434		} else if (done + size < count) {
1435			pr_warning("Line length is too long: "
1436				   "Should be less than %d.", WRITE_BUFSIZE);
1437			ret = -EINVAL;
1438			goto out;
1439		}
1440		done += size;
1441		/* Remove comments */
1442		tmp = strchr(kbuf, '#');
1443		if (tmp)
1444			*tmp = '\0';
1445
1446		ret = command_trace_probe(kbuf);
1447		if (ret)
1448			goto out;
1449	}
1450	ret = done;
1451out:
1452	kfree(kbuf);
1453	return ret;
1454}
1455
1456static const struct file_operations kprobe_events_ops = {
1457	.owner          = THIS_MODULE,
1458	.open           = probes_open,
1459	.read           = seq_read,
1460	.llseek         = seq_lseek,
1461	.release        = seq_release,
1462	.write		= probes_write,
1463};
1464
1465/* Probes profiling interfaces */
1466static int probes_profile_seq_show(struct seq_file *m, void *v)
1467{
1468	struct trace_probe *tp = v;
1469
1470	seq_printf(m, "  %-44s %15lu %15lu\n", tp->call.name, tp->nhit,
1471		   tp->rp.kp.nmissed);
 
 
1472
1473	return 0;
1474}
1475
1476static const struct seq_operations profile_seq_op = {
1477	.start  = probes_seq_start,
1478	.next   = probes_seq_next,
1479	.stop   = probes_seq_stop,
1480	.show   = probes_profile_seq_show
1481};
1482
1483static int profile_open(struct inode *inode, struct file *file)
1484{
1485	return seq_open(file, &profile_seq_op);
1486}
1487
1488static const struct file_operations kprobe_profile_ops = {
1489	.owner          = THIS_MODULE,
1490	.open           = profile_open,
1491	.read           = seq_read,
1492	.llseek         = seq_lseek,
1493	.release        = seq_release,
1494};
1495
1496/* Sum up total data length for dynamic arraies (strings) */
1497static __kprobes int __get_data_size(struct trace_probe *tp,
1498				     struct pt_regs *regs)
1499{
1500	int i, ret = 0;
1501	u32 len;
1502
1503	for (i = 0; i < tp->nr_args; i++)
1504		if (unlikely(tp->args[i].fetch_size.fn)) {
1505			call_fetch(&tp->args[i].fetch_size, regs, &len);
1506			ret += len;
1507		}
1508
1509	return ret;
1510}
1511
1512/* Store the value of each argument */
1513static __kprobes void store_trace_args(int ent_size, struct trace_probe *tp,
1514				       struct pt_regs *regs,
1515				       u8 *data, int maxlen)
1516{
1517	int i;
1518	u32 end = tp->size;
1519	u32 *dl;	/* Data (relative) location */
1520
1521	for (i = 0; i < tp->nr_args; i++) {
1522		if (unlikely(tp->args[i].fetch_size.fn)) {
1523			/*
1524			 * First, we set the relative location and
1525			 * maximum data length to *dl
1526			 */
1527			dl = (u32 *)(data + tp->args[i].offset);
1528			*dl = make_data_rloc(maxlen, end - tp->args[i].offset);
1529			/* Then try to fetch string or dynamic array data */
1530			call_fetch(&tp->args[i].fetch, regs, dl);
1531			/* Reduce maximum length */
1532			end += get_rloc_len(*dl);
1533			maxlen -= get_rloc_len(*dl);
1534			/* Trick here, convert data_rloc to data_loc */
1535			*dl = convert_rloc_to_loc(*dl,
1536				 ent_size + tp->args[i].offset);
1537		} else
1538			/* Just fetching data normally */
1539			call_fetch(&tp->args[i].fetch, regs,
1540				   data + tp->args[i].offset);
1541	}
1542}
1543
1544/* Kprobe handler */
1545static __kprobes void kprobe_trace_func(struct kprobe *kp, struct pt_regs *regs)
 
 
1546{
1547	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1548	struct kprobe_trace_entry_head *entry;
1549	struct ring_buffer_event *event;
1550	struct ring_buffer *buffer;
1551	int size, dsize, pc;
1552	unsigned long irq_flags;
1553	struct ftrace_event_call *call = &tp->call;
1554
1555	tp->nhit++;
 
 
 
1556
1557	local_save_flags(irq_flags);
1558	pc = preempt_count();
1559
1560	dsize = __get_data_size(tp, regs);
1561	size = sizeof(*entry) + tp->size + dsize;
1562
1563	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1564						  size, irq_flags, pc);
 
1565	if (!event)
1566		return;
1567
1568	entry = ring_buffer_event_data(event);
1569	entry->ip = (unsigned long)kp->addr;
1570	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 
 
 
 
1571
1572	if (!filter_current_check_discard(buffer, call, entry, event))
1573		trace_nowake_buffer_unlock_commit_regs(buffer, event,
1574						       irq_flags, pc, regs);
 
 
 
 
1575}
 
1576
1577/* Kretprobe handler */
1578static __kprobes void kretprobe_trace_func(struct kretprobe_instance *ri,
1579					  struct pt_regs *regs)
 
 
1580{
1581	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1582	struct kretprobe_trace_entry_head *entry;
1583	struct ring_buffer_event *event;
1584	struct ring_buffer *buffer;
1585	int size, pc, dsize;
1586	unsigned long irq_flags;
1587	struct ftrace_event_call *call = &tp->call;
 
 
 
 
 
1588
1589	local_save_flags(irq_flags);
1590	pc = preempt_count();
1591
1592	dsize = __get_data_size(tp, regs);
1593	size = sizeof(*entry) + tp->size + dsize;
1594
1595	event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
1596						  size, irq_flags, pc);
 
1597	if (!event)
1598		return;
1599
1600	entry = ring_buffer_event_data(event);
1601	entry->func = (unsigned long)tp->rp.kp.addr;
1602	entry->ret_ip = (unsigned long)ri->ret_addr;
1603	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
 
 
 
 
 
 
 
 
 
 
1604
1605	if (!filter_current_check_discard(buffer, call, entry, event))
1606		trace_nowake_buffer_unlock_commit_regs(buffer, event,
1607						       irq_flags, pc, regs);
1608}
 
1609
1610/* Event entry printers */
1611enum print_line_t
1612print_kprobe_event(struct trace_iterator *iter, int flags,
1613		   struct trace_event *event)
1614{
1615	struct kprobe_trace_entry_head *field;
1616	struct trace_seq *s = &iter->seq;
1617	struct trace_probe *tp;
1618	u8 *data;
1619	int i;
1620
1621	field = (struct kprobe_trace_entry_head *)iter->ent;
1622	tp = container_of(event, struct trace_probe, call.event);
1623
1624	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1625		goto partial;
1626
1627	if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1628		goto partial;
1629
1630	if (!trace_seq_puts(s, ")"))
1631		goto partial;
1632
1633	data = (u8 *)&field[1];
1634	for (i = 0; i < tp->nr_args; i++)
1635		if (!tp->args[i].type->print(s, tp->args[i].name,
1636					     data + tp->args[i].offset, field))
1637			goto partial;
1638
1639	if (!trace_seq_puts(s, "\n"))
1640		goto partial;
1641
1642	return TRACE_TYPE_HANDLED;
1643partial:
1644	return TRACE_TYPE_PARTIAL_LINE;
1645}
1646
1647enum print_line_t
1648print_kretprobe_event(struct trace_iterator *iter, int flags,
1649		      struct trace_event *event)
1650{
1651	struct kretprobe_trace_entry_head *field;
1652	struct trace_seq *s = &iter->seq;
1653	struct trace_probe *tp;
1654	u8 *data;
1655	int i;
1656
1657	field = (struct kretprobe_trace_entry_head *)iter->ent;
1658	tp = container_of(event, struct trace_probe, call.event);
1659
1660	if (!trace_seq_printf(s, "%s: (", tp->call.name))
1661		goto partial;
1662
1663	if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1664		goto partial;
1665
1666	if (!trace_seq_puts(s, " <- "))
1667		goto partial;
1668
1669	if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1670		goto partial;
1671
1672	if (!trace_seq_puts(s, ")"))
1673		goto partial;
1674
1675	data = (u8 *)&field[1];
1676	for (i = 0; i < tp->nr_args; i++)
1677		if (!tp->args[i].type->print(s, tp->args[i].name,
1678					     data + tp->args[i].offset, field))
1679			goto partial;
1680
1681	if (!trace_seq_puts(s, "\n"))
1682		goto partial;
 
 
 
1683
1684	return TRACE_TYPE_HANDLED;
1685partial:
1686	return TRACE_TYPE_PARTIAL_LINE;
1687}
1688
1689#undef DEFINE_FIELD
1690#define DEFINE_FIELD(type, item, name, is_signed)			\
1691	do {								\
1692		ret = trace_define_field(event_call, #type, name,	\
1693					 offsetof(typeof(field), item),	\
1694					 sizeof(field.item), is_signed, \
1695					 FILTER_OTHER);			\
1696		if (ret)						\
1697			return ret;					\
1698	} while (0)
1699
1700static int kprobe_event_define_fields(struct ftrace_event_call *event_call)
1701{
1702	int ret, i;
1703	struct kprobe_trace_entry_head field;
1704	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1705
1706	DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1707	/* Set argument names as fields */
1708	for (i = 0; i < tp->nr_args; i++) {
1709		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1710					 tp->args[i].name,
1711					 sizeof(field) + tp->args[i].offset,
1712					 tp->args[i].type->size,
1713					 tp->args[i].type->is_signed,
 
 
1714					 FILTER_OTHER);
1715		if (ret)
1716			return ret;
1717	}
1718	return 0;
1719}
1720
1721static int kretprobe_event_define_fields(struct ftrace_event_call *event_call)
1722{
1723	int ret, i;
1724	struct kretprobe_trace_entry_head field;
1725	struct trace_probe *tp = (struct trace_probe *)event_call->data;
1726
1727	DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1728	DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1729	/* Set argument names as fields */
1730	for (i = 0; i < tp->nr_args; i++) {
1731		ret = trace_define_field(event_call, tp->args[i].type->fmttype,
1732					 tp->args[i].name,
1733					 sizeof(field) + tp->args[i].offset,
1734					 tp->args[i].type->size,
1735					 tp->args[i].type->is_signed,
 
 
1736					 FILTER_OTHER);
1737		if (ret)
1738			return ret;
1739	}
1740	return 0;
1741}
1742
1743static int __set_print_fmt(struct trace_probe *tp, char *buf, int len)
1744{
1745	int i;
1746	int pos = 0;
1747
1748	const char *fmt, *arg;
1749
1750	if (!trace_probe_is_return(tp)) {
1751		fmt = "(%lx)";
1752		arg = "REC->" FIELD_STRING_IP;
1753	} else {
1754		fmt = "(%lx <- %lx)";
1755		arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
1756	}
1757
1758	/* When len=0, we just calculate the needed length */
1759#define LEN_OR_ZERO (len ? len - pos : 0)
1760
1761	pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
1762
1763	for (i = 0; i < tp->nr_args; i++) {
1764		pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
1765				tp->args[i].name, tp->args[i].type->fmt);
1766	}
1767
1768	pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
1769
1770	for (i = 0; i < tp->nr_args; i++) {
1771		if (strcmp(tp->args[i].type->name, "string") == 0)
1772			pos += snprintf(buf + pos, LEN_OR_ZERO,
1773					", __get_str(%s)",
1774					tp->args[i].name);
1775		else
1776			pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
1777					tp->args[i].name);
1778	}
1779
1780#undef LEN_OR_ZERO
1781
1782	/* return the length of print_fmt */
1783	return pos;
1784}
1785
1786static int set_print_fmt(struct trace_probe *tp)
1787{
1788	int len;
1789	char *print_fmt;
1790
1791	/* First: called with 0 length to calculate the needed length */
1792	len = __set_print_fmt(tp, NULL, 0);
1793	print_fmt = kmalloc(len + 1, GFP_KERNEL);
1794	if (!print_fmt)
1795		return -ENOMEM;
1796
1797	/* Second: actually write the @print_fmt */
1798	__set_print_fmt(tp, print_fmt, len + 1);
1799	tp->call.print_fmt = print_fmt;
1800
1801	return 0;
1802}
1803
1804#ifdef CONFIG_PERF_EVENTS
1805
1806/* Kprobe profile handler */
1807static __kprobes void kprobe_perf_func(struct kprobe *kp,
1808					 struct pt_regs *regs)
1809{
1810	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
1811	struct ftrace_event_call *call = &tp->call;
1812	struct kprobe_trace_entry_head *entry;
1813	struct hlist_head *head;
1814	int size, __size, dsize;
1815	int rctx;
1816
1817	dsize = __get_data_size(tp, regs);
1818	__size = sizeof(*entry) + tp->size + dsize;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1819	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1820	size -= sizeof(u32);
1821	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1822		     "profile buffer not large enough"))
1823		return;
1824
1825	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1826	if (!entry)
1827		return;
1828
1829	entry->ip = (unsigned long)kp->addr;
1830	memset(&entry[1], 0, dsize);
1831	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1832
1833	head = this_cpu_ptr(call->perf_events);
1834	perf_trace_buf_submit(entry, size, rctx, entry->ip, 1, regs, head);
1835}
 
1836
1837/* Kretprobe profile handler */
1838static __kprobes void kretprobe_perf_func(struct kretprobe_instance *ri,
1839					    struct pt_regs *regs)
 
1840{
1841	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1842	struct ftrace_event_call *call = &tp->call;
1843	struct kretprobe_trace_entry_head *entry;
1844	struct hlist_head *head;
1845	int size, __size, dsize;
1846	int rctx;
1847
1848	dsize = __get_data_size(tp, regs);
1849	__size = sizeof(*entry) + tp->size + dsize;
 
 
 
 
 
 
 
1850	size = ALIGN(__size + sizeof(u32), sizeof(u64));
1851	size -= sizeof(u32);
1852	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
1853		     "profile buffer not large enough"))
1854		return;
1855
1856	entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1857	if (!entry)
1858		return;
1859
1860	entry->func = (unsigned long)tp->rp.kp.addr;
1861	entry->ret_ip = (unsigned long)ri->ret_addr;
1862	store_trace_args(sizeof(*entry), tp, regs, (u8 *)&entry[1], dsize);
1863
1864	head = this_cpu_ptr(call->perf_events);
1865	perf_trace_buf_submit(entry, size, rctx, entry->ret_ip, 1, regs, head);
1866}
 
1867#endif	/* CONFIG_PERF_EVENTS */
1868
1869static __kprobes
1870int kprobe_register(struct ftrace_event_call *event, enum trace_reg type)
 
 
 
 
 
 
1871{
1872	struct trace_probe *tp = (struct trace_probe *)event->data;
 
1873
1874	switch (type) {
1875	case TRACE_REG_REGISTER:
1876		return enable_trace_probe(tp, TP_FLAG_TRACE);
1877	case TRACE_REG_UNREGISTER:
1878		disable_trace_probe(tp, TP_FLAG_TRACE);
1879		return 0;
1880
1881#ifdef CONFIG_PERF_EVENTS
1882	case TRACE_REG_PERF_REGISTER:
1883		return enable_trace_probe(tp, TP_FLAG_PROFILE);
1884	case TRACE_REG_PERF_UNREGISTER:
1885		disable_trace_probe(tp, TP_FLAG_PROFILE);
 
 
 
 
1886		return 0;
1887#endif
1888	}
1889	return 0;
1890}
1891
1892static __kprobes
1893int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1894{
1895	struct trace_probe *tp = container_of(kp, struct trace_probe, rp.kp);
 
 
 
1896
1897	if (tp->flags & TP_FLAG_TRACE)
1898		kprobe_trace_func(kp, regs);
1899#ifdef CONFIG_PERF_EVENTS
1900	if (tp->flags & TP_FLAG_PROFILE)
1901		kprobe_perf_func(kp, regs);
1902#endif
1903	return 0;	/* We don't tweek kernel, so just return 0 */
1904}
 
1905
1906static __kprobes
1907int kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1908{
1909	struct trace_probe *tp = container_of(ri->rp, struct trace_probe, rp);
1910
1911	if (tp->flags & TP_FLAG_TRACE)
1912		kretprobe_trace_func(ri, regs);
 
 
1913#ifdef CONFIG_PERF_EVENTS
1914	if (tp->flags & TP_FLAG_PROFILE)
1915		kretprobe_perf_func(ri, regs);
1916#endif
1917	return 0;	/* We don't tweek kernel, so just return 0 */
1918}
 
1919
1920static struct trace_event_functions kretprobe_funcs = {
1921	.trace		= print_kretprobe_event
1922};
1923
1924static struct trace_event_functions kprobe_funcs = {
1925	.trace		= print_kprobe_event
1926};
1927
1928static int register_probe_event(struct trace_probe *tp)
 
1929{
1930	struct ftrace_event_call *call = &tp->call;
1931	int ret;
1932
1933	/* Initialize ftrace_event_call */
1934	INIT_LIST_HEAD(&call->class->fields);
1935	if (trace_probe_is_return(tp)) {
1936		call->event.funcs = &kretprobe_funcs;
1937		call->class->define_fields = kretprobe_event_define_fields;
1938	} else {
1939		call->event.funcs = &kprobe_funcs;
1940		call->class->define_fields = kprobe_event_define_fields;
1941	}
1942	if (set_print_fmt(tp) < 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
1943		return -ENOMEM;
1944	ret = register_ftrace_event(&call->event);
1945	if (!ret) {
1946		kfree(call->print_fmt);
1947		return -ENODEV;
1948	}
1949	call->flags = 0;
1950	call->class->reg = kprobe_register;
1951	call->data = tp;
1952	ret = trace_add_event_call(call);
1953	if (ret) {
1954		pr_info("Failed to register kprobe event: %s\n", call->name);
 
1955		kfree(call->print_fmt);
1956		unregister_ftrace_event(&call->event);
1957	}
1958	return ret;
1959}
1960
1961static void unregister_probe_event(struct trace_probe *tp)
1962{
 
 
1963	/* tp->event is unregistered in trace_remove_event_call() */
1964	trace_remove_event_call(&tp->call);
1965	kfree(tp->call.print_fmt);
 
 
1966}
1967
1968/* Make a debugfs interface for controlling probe points */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1969static __init int init_kprobe_trace(void)
1970{
1971	struct dentry *d_tracer;
1972	struct dentry *entry;
1973
1974	if (register_module_notifier(&trace_probe_module_nb))
1975		return -EINVAL;
1976
1977	d_tracer = tracing_init_dentry();
1978	if (!d_tracer)
1979		return 0;
1980
1981	entry = debugfs_create_file("kprobe_events", 0644, d_tracer,
1982				    NULL, &kprobe_events_ops);
1983
1984	/* Event list interface */
1985	if (!entry)
1986		pr_warning("Could not create debugfs "
1987			   "'kprobe_events' entry\n");
1988
1989	/* Profile interface */
1990	entry = debugfs_create_file("kprobe_profile", 0444, d_tracer,
1991				    NULL, &kprobe_profile_ops);
1992
1993	if (!entry)
1994		pr_warning("Could not create debugfs "
1995			   "'kprobe_profile' entry\n");
1996	return 0;
1997}
1998fs_initcall(init_kprobe_trace);
1999
2000
2001#ifdef CONFIG_FTRACE_STARTUP_TEST
2002
2003/*
2004 * The "__used" keeps gcc from removing the function symbol
2005 * from the kallsyms table.
 
2006 */
2007static __used int kprobe_trace_selftest_target(int a1, int a2, int a3,
2008					       int a4, int a5, int a6)
2009{
2010	return a1 + a2 + a3 + a4 + a5 + a6;
2011}
2012
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2013static __init int kprobe_trace_self_tests_init(void)
2014{
2015	int ret, warn = 0;
2016	int (*target)(int, int, int, int, int, int);
2017	struct trace_probe *tp;
 
 
 
 
2018
2019	target = kprobe_trace_selftest_target;
2020
2021	pr_info("Testing kprobe tracing: ");
2022
2023	ret = command_trace_probe("p:testprobe kprobe_trace_selftest_target "
2024				  "$stack $stack0 +0($stack)");
 
2025	if (WARN_ON_ONCE(ret)) {
2026		pr_warning("error on probing function entry.\n");
2027		warn++;
2028	} else {
2029		/* Enable trace point */
2030		tp = find_trace_probe("testprobe", KPROBE_EVENT_SYSTEM);
2031		if (WARN_ON_ONCE(tp == NULL)) {
2032			pr_warning("error on getting new probe.\n");
2033			warn++;
2034		} else
2035			enable_trace_probe(tp, TP_FLAG_TRACE);
 
 
 
 
 
 
2036	}
2037
2038	ret = command_trace_probe("r:testprobe2 kprobe_trace_selftest_target "
2039				  "$retval");
2040	if (WARN_ON_ONCE(ret)) {
2041		pr_warning("error on probing function return.\n");
2042		warn++;
2043	} else {
2044		/* Enable trace point */
2045		tp = find_trace_probe("testprobe2", KPROBE_EVENT_SYSTEM);
2046		if (WARN_ON_ONCE(tp == NULL)) {
2047			pr_warning("error on getting new probe.\n");
2048			warn++;
2049		} else
2050			enable_trace_probe(tp, TP_FLAG_TRACE);
 
 
 
 
 
 
2051	}
2052
2053	if (warn)
2054		goto end;
2055
2056	ret = target(1, 2, 3, 4, 5, 6);
2057
2058	ret = command_trace_probe("-:testprobe");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2059	if (WARN_ON_ONCE(ret)) {
2060		pr_warning("error on deleting a probe.\n");
2061		warn++;
2062	}
2063
2064	ret = command_trace_probe("-:testprobe2");
2065	if (WARN_ON_ONCE(ret)) {
2066		pr_warning("error on deleting a probe.\n");
2067		warn++;
2068	}
2069
2070end:
2071	release_all_trace_probes();
 
 
 
 
 
2072	if (warn)
2073		pr_cont("NG: Some tests are failed. Please check them.\n");
2074	else
2075		pr_cont("OK\n");
2076	return 0;
2077}
2078
2079late_initcall(kprobe_trace_self_tests_init);
2080
2081#endif