Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * uprobes-based tracing events
   4 *
   5 * Copyright (C) IBM Corporation, 2010-2012
   6 * Author:	Srikar Dronamraju <srikar@linux.vnet.ibm.com>
   7 */
   8#define pr_fmt(fmt)	"trace_uprobe: " fmt
   9
  10#include <linux/bpf-cgroup.h>
  11#include <linux/security.h>
  12#include <linux/ctype.h>
  13#include <linux/module.h>
  14#include <linux/uaccess.h>
  15#include <linux/uprobes.h>
  16#include <linux/namei.h>
  17#include <linux/string.h>
  18#include <linux/rculist.h>
  19#include <linux/filter.h>
  20
  21#include "trace_dynevent.h"
  22#include "trace_probe.h"
  23#include "trace_probe_tmpl.h"
  24
  25#define UPROBE_EVENT_SYSTEM	"uprobes"
  26
  27struct uprobe_trace_entry_head {
  28	struct trace_entry	ent;
  29	unsigned long		vaddr[];
  30};
  31
  32#define SIZEOF_TRACE_ENTRY(is_return)			\
  33	(sizeof(struct uprobe_trace_entry_head) +	\
  34	 sizeof(unsigned long) * (is_return ? 2 : 1))
  35
  36#define DATAOF_TRACE_ENTRY(entry, is_return)		\
  37	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  38
  39static int trace_uprobe_create(const char *raw_command);
  40static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  41static int trace_uprobe_release(struct dyn_event *ev);
  42static bool trace_uprobe_is_busy(struct dyn_event *ev);
  43static bool trace_uprobe_match(const char *system, const char *event,
  44			int argc, const char **argv, struct dyn_event *ev);
  45
  46static struct dyn_event_operations trace_uprobe_ops = {
  47	.create = trace_uprobe_create,
  48	.show = trace_uprobe_show,
  49	.is_busy = trace_uprobe_is_busy,
  50	.free = trace_uprobe_release,
  51	.match = trace_uprobe_match,
  52};
  53
  54/*
  55 * uprobe event core functions
  56 */
  57struct trace_uprobe {
  58	struct dyn_event		devent;
  59	struct uprobe_consumer		consumer;
  60	struct path			path;
  61	struct inode			*inode;
  62	char				*filename;
  63	unsigned long			offset;
  64	unsigned long			ref_ctr_offset;
  65	unsigned long			nhit;
  66	struct trace_probe		tp;
  67};
  68
  69static bool is_trace_uprobe(struct dyn_event *ev)
  70{
  71	return ev->ops == &trace_uprobe_ops;
  72}
  73
  74static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  75{
  76	return container_of(ev, struct trace_uprobe, devent);
  77}
  78
  79/**
  80 * for_each_trace_uprobe - iterate over the trace_uprobe list
  81 * @pos:	the struct trace_uprobe * for each entry
  82 * @dpos:	the struct dyn_event * to use as a loop cursor
  83 */
  84#define for_each_trace_uprobe(pos, dpos)	\
  85	for_each_dyn_event(dpos)		\
  86		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  87
  88static int register_uprobe_event(struct trace_uprobe *tu);
  89static int unregister_uprobe_event(struct trace_uprobe *tu);
  90
 
 
 
 
 
  91static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  92static int uretprobe_dispatcher(struct uprobe_consumer *con,
  93				unsigned long func, struct pt_regs *regs);
  94
  95#ifdef CONFIG_STACK_GROWSUP
  96static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  97{
  98	return addr - (n * sizeof(long));
  99}
 100#else
 101static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 102{
 103	return addr + (n * sizeof(long));
 104}
 105#endif
 106
 107static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 108{
 109	unsigned long ret;
 110	unsigned long addr = user_stack_pointer(regs);
 111
 112	addr = adjust_stack_addr(addr, n);
 113
 114	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 115		return 0;
 116
 117	return ret;
 118}
 119
 120/*
 121 * Uprobes-specific fetch functions
 122 */
 123static nokprobe_inline int
 124probe_mem_read(void *dest, void *src, size_t size)
 125{
 126	void __user *vaddr = (void __force __user *)src;
 127
 128	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 129}
 130
 131static nokprobe_inline int
 132probe_mem_read_user(void *dest, void *src, size_t size)
 133{
 134	return probe_mem_read(dest, src, size);
 135}
 136
 137/*
 138 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 139 * length and relative data location.
 140 */
 141static nokprobe_inline int
 142fetch_store_string(unsigned long addr, void *dest, void *base)
 143{
 144	long ret;
 145	u32 loc = *(u32 *)dest;
 146	int maxlen  = get_loc_len(loc);
 147	u8 *dst = get_loc_data(dest, base);
 148	void __user *src = (void __force __user *) addr;
 149
 150	if (unlikely(!maxlen))
 151		return -ENOMEM;
 152
 153	if (addr == FETCH_TOKEN_COMM)
 154		ret = strscpy(dst, current->comm, maxlen);
 155	else
 156		ret = strncpy_from_user(dst, src, maxlen);
 157	if (ret >= 0) {
 158		if (ret == maxlen)
 159			dst[ret - 1] = '\0';
 160		else
 161			/*
 162			 * Include the terminating null byte. In this case it
 163			 * was copied by strncpy_from_user but not accounted
 164			 * for in ret.
 165			 */
 166			ret++;
 167		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 168	} else
 169		*(u32 *)dest = make_data_loc(0, (void *)dst - base);
 170
 171	return ret;
 172}
 173
 174static nokprobe_inline int
 175fetch_store_string_user(unsigned long addr, void *dest, void *base)
 176{
 177	return fetch_store_string(addr, dest, base);
 178}
 179
 180/* Return the length of string -- including null terminal byte */
 181static nokprobe_inline int
 182fetch_store_strlen(unsigned long addr)
 183{
 184	int len;
 185	void __user *vaddr = (void __force __user *) addr;
 186
 187	if (addr == FETCH_TOKEN_COMM)
 188		len = strlen(current->comm) + 1;
 189	else
 190		len = strnlen_user(vaddr, MAX_STRING_SIZE);
 191
 192	return (len > MAX_STRING_SIZE) ? 0 : len;
 193}
 194
 195static nokprobe_inline int
 196fetch_store_strlen_user(unsigned long addr)
 197{
 198	return fetch_store_strlen(addr);
 199}
 200
 201static unsigned long translate_user_vaddr(unsigned long file_offset)
 202{
 203	unsigned long base_addr;
 204	struct uprobe_dispatch_data *udd;
 205
 206	udd = (void *) current->utask->vaddr;
 207
 208	base_addr = udd->bp_addr - udd->tu->offset;
 209	return base_addr + file_offset;
 210}
 211
 212/* Note that we don't verify it, since the code does not come from user space */
 213static int
 214process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 215		   void *base)
 216{
 217	struct pt_regs *regs = rec;
 218	unsigned long val;
 219	int ret;
 220
 221	/* 1st stage: get value from context */
 222	switch (code->op) {
 223	case FETCH_OP_REG:
 224		val = regs_get_register(regs, code->param);
 225		break;
 226	case FETCH_OP_STACK:
 227		val = get_user_stack_nth(regs, code->param);
 228		break;
 229	case FETCH_OP_STACKP:
 230		val = user_stack_pointer(regs);
 231		break;
 232	case FETCH_OP_RETVAL:
 233		val = regs_return_value(regs);
 234		break;
 
 
 
 235	case FETCH_OP_COMM:
 236		val = FETCH_TOKEN_COMM;
 237		break;
 
 
 
 238	case FETCH_OP_FOFFS:
 239		val = translate_user_vaddr(code->immediate);
 240		break;
 241	default:
 242		ret = process_common_fetch_insn(code, &val);
 243		if (ret < 0)
 244			return ret;
 245	}
 246	code++;
 247
 248	return process_fetch_insn_bottom(code, val, dest, base);
 249}
 250NOKPROBE_SYMBOL(process_fetch_insn)
 251
 252static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 253{
 254	rwlock_init(&filter->rwlock);
 255	filter->nr_systemwide = 0;
 256	INIT_LIST_HEAD(&filter->perf_events);
 257}
 258
 259static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 260{
 261	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 262}
 263
 264static inline bool is_ret_probe(struct trace_uprobe *tu)
 265{
 266	return tu->consumer.ret_handler != NULL;
 267}
 268
 269static bool trace_uprobe_is_busy(struct dyn_event *ev)
 270{
 271	struct trace_uprobe *tu = to_trace_uprobe(ev);
 272
 273	return trace_probe_is_enabled(&tu->tp);
 274}
 275
 276static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 277					    int argc, const char **argv)
 278{
 279	char buf[MAX_ARGSTR_LEN + 1];
 280	int len;
 281
 282	if (!argc)
 283		return true;
 284
 285	len = strlen(tu->filename);
 286	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 287		return false;
 288
 289	if (tu->ref_ctr_offset == 0)
 290		snprintf(buf, sizeof(buf), "0x%0*lx",
 291				(int)(sizeof(void *) * 2), tu->offset);
 292	else
 293		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 294				(int)(sizeof(void *) * 2), tu->offset,
 295				tu->ref_ctr_offset);
 296	if (strcmp(buf, &argv[0][len + 1]))
 297		return false;
 298
 299	argc--; argv++;
 300
 301	return trace_probe_match_command_args(&tu->tp, argc, argv);
 302}
 303
 304static bool trace_uprobe_match(const char *system, const char *event,
 305			int argc, const char **argv, struct dyn_event *ev)
 306{
 307	struct trace_uprobe *tu = to_trace_uprobe(ev);
 308
 309	return (event[0] == '\0' ||
 310		strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 311	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 312	   trace_uprobe_match_command_head(tu, argc, argv);
 313}
 314
 315static nokprobe_inline struct trace_uprobe *
 316trace_uprobe_primary_from_call(struct trace_event_call *call)
 317{
 318	struct trace_probe *tp;
 319
 320	tp = trace_probe_primary_from_call(call);
 321	if (WARN_ON_ONCE(!tp))
 322		return NULL;
 323
 324	return container_of(tp, struct trace_uprobe, tp);
 325}
 326
 327/*
 328 * Allocate new trace_uprobe and initialize it (including uprobes).
 329 */
 330static struct trace_uprobe *
 331alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 332{
 333	struct trace_uprobe *tu;
 334	int ret;
 335
 336	tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 337	if (!tu)
 338		return ERR_PTR(-ENOMEM);
 339
 340	ret = trace_probe_init(&tu->tp, event, group, true);
 341	if (ret < 0)
 342		goto error;
 343
 344	dyn_event_init(&tu->devent, &trace_uprobe_ops);
 345	tu->consumer.handler = uprobe_dispatcher;
 346	if (is_ret)
 347		tu->consumer.ret_handler = uretprobe_dispatcher;
 348	init_trace_uprobe_filter(tu->tp.event->filter);
 349	return tu;
 350
 351error:
 352	kfree(tu);
 353
 354	return ERR_PTR(ret);
 355}
 356
 357static void free_trace_uprobe(struct trace_uprobe *tu)
 358{
 359	if (!tu)
 360		return;
 361
 362	path_put(&tu->path);
 363	trace_probe_cleanup(&tu->tp);
 364	kfree(tu->filename);
 365	kfree(tu);
 366}
 367
 368static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 369{
 370	struct dyn_event *pos;
 371	struct trace_uprobe *tu;
 372
 373	for_each_trace_uprobe(tu, pos)
 374		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 375		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 376			return tu;
 377
 378	return NULL;
 379}
 380
 381/* Unregister a trace_uprobe and probe_event */
 382static int unregister_trace_uprobe(struct trace_uprobe *tu)
 383{
 384	int ret;
 385
 386	if (trace_probe_has_sibling(&tu->tp))
 387		goto unreg;
 388
 389	/* If there's a reference to the dynamic event */
 390	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 391		return -EBUSY;
 392
 393	ret = unregister_uprobe_event(tu);
 394	if (ret)
 395		return ret;
 396
 397unreg:
 398	dyn_event_remove(&tu->devent);
 399	trace_probe_unlink(&tu->tp);
 400	free_trace_uprobe(tu);
 401	return 0;
 402}
 403
 404static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 405					 struct trace_uprobe *comp)
 406{
 407	struct trace_probe_event *tpe = orig->tp.event;
 408	struct inode *comp_inode = d_real_inode(comp->path.dentry);
 409	int i;
 410
 411	list_for_each_entry(orig, &tpe->probes, tp.list) {
 412		if (comp_inode != d_real_inode(orig->path.dentry) ||
 413		    comp->offset != orig->offset)
 414			continue;
 415
 416		/*
 417		 * trace_probe_compare_arg_type() ensured that nr_args and
 418		 * each argument name and type are same. Let's compare comm.
 419		 */
 420		for (i = 0; i < orig->tp.nr_args; i++) {
 421			if (strcmp(orig->tp.args[i].comm,
 422				   comp->tp.args[i].comm))
 423				break;
 424		}
 425
 426		if (i == orig->tp.nr_args)
 427			return true;
 428	}
 429
 430	return false;
 431}
 432
 433static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 434{
 435	int ret;
 436
 437	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 438	if (ret) {
 439		/* Note that argument starts index = 2 */
 440		trace_probe_log_set_index(ret + 1);
 441		trace_probe_log_err(0, DIFF_ARG_TYPE);
 442		return -EEXIST;
 443	}
 444	if (trace_uprobe_has_same_uprobe(to, tu)) {
 445		trace_probe_log_set_index(0);
 446		trace_probe_log_err(0, SAME_PROBE);
 447		return -EEXIST;
 448	}
 449
 450	/* Append to existing event */
 451	ret = trace_probe_append(&tu->tp, &to->tp);
 452	if (!ret)
 453		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 454
 455	return ret;
 456}
 457
 458/*
 459 * Uprobe with multiple reference counter is not allowed. i.e.
 460 * If inode and offset matches, reference counter offset *must*
 461 * match as well. Though, there is one exception: If user is
 462 * replacing old trace_uprobe with new one(same group/event),
 463 * then we allow same uprobe with new reference counter as far
 464 * as the new one does not conflict with any other existing
 465 * ones.
 466 */
 467static int validate_ref_ctr_offset(struct trace_uprobe *new)
 468{
 469	struct dyn_event *pos;
 470	struct trace_uprobe *tmp;
 471	struct inode *new_inode = d_real_inode(new->path.dentry);
 472
 473	for_each_trace_uprobe(tmp, pos) {
 474		if (new_inode == d_real_inode(tmp->path.dentry) &&
 475		    new->offset == tmp->offset &&
 476		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
 477			pr_warn("Reference counter offset mismatch.");
 478			return -EINVAL;
 479		}
 480	}
 481	return 0;
 482}
 483
 484/* Register a trace_uprobe and probe_event */
 485static int register_trace_uprobe(struct trace_uprobe *tu)
 486{
 487	struct trace_uprobe *old_tu;
 488	int ret;
 489
 490	mutex_lock(&event_mutex);
 491
 492	ret = validate_ref_ctr_offset(tu);
 493	if (ret)
 494		goto end;
 495
 496	/* register as an event */
 497	old_tu = find_probe_event(trace_probe_name(&tu->tp),
 498				  trace_probe_group_name(&tu->tp));
 499	if (old_tu) {
 500		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 501			trace_probe_log_set_index(0);
 502			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 503			ret = -EEXIST;
 504		} else {
 505			ret = append_trace_uprobe(tu, old_tu);
 506		}
 507		goto end;
 508	}
 509
 510	ret = register_uprobe_event(tu);
 511	if (ret) {
 512		if (ret == -EEXIST) {
 513			trace_probe_log_set_index(0);
 514			trace_probe_log_err(0, EVENT_EXIST);
 515		} else
 516			pr_warn("Failed to register probe event(%d)\n", ret);
 517		goto end;
 518	}
 519
 520	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 521
 522end:
 523	mutex_unlock(&event_mutex);
 524
 525	return ret;
 526}
 527
 528/*
 529 * Argument syntax:
 530 *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 531 */
 532static int __trace_uprobe_create(int argc, const char **argv)
 533{
 534	struct trace_uprobe *tu;
 535	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 536	char *arg, *filename, *rctr, *rctr_end, *tmp;
 537	char buf[MAX_EVENT_NAME_LEN];
 538	char gbuf[MAX_EVENT_NAME_LEN];
 539	enum probe_print_type ptype;
 540	struct path path;
 541	unsigned long offset, ref_ctr_offset;
 542	bool is_return = false;
 543	int i, ret;
 544
 545	ref_ctr_offset = 0;
 546
 547	switch (argv[0][0]) {
 548	case 'r':
 549		is_return = true;
 550		break;
 551	case 'p':
 552		break;
 553	default:
 554		return -ECANCELED;
 555	}
 556
 557	if (argc < 2)
 558		return -ECANCELED;
 559
 560	if (argv[0][1] == ':')
 561		event = &argv[0][2];
 562
 563	if (!strchr(argv[1], '/'))
 564		return -ECANCELED;
 565
 566	filename = kstrdup(argv[1], GFP_KERNEL);
 567	if (!filename)
 568		return -ENOMEM;
 569
 570	/* Find the last occurrence, in case the path contains ':' too. */
 571	arg = strrchr(filename, ':');
 572	if (!arg || !isdigit(arg[1])) {
 573		kfree(filename);
 574		return -ECANCELED;
 575	}
 576
 577	trace_probe_log_init("trace_uprobe", argc, argv);
 578	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
 579
 580	*arg++ = '\0';
 581	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 582	if (ret) {
 583		trace_probe_log_err(0, FILE_NOT_FOUND);
 584		kfree(filename);
 585		trace_probe_log_clear();
 586		return ret;
 587	}
 588	if (!d_is_reg(path.dentry)) {
 589		trace_probe_log_err(0, NO_REGULAR_FILE);
 590		ret = -EINVAL;
 591		goto fail_address_parse;
 592	}
 593
 594	/* Parse reference counter offset if specified. */
 595	rctr = strchr(arg, '(');
 596	if (rctr) {
 597		rctr_end = strchr(rctr, ')');
 598		if (!rctr_end) {
 599			ret = -EINVAL;
 600			rctr_end = rctr + strlen(rctr);
 601			trace_probe_log_err(rctr_end - filename,
 602					    REFCNT_OPEN_BRACE);
 603			goto fail_address_parse;
 604		} else if (rctr_end[1] != '\0') {
 605			ret = -EINVAL;
 606			trace_probe_log_err(rctr_end + 1 - filename,
 607					    BAD_REFCNT_SUFFIX);
 608			goto fail_address_parse;
 609		}
 610
 611		*rctr++ = '\0';
 612		*rctr_end = '\0';
 613		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 614		if (ret) {
 615			trace_probe_log_err(rctr - filename, BAD_REFCNT);
 616			goto fail_address_parse;
 617		}
 618	}
 619
 620	/* Check if there is %return suffix */
 621	tmp = strchr(arg, '%');
 622	if (tmp) {
 623		if (!strcmp(tmp, "%return")) {
 624			*tmp = '\0';
 625			is_return = true;
 626		} else {
 627			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 628			ret = -EINVAL;
 629			goto fail_address_parse;
 630		}
 631	}
 632
 633	/* Parse uprobe offset. */
 634	ret = kstrtoul(arg, 0, &offset);
 635	if (ret) {
 636		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 637		goto fail_address_parse;
 638	}
 639
 640	/* setup a probe */
 641	trace_probe_log_set_index(0);
 642	if (event) {
 643		ret = traceprobe_parse_event_name(&event, &group, gbuf,
 644						  event - argv[0]);
 645		if (ret)
 646			goto fail_address_parse;
 647	}
 648
 649	if (!event) {
 650		char *tail;
 651		char *ptr;
 652
 653		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 654		if (!tail) {
 655			ret = -ENOMEM;
 656			goto fail_address_parse;
 657		}
 658
 659		ptr = strpbrk(tail, ".-_");
 660		if (ptr)
 661			*ptr = '\0';
 662
 663		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 664		event = buf;
 665		kfree(tail);
 666	}
 667
 668	argc -= 2;
 669	argv += 2;
 670
 671	tu = alloc_trace_uprobe(group, event, argc, is_return);
 672	if (IS_ERR(tu)) {
 673		ret = PTR_ERR(tu);
 674		/* This must return -ENOMEM otherwise there is a bug */
 675		WARN_ON_ONCE(ret != -ENOMEM);
 676		goto fail_address_parse;
 677	}
 678	tu->offset = offset;
 679	tu->ref_ctr_offset = ref_ctr_offset;
 680	tu->path = path;
 681	tu->filename = filename;
 682
 683	/* parse arguments */
 684	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 685		struct traceprobe_parse_context ctx = {
 686			.flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
 687		};
 688
 689		trace_probe_log_set_index(i + 2);
 690		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
 691		traceprobe_finish_parse(&ctx);
 
 692		if (ret)
 693			goto error;
 694	}
 695
 696	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 697	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 698	if (ret < 0)
 699		goto error;
 700
 701	ret = register_trace_uprobe(tu);
 702	if (!ret)
 703		goto out;
 704
 705error:
 706	free_trace_uprobe(tu);
 707out:
 708	trace_probe_log_clear();
 709	return ret;
 710
 711fail_address_parse:
 712	trace_probe_log_clear();
 713	path_put(&path);
 714	kfree(filename);
 715
 716	return ret;
 717}
 718
 719int trace_uprobe_create(const char *raw_command)
 720{
 721	return trace_probe_create(raw_command, __trace_uprobe_create);
 722}
 723
 724static int create_or_delete_trace_uprobe(const char *raw_command)
 725{
 726	int ret;
 727
 728	if (raw_command[0] == '-')
 729		return dyn_event_release(raw_command, &trace_uprobe_ops);
 730
 731	ret = trace_uprobe_create(raw_command);
 732	return ret == -ECANCELED ? -EINVAL : ret;
 733}
 734
 735static int trace_uprobe_release(struct dyn_event *ev)
 736{
 737	struct trace_uprobe *tu = to_trace_uprobe(ev);
 738
 739	return unregister_trace_uprobe(tu);
 740}
 741
 742/* Probes listing interfaces */
 743static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 744{
 745	struct trace_uprobe *tu = to_trace_uprobe(ev);
 746	char c = is_ret_probe(tu) ? 'r' : 'p';
 747	int i;
 748
 749	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 750			trace_probe_name(&tu->tp), tu->filename,
 751			(int)(sizeof(void *) * 2), tu->offset);
 752
 753	if (tu->ref_ctr_offset)
 754		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 755
 756	for (i = 0; i < tu->tp.nr_args; i++)
 757		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 758
 759	seq_putc(m, '\n');
 760	return 0;
 761}
 762
 763static int probes_seq_show(struct seq_file *m, void *v)
 764{
 765	struct dyn_event *ev = v;
 766
 767	if (!is_trace_uprobe(ev))
 768		return 0;
 769
 770	return trace_uprobe_show(m, ev);
 771}
 772
 773static const struct seq_operations probes_seq_op = {
 774	.start  = dyn_event_seq_start,
 775	.next   = dyn_event_seq_next,
 776	.stop   = dyn_event_seq_stop,
 777	.show   = probes_seq_show
 778};
 779
 780static int probes_open(struct inode *inode, struct file *file)
 781{
 782	int ret;
 783
 784	ret = security_locked_down(LOCKDOWN_TRACEFS);
 785	if (ret)
 786		return ret;
 787
 788	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 789		ret = dyn_events_release_all(&trace_uprobe_ops);
 790		if (ret)
 791			return ret;
 792	}
 793
 794	return seq_open(file, &probes_seq_op);
 795}
 796
 797static ssize_t probes_write(struct file *file, const char __user *buffer,
 798			    size_t count, loff_t *ppos)
 799{
 800	return trace_parse_run_command(file, buffer, count, ppos,
 801					create_or_delete_trace_uprobe);
 802}
 803
 804static const struct file_operations uprobe_events_ops = {
 805	.owner		= THIS_MODULE,
 806	.open		= probes_open,
 807	.read		= seq_read,
 808	.llseek		= seq_lseek,
 809	.release	= seq_release,
 810	.write		= probes_write,
 811};
 812
 813/* Probes profiling interfaces */
 814static int probes_profile_seq_show(struct seq_file *m, void *v)
 815{
 816	struct dyn_event *ev = v;
 817	struct trace_uprobe *tu;
 818
 819	if (!is_trace_uprobe(ev))
 820		return 0;
 821
 822	tu = to_trace_uprobe(ev);
 823	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 824			trace_probe_name(&tu->tp), tu->nhit);
 825	return 0;
 826}
 827
 828static const struct seq_operations profile_seq_op = {
 829	.start  = dyn_event_seq_start,
 830	.next   = dyn_event_seq_next,
 831	.stop   = dyn_event_seq_stop,
 832	.show	= probes_profile_seq_show
 833};
 834
 835static int profile_open(struct inode *inode, struct file *file)
 836{
 837	int ret;
 838
 839	ret = security_locked_down(LOCKDOWN_TRACEFS);
 840	if (ret)
 841		return ret;
 842
 843	return seq_open(file, &profile_seq_op);
 844}
 845
 846static const struct file_operations uprobe_profile_ops = {
 847	.owner		= THIS_MODULE,
 848	.open		= profile_open,
 849	.read		= seq_read,
 850	.llseek		= seq_lseek,
 851	.release	= seq_release,
 852};
 853
 854struct uprobe_cpu_buffer {
 855	struct mutex mutex;
 856	void *buf;
 857};
 858static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 859static int uprobe_buffer_refcnt;
 860
 861static int uprobe_buffer_init(void)
 862{
 863	int cpu, err_cpu;
 864
 865	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 866	if (uprobe_cpu_buffer == NULL)
 867		return -ENOMEM;
 868
 869	for_each_possible_cpu(cpu) {
 870		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 871						  GFP_KERNEL, 0);
 872		if (p == NULL) {
 873			err_cpu = cpu;
 874			goto err;
 875		}
 876		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 877		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 878	}
 879
 880	return 0;
 881
 882err:
 883	for_each_possible_cpu(cpu) {
 884		if (cpu == err_cpu)
 885			break;
 886		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 887	}
 888
 889	free_percpu(uprobe_cpu_buffer);
 890	return -ENOMEM;
 891}
 892
 893static int uprobe_buffer_enable(void)
 894{
 895	int ret = 0;
 896
 897	BUG_ON(!mutex_is_locked(&event_mutex));
 898
 899	if (uprobe_buffer_refcnt++ == 0) {
 900		ret = uprobe_buffer_init();
 901		if (ret < 0)
 902			uprobe_buffer_refcnt--;
 903	}
 904
 905	return ret;
 906}
 907
 908static void uprobe_buffer_disable(void)
 909{
 910	int cpu;
 911
 912	BUG_ON(!mutex_is_locked(&event_mutex));
 913
 914	if (--uprobe_buffer_refcnt == 0) {
 915		for_each_possible_cpu(cpu)
 916			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 917							     cpu)->buf);
 918
 919		free_percpu(uprobe_cpu_buffer);
 920		uprobe_cpu_buffer = NULL;
 921	}
 922}
 923
 924static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 925{
 926	struct uprobe_cpu_buffer *ucb;
 927	int cpu;
 928
 929	cpu = raw_smp_processor_id();
 930	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 931
 932	/*
 933	 * Use per-cpu buffers for fastest access, but we might migrate
 934	 * so the mutex makes sure we have sole access to it.
 935	 */
 936	mutex_lock(&ucb->mutex);
 937
 938	return ucb;
 939}
 940
 941static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 942{
 943	mutex_unlock(&ucb->mutex);
 944}
 945
 946static void __uprobe_trace_func(struct trace_uprobe *tu,
 947				unsigned long func, struct pt_regs *regs,
 948				struct uprobe_cpu_buffer *ucb, int dsize,
 949				struct trace_event_file *trace_file)
 950{
 951	struct uprobe_trace_entry_head *entry;
 952	struct trace_event_buffer fbuffer;
 953	void *data;
 954	int size, esize;
 955	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 956
 957	WARN_ON(call != trace_file->event_call);
 958
 959	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 960		return;
 961
 962	if (trace_trigger_soft_disabled(trace_file))
 963		return;
 964
 965	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 966	size = esize + tu->tp.size + dsize;
 967	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 968	if (!entry)
 969		return;
 970
 971	if (is_ret_probe(tu)) {
 972		entry->vaddr[0] = func;
 973		entry->vaddr[1] = instruction_pointer(regs);
 974		data = DATAOF_TRACE_ENTRY(entry, true);
 975	} else {
 976		entry->vaddr[0] = instruction_pointer(regs);
 977		data = DATAOF_TRACE_ENTRY(entry, false);
 978	}
 979
 980	memcpy(data, ucb->buf, tu->tp.size + dsize);
 981
 982	trace_event_buffer_commit(&fbuffer);
 983}
 984
 985/* uprobe handler */
 986static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 987			     struct uprobe_cpu_buffer *ucb, int dsize)
 988{
 989	struct event_file_link *link;
 990
 991	if (is_ret_probe(tu))
 992		return 0;
 993
 994	rcu_read_lock();
 995	trace_probe_for_each_link_rcu(link, &tu->tp)
 996		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 997	rcu_read_unlock();
 998
 999	return 0;
1000}
1001
1002static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1003				 struct pt_regs *regs,
1004				 struct uprobe_cpu_buffer *ucb, int dsize)
1005{
1006	struct event_file_link *link;
1007
1008	rcu_read_lock();
1009	trace_probe_for_each_link_rcu(link, &tu->tp)
1010		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1011	rcu_read_unlock();
1012}
1013
1014/* Event entry printers */
1015static enum print_line_t
1016print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1017{
1018	struct uprobe_trace_entry_head *entry;
1019	struct trace_seq *s = &iter->seq;
1020	struct trace_uprobe *tu;
1021	u8 *data;
1022
1023	entry = (struct uprobe_trace_entry_head *)iter->ent;
1024	tu = trace_uprobe_primary_from_call(
1025		container_of(event, struct trace_event_call, event));
1026	if (unlikely(!tu))
1027		goto out;
1028
1029	if (is_ret_probe(tu)) {
1030		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1031				 trace_probe_name(&tu->tp),
1032				 entry->vaddr[1], entry->vaddr[0]);
1033		data = DATAOF_TRACE_ENTRY(entry, true);
1034	} else {
1035		trace_seq_printf(s, "%s: (0x%lx)",
1036				 trace_probe_name(&tu->tp),
1037				 entry->vaddr[0]);
1038		data = DATAOF_TRACE_ENTRY(entry, false);
1039	}
1040
1041	if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1042		goto out;
1043
1044	trace_seq_putc(s, '\n');
1045
1046 out:
1047	return trace_handle_return(s);
1048}
1049
1050typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1051				enum uprobe_filter_ctx ctx,
1052				struct mm_struct *mm);
1053
1054static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1055{
1056	int ret;
1057
1058	tu->consumer.filter = filter;
1059	tu->inode = d_real_inode(tu->path.dentry);
1060
1061	if (tu->ref_ctr_offset)
1062		ret = uprobe_register_refctr(tu->inode, tu->offset,
1063				tu->ref_ctr_offset, &tu->consumer);
1064	else
1065		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1066
1067	if (ret)
1068		tu->inode = NULL;
1069
1070	return ret;
1071}
1072
1073static void __probe_event_disable(struct trace_probe *tp)
1074{
1075	struct trace_uprobe *tu;
1076
1077	tu = container_of(tp, struct trace_uprobe, tp);
1078	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1079
1080	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1081		if (!tu->inode)
1082			continue;
1083
1084		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1085		tu->inode = NULL;
1086	}
1087}
1088
1089static int probe_event_enable(struct trace_event_call *call,
1090			struct trace_event_file *file, filter_func_t filter)
1091{
1092	struct trace_probe *tp;
1093	struct trace_uprobe *tu;
1094	bool enabled;
1095	int ret;
1096
1097	tp = trace_probe_primary_from_call(call);
1098	if (WARN_ON_ONCE(!tp))
1099		return -ENODEV;
1100	enabled = trace_probe_is_enabled(tp);
1101
1102	/* This may also change "enabled" state */
1103	if (file) {
1104		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1105			return -EINTR;
1106
1107		ret = trace_probe_add_file(tp, file);
1108		if (ret < 0)
1109			return ret;
1110	} else {
1111		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1112			return -EINTR;
1113
1114		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1115	}
1116
1117	tu = container_of(tp, struct trace_uprobe, tp);
1118	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1119
1120	if (enabled)
1121		return 0;
1122
1123	ret = uprobe_buffer_enable();
1124	if (ret)
1125		goto err_flags;
1126
1127	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1128		ret = trace_uprobe_enable(tu, filter);
1129		if (ret) {
1130			__probe_event_disable(tp);
1131			goto err_buffer;
1132		}
1133	}
1134
1135	return 0;
1136
1137 err_buffer:
1138	uprobe_buffer_disable();
1139
1140 err_flags:
1141	if (file)
1142		trace_probe_remove_file(tp, file);
1143	else
1144		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1145
1146	return ret;
1147}
1148
1149static void probe_event_disable(struct trace_event_call *call,
1150				struct trace_event_file *file)
1151{
1152	struct trace_probe *tp;
1153
1154	tp = trace_probe_primary_from_call(call);
1155	if (WARN_ON_ONCE(!tp))
1156		return;
1157
1158	if (!trace_probe_is_enabled(tp))
1159		return;
1160
1161	if (file) {
1162		if (trace_probe_remove_file(tp, file) < 0)
1163			return;
1164
1165		if (trace_probe_is_enabled(tp))
1166			return;
1167	} else
1168		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1169
1170	__probe_event_disable(tp);
1171	uprobe_buffer_disable();
1172}
1173
1174static int uprobe_event_define_fields(struct trace_event_call *event_call)
1175{
1176	int ret, size;
1177	struct uprobe_trace_entry_head field;
1178	struct trace_uprobe *tu;
1179
1180	tu = trace_uprobe_primary_from_call(event_call);
1181	if (unlikely(!tu))
1182		return -ENODEV;
1183
1184	if (is_ret_probe(tu)) {
1185		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1186		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1187		size = SIZEOF_TRACE_ENTRY(true);
1188	} else {
1189		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1190		size = SIZEOF_TRACE_ENTRY(false);
1191	}
1192
1193	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1194}
1195
1196#ifdef CONFIG_PERF_EVENTS
1197static bool
1198__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1199{
1200	struct perf_event *event;
1201
1202	if (filter->nr_systemwide)
1203		return true;
1204
1205	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1206		if (event->hw.target->mm == mm)
1207			return true;
1208	}
1209
1210	return false;
1211}
1212
1213static inline bool
1214trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1215			  struct perf_event *event)
1216{
1217	return __uprobe_perf_filter(filter, event->hw.target->mm);
1218}
1219
1220static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1221				       struct perf_event *event)
1222{
1223	bool done;
1224
1225	write_lock(&filter->rwlock);
1226	if (event->hw.target) {
1227		list_del(&event->hw.tp_list);
1228		done = filter->nr_systemwide ||
1229			(event->hw.target->flags & PF_EXITING) ||
1230			trace_uprobe_filter_event(filter, event);
1231	} else {
1232		filter->nr_systemwide--;
1233		done = filter->nr_systemwide;
1234	}
1235	write_unlock(&filter->rwlock);
1236
1237	return done;
1238}
1239
1240/* This returns true if the filter always covers target mm */
1241static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1242				    struct perf_event *event)
1243{
1244	bool done;
1245
1246	write_lock(&filter->rwlock);
1247	if (event->hw.target) {
1248		/*
1249		 * event->parent != NULL means copy_process(), we can avoid
1250		 * uprobe_apply(). current->mm must be probed and we can rely
1251		 * on dup_mmap() which preserves the already installed bp's.
1252		 *
1253		 * attr.enable_on_exec means that exec/mmap will install the
1254		 * breakpoints we need.
1255		 */
1256		done = filter->nr_systemwide ||
1257			event->parent || event->attr.enable_on_exec ||
1258			trace_uprobe_filter_event(filter, event);
1259		list_add(&event->hw.tp_list, &filter->perf_events);
1260	} else {
1261		done = filter->nr_systemwide;
1262		filter->nr_systemwide++;
1263	}
1264	write_unlock(&filter->rwlock);
1265
1266	return done;
1267}
1268
1269static int uprobe_perf_close(struct trace_event_call *call,
1270			     struct perf_event *event)
1271{
1272	struct trace_probe *tp;
1273	struct trace_uprobe *tu;
1274	int ret = 0;
1275
1276	tp = trace_probe_primary_from_call(call);
1277	if (WARN_ON_ONCE(!tp))
1278		return -ENODEV;
1279
1280	tu = container_of(tp, struct trace_uprobe, tp);
1281	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1282		return 0;
1283
1284	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1285		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1286		if (ret)
1287			break;
1288	}
1289
1290	return ret;
1291}
1292
1293static int uprobe_perf_open(struct trace_event_call *call,
1294			    struct perf_event *event)
1295{
1296	struct trace_probe *tp;
1297	struct trace_uprobe *tu;
1298	int err = 0;
1299
1300	tp = trace_probe_primary_from_call(call);
1301	if (WARN_ON_ONCE(!tp))
1302		return -ENODEV;
1303
1304	tu = container_of(tp, struct trace_uprobe, tp);
1305	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1306		return 0;
1307
1308	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1309		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1310		if (err) {
1311			uprobe_perf_close(call, event);
1312			break;
1313		}
1314	}
1315
1316	return err;
1317}
1318
1319static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1320				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1321{
1322	struct trace_uprobe_filter *filter;
1323	struct trace_uprobe *tu;
1324	int ret;
1325
1326	tu = container_of(uc, struct trace_uprobe, consumer);
1327	filter = tu->tp.event->filter;
1328
1329	read_lock(&filter->rwlock);
1330	ret = __uprobe_perf_filter(filter, mm);
1331	read_unlock(&filter->rwlock);
1332
1333	return ret;
1334}
1335
1336static void __uprobe_perf_func(struct trace_uprobe *tu,
1337			       unsigned long func, struct pt_regs *regs,
1338			       struct uprobe_cpu_buffer *ucb, int dsize)
1339{
1340	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1341	struct uprobe_trace_entry_head *entry;
1342	struct hlist_head *head;
1343	void *data;
1344	int size, esize;
1345	int rctx;
1346
1347#ifdef CONFIG_BPF_EVENTS
1348	if (bpf_prog_array_valid(call)) {
1349		u32 ret;
1350
1351		ret = bpf_prog_run_array_uprobe(call->prog_array, regs, bpf_prog_run);
1352		if (!ret)
1353			return;
1354	}
1355#endif /* CONFIG_BPF_EVENTS */
1356
1357	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1358
1359	size = esize + tu->tp.size + dsize;
1360	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1361	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1362		return;
1363
1364	preempt_disable();
1365	head = this_cpu_ptr(call->perf_events);
1366	if (hlist_empty(head))
1367		goto out;
1368
1369	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1370	if (!entry)
1371		goto out;
1372
1373	if (is_ret_probe(tu)) {
1374		entry->vaddr[0] = func;
1375		entry->vaddr[1] = instruction_pointer(regs);
1376		data = DATAOF_TRACE_ENTRY(entry, true);
1377	} else {
1378		entry->vaddr[0] = instruction_pointer(regs);
1379		data = DATAOF_TRACE_ENTRY(entry, false);
1380	}
1381
1382	memcpy(data, ucb->buf, tu->tp.size + dsize);
1383
1384	if (size - esize > tu->tp.size + dsize) {
1385		int len = tu->tp.size + dsize;
1386
1387		memset(data + len, 0, size - esize - len);
1388	}
1389
1390	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1391			      head, NULL);
1392 out:
1393	preempt_enable();
1394}
1395
1396/* uprobe profile handler */
1397static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1398			    struct uprobe_cpu_buffer *ucb, int dsize)
1399{
1400	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1401		return UPROBE_HANDLER_REMOVE;
1402
1403	if (!is_ret_probe(tu))
1404		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1405	return 0;
1406}
1407
1408static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1409				struct pt_regs *regs,
1410				struct uprobe_cpu_buffer *ucb, int dsize)
1411{
1412	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1413}
1414
1415int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1416			const char **filename, u64 *probe_offset,
1417			u64 *probe_addr, bool perf_type_tracepoint)
1418{
1419	const char *pevent = trace_event_name(event->tp_event);
1420	const char *group = event->tp_event->class->system;
1421	struct trace_uprobe *tu;
1422
1423	if (perf_type_tracepoint)
1424		tu = find_probe_event(pevent, group);
1425	else
1426		tu = trace_uprobe_primary_from_call(event->tp_event);
1427	if (!tu)
1428		return -EINVAL;
1429
1430	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1431				    : BPF_FD_TYPE_UPROBE;
1432	*filename = tu->filename;
1433	*probe_offset = tu->offset;
1434	*probe_addr = 0;
1435	return 0;
1436}
1437#endif	/* CONFIG_PERF_EVENTS */
1438
1439static int
1440trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1441		      void *data)
1442{
1443	struct trace_event_file *file = data;
1444
1445	switch (type) {
1446	case TRACE_REG_REGISTER:
1447		return probe_event_enable(event, file, NULL);
1448
1449	case TRACE_REG_UNREGISTER:
1450		probe_event_disable(event, file);
1451		return 0;
1452
1453#ifdef CONFIG_PERF_EVENTS
1454	case TRACE_REG_PERF_REGISTER:
1455		return probe_event_enable(event, NULL, uprobe_perf_filter);
1456
1457	case TRACE_REG_PERF_UNREGISTER:
1458		probe_event_disable(event, NULL);
1459		return 0;
1460
1461	case TRACE_REG_PERF_OPEN:
1462		return uprobe_perf_open(event, data);
1463
1464	case TRACE_REG_PERF_CLOSE:
1465		return uprobe_perf_close(event, data);
1466
1467#endif
1468	default:
1469		return 0;
1470	}
1471}
1472
1473static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1474{
1475	struct trace_uprobe *tu;
1476	struct uprobe_dispatch_data udd;
1477	struct uprobe_cpu_buffer *ucb;
1478	int dsize, esize;
1479	int ret = 0;
1480
1481
1482	tu = container_of(con, struct trace_uprobe, consumer);
1483	tu->nhit++;
1484
1485	udd.tu = tu;
1486	udd.bp_addr = instruction_pointer(regs);
1487
1488	current->utask->vaddr = (unsigned long) &udd;
1489
1490	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1491		return 0;
1492
1493	dsize = __get_data_size(&tu->tp, regs);
1494	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1495
1496	ucb = uprobe_buffer_get();
1497	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1498
1499	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1500		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1501
1502#ifdef CONFIG_PERF_EVENTS
1503	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1504		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1505#endif
1506	uprobe_buffer_put(ucb);
1507	return ret;
1508}
1509
1510static int uretprobe_dispatcher(struct uprobe_consumer *con,
1511				unsigned long func, struct pt_regs *regs)
1512{
1513	struct trace_uprobe *tu;
1514	struct uprobe_dispatch_data udd;
1515	struct uprobe_cpu_buffer *ucb;
1516	int dsize, esize;
1517
1518	tu = container_of(con, struct trace_uprobe, consumer);
1519
1520	udd.tu = tu;
1521	udd.bp_addr = func;
1522
1523	current->utask->vaddr = (unsigned long) &udd;
1524
1525	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1526		return 0;
1527
1528	dsize = __get_data_size(&tu->tp, regs);
1529	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1530
1531	ucb = uprobe_buffer_get();
1532	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1533
1534	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1535		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1536
1537#ifdef CONFIG_PERF_EVENTS
1538	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1539		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1540#endif
1541	uprobe_buffer_put(ucb);
1542	return 0;
1543}
1544
1545static struct trace_event_functions uprobe_funcs = {
1546	.trace		= print_uprobe_event
1547};
1548
1549static struct trace_event_fields uprobe_fields_array[] = {
1550	{ .type = TRACE_FUNCTION_TYPE,
1551	  .define_fields = uprobe_event_define_fields },
1552	{}
1553};
1554
1555static inline void init_trace_event_call(struct trace_uprobe *tu)
1556{
1557	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1558	call->event.funcs = &uprobe_funcs;
1559	call->class->fields_array = uprobe_fields_array;
1560
1561	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1562	call->class->reg = trace_uprobe_register;
1563}
1564
1565static int register_uprobe_event(struct trace_uprobe *tu)
1566{
1567	init_trace_event_call(tu);
1568
1569	return trace_probe_register_event_call(&tu->tp);
1570}
1571
1572static int unregister_uprobe_event(struct trace_uprobe *tu)
1573{
1574	return trace_probe_unregister_event_call(&tu->tp);
1575}
1576
1577#ifdef CONFIG_PERF_EVENTS
1578struct trace_event_call *
1579create_local_trace_uprobe(char *name, unsigned long offs,
1580			  unsigned long ref_ctr_offset, bool is_return)
1581{
1582	enum probe_print_type ptype;
1583	struct trace_uprobe *tu;
1584	struct path path;
1585	int ret;
1586
1587	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1588	if (ret)
1589		return ERR_PTR(ret);
1590
1591	if (!d_is_reg(path.dentry)) {
1592		path_put(&path);
1593		return ERR_PTR(-EINVAL);
1594	}
1595
1596	/*
1597	 * local trace_kprobes are not added to dyn_event, so they are never
1598	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1599	 * duplicated name "DUMMY_EVENT" here.
1600	 */
1601	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1602				is_return);
1603
1604	if (IS_ERR(tu)) {
1605		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1606			(int)PTR_ERR(tu));
1607		path_put(&path);
1608		return ERR_CAST(tu);
1609	}
1610
1611	tu->offset = offs;
1612	tu->path = path;
1613	tu->ref_ctr_offset = ref_ctr_offset;
1614	tu->filename = kstrdup(name, GFP_KERNEL);
1615	if (!tu->filename) {
1616		ret = -ENOMEM;
1617		goto error;
1618	}
1619
1620	init_trace_event_call(tu);
1621
1622	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1623	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1624		ret = -ENOMEM;
1625		goto error;
1626	}
1627
1628	return trace_probe_event_call(&tu->tp);
1629error:
1630	free_trace_uprobe(tu);
1631	return ERR_PTR(ret);
1632}
1633
1634void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1635{
1636	struct trace_uprobe *tu;
1637
1638	tu = trace_uprobe_primary_from_call(event_call);
1639
1640	free_trace_uprobe(tu);
1641}
1642#endif /* CONFIG_PERF_EVENTS */
1643
1644/* Make a trace interface for controlling probe points */
1645static __init int init_uprobe_trace(void)
1646{
1647	int ret;
1648
1649	ret = dyn_event_register(&trace_uprobe_ops);
1650	if (ret)
1651		return ret;
1652
1653	ret = tracing_init_dentry();
1654	if (ret)
1655		return 0;
1656
1657	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1658				    NULL, &uprobe_events_ops);
1659	/* Profile interface */
1660	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1661				    NULL, &uprobe_profile_ops);
1662	return 0;
1663}
1664
1665fs_initcall(init_uprobe_trace);
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * uprobes-based tracing events
   4 *
   5 * Copyright (C) IBM Corporation, 2010-2012
   6 * Author:	Srikar Dronamraju <srikar@linux.vnet.ibm.com>
   7 */
   8#define pr_fmt(fmt)	"trace_uprobe: " fmt
   9
  10#include <linux/bpf-cgroup.h>
  11#include <linux/security.h>
  12#include <linux/ctype.h>
  13#include <linux/module.h>
  14#include <linux/uaccess.h>
  15#include <linux/uprobes.h>
  16#include <linux/namei.h>
  17#include <linux/string.h>
  18#include <linux/rculist.h>
  19#include <linux/filter.h>
  20
  21#include "trace_dynevent.h"
  22#include "trace_probe.h"
  23#include "trace_probe_tmpl.h"
  24
  25#define UPROBE_EVENT_SYSTEM	"uprobes"
  26
  27struct uprobe_trace_entry_head {
  28	struct trace_entry	ent;
  29	unsigned long		vaddr[];
  30};
  31
  32#define SIZEOF_TRACE_ENTRY(is_return)			\
  33	(sizeof(struct uprobe_trace_entry_head) +	\
  34	 sizeof(unsigned long) * (is_return ? 2 : 1))
  35
  36#define DATAOF_TRACE_ENTRY(entry, is_return)		\
  37	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  38
  39static int trace_uprobe_create(const char *raw_command);
  40static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  41static int trace_uprobe_release(struct dyn_event *ev);
  42static bool trace_uprobe_is_busy(struct dyn_event *ev);
  43static bool trace_uprobe_match(const char *system, const char *event,
  44			int argc, const char **argv, struct dyn_event *ev);
  45
  46static struct dyn_event_operations trace_uprobe_ops = {
  47	.create = trace_uprobe_create,
  48	.show = trace_uprobe_show,
  49	.is_busy = trace_uprobe_is_busy,
  50	.free = trace_uprobe_release,
  51	.match = trace_uprobe_match,
  52};
  53
  54/*
  55 * uprobe event core functions
  56 */
  57struct trace_uprobe {
  58	struct dyn_event		devent;
  59	struct uprobe_consumer		consumer;
  60	struct path			path;
  61	struct inode			*inode;
  62	char				*filename;
  63	unsigned long			offset;
  64	unsigned long			ref_ctr_offset;
  65	unsigned long			nhit;
  66	struct trace_probe		tp;
  67};
  68
  69static bool is_trace_uprobe(struct dyn_event *ev)
  70{
  71	return ev->ops == &trace_uprobe_ops;
  72}
  73
  74static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  75{
  76	return container_of(ev, struct trace_uprobe, devent);
  77}
  78
  79/**
  80 * for_each_trace_uprobe - iterate over the trace_uprobe list
  81 * @pos:	the struct trace_uprobe * for each entry
  82 * @dpos:	the struct dyn_event * to use as a loop cursor
  83 */
  84#define for_each_trace_uprobe(pos, dpos)	\
  85	for_each_dyn_event(dpos)		\
  86		if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  87
  88static int register_uprobe_event(struct trace_uprobe *tu);
  89static int unregister_uprobe_event(struct trace_uprobe *tu);
  90
  91struct uprobe_dispatch_data {
  92	struct trace_uprobe	*tu;
  93	unsigned long		bp_addr;
  94};
  95
  96static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  97static int uretprobe_dispatcher(struct uprobe_consumer *con,
  98				unsigned long func, struct pt_regs *regs);
  99
 100#ifdef CONFIG_STACK_GROWSUP
 101static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 102{
 103	return addr - (n * sizeof(long));
 104}
 105#else
 106static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 107{
 108	return addr + (n * sizeof(long));
 109}
 110#endif
 111
 112static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 113{
 114	unsigned long ret;
 115	unsigned long addr = user_stack_pointer(regs);
 116
 117	addr = adjust_stack_addr(addr, n);
 118
 119	if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 120		return 0;
 121
 122	return ret;
 123}
 124
 125/*
 126 * Uprobes-specific fetch functions
 127 */
 128static nokprobe_inline int
 129probe_mem_read(void *dest, void *src, size_t size)
 130{
 131	void __user *vaddr = (void __force __user *)src;
 132
 133	return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 134}
 135
 136static nokprobe_inline int
 137probe_mem_read_user(void *dest, void *src, size_t size)
 138{
 139	return probe_mem_read(dest, src, size);
 140}
 141
 142/*
 143 * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 144 * length and relative data location.
 145 */
 146static nokprobe_inline int
 147fetch_store_string(unsigned long addr, void *dest, void *base)
 148{
 149	long ret;
 150	u32 loc = *(u32 *)dest;
 151	int maxlen  = get_loc_len(loc);
 152	u8 *dst = get_loc_data(dest, base);
 153	void __user *src = (void __force __user *) addr;
 154
 155	if (unlikely(!maxlen))
 156		return -ENOMEM;
 157
 158	if (addr == FETCH_TOKEN_COMM)
 159		ret = strlcpy(dst, current->comm, maxlen);
 160	else
 161		ret = strncpy_from_user(dst, src, maxlen);
 162	if (ret >= 0) {
 163		if (ret == maxlen)
 164			dst[ret - 1] = '\0';
 165		else
 166			/*
 167			 * Include the terminating null byte. In this case it
 168			 * was copied by strncpy_from_user but not accounted
 169			 * for in ret.
 170			 */
 171			ret++;
 172		*(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 173	}
 
 174
 175	return ret;
 176}
 177
 178static nokprobe_inline int
 179fetch_store_string_user(unsigned long addr, void *dest, void *base)
 180{
 181	return fetch_store_string(addr, dest, base);
 182}
 183
 184/* Return the length of string -- including null terminal byte */
 185static nokprobe_inline int
 186fetch_store_strlen(unsigned long addr)
 187{
 188	int len;
 189	void __user *vaddr = (void __force __user *) addr;
 190
 191	if (addr == FETCH_TOKEN_COMM)
 192		len = strlen(current->comm) + 1;
 193	else
 194		len = strnlen_user(vaddr, MAX_STRING_SIZE);
 195
 196	return (len > MAX_STRING_SIZE) ? 0 : len;
 197}
 198
 199static nokprobe_inline int
 200fetch_store_strlen_user(unsigned long addr)
 201{
 202	return fetch_store_strlen(addr);
 203}
 204
 205static unsigned long translate_user_vaddr(unsigned long file_offset)
 206{
 207	unsigned long base_addr;
 208	struct uprobe_dispatch_data *udd;
 209
 210	udd = (void *) current->utask->vaddr;
 211
 212	base_addr = udd->bp_addr - udd->tu->offset;
 213	return base_addr + file_offset;
 214}
 215
 216/* Note that we don't verify it, since the code does not come from user space */
 217static int
 218process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 219		   void *base)
 220{
 221	struct pt_regs *regs = rec;
 222	unsigned long val;
 
 223
 224	/* 1st stage: get value from context */
 225	switch (code->op) {
 226	case FETCH_OP_REG:
 227		val = regs_get_register(regs, code->param);
 228		break;
 229	case FETCH_OP_STACK:
 230		val = get_user_stack_nth(regs, code->param);
 231		break;
 232	case FETCH_OP_STACKP:
 233		val = user_stack_pointer(regs);
 234		break;
 235	case FETCH_OP_RETVAL:
 236		val = regs_return_value(regs);
 237		break;
 238	case FETCH_OP_IMM:
 239		val = code->immediate;
 240		break;
 241	case FETCH_OP_COMM:
 242		val = FETCH_TOKEN_COMM;
 243		break;
 244	case FETCH_OP_DATA:
 245		val = (unsigned long)code->data;
 246		break;
 247	case FETCH_OP_FOFFS:
 248		val = translate_user_vaddr(code->immediate);
 249		break;
 250	default:
 251		return -EILSEQ;
 
 
 252	}
 253	code++;
 254
 255	return process_fetch_insn_bottom(code, val, dest, base);
 256}
 257NOKPROBE_SYMBOL(process_fetch_insn)
 258
 259static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 260{
 261	rwlock_init(&filter->rwlock);
 262	filter->nr_systemwide = 0;
 263	INIT_LIST_HEAD(&filter->perf_events);
 264}
 265
 266static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 267{
 268	return !filter->nr_systemwide && list_empty(&filter->perf_events);
 269}
 270
 271static inline bool is_ret_probe(struct trace_uprobe *tu)
 272{
 273	return tu->consumer.ret_handler != NULL;
 274}
 275
 276static bool trace_uprobe_is_busy(struct dyn_event *ev)
 277{
 278	struct trace_uprobe *tu = to_trace_uprobe(ev);
 279
 280	return trace_probe_is_enabled(&tu->tp);
 281}
 282
 283static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 284					    int argc, const char **argv)
 285{
 286	char buf[MAX_ARGSTR_LEN + 1];
 287	int len;
 288
 289	if (!argc)
 290		return true;
 291
 292	len = strlen(tu->filename);
 293	if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 294		return false;
 295
 296	if (tu->ref_ctr_offset == 0)
 297		snprintf(buf, sizeof(buf), "0x%0*lx",
 298				(int)(sizeof(void *) * 2), tu->offset);
 299	else
 300		snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 301				(int)(sizeof(void *) * 2), tu->offset,
 302				tu->ref_ctr_offset);
 303	if (strcmp(buf, &argv[0][len + 1]))
 304		return false;
 305
 306	argc--; argv++;
 307
 308	return trace_probe_match_command_args(&tu->tp, argc, argv);
 309}
 310
 311static bool trace_uprobe_match(const char *system, const char *event,
 312			int argc, const char **argv, struct dyn_event *ev)
 313{
 314	struct trace_uprobe *tu = to_trace_uprobe(ev);
 315
 316	return (event[0] == '\0' ||
 317		strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 318	   (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 319	   trace_uprobe_match_command_head(tu, argc, argv);
 320}
 321
 322static nokprobe_inline struct trace_uprobe *
 323trace_uprobe_primary_from_call(struct trace_event_call *call)
 324{
 325	struct trace_probe *tp;
 326
 327	tp = trace_probe_primary_from_call(call);
 328	if (WARN_ON_ONCE(!tp))
 329		return NULL;
 330
 331	return container_of(tp, struct trace_uprobe, tp);
 332}
 333
 334/*
 335 * Allocate new trace_uprobe and initialize it (including uprobes).
 336 */
 337static struct trace_uprobe *
 338alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 339{
 340	struct trace_uprobe *tu;
 341	int ret;
 342
 343	tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 344	if (!tu)
 345		return ERR_PTR(-ENOMEM);
 346
 347	ret = trace_probe_init(&tu->tp, event, group, true);
 348	if (ret < 0)
 349		goto error;
 350
 351	dyn_event_init(&tu->devent, &trace_uprobe_ops);
 352	tu->consumer.handler = uprobe_dispatcher;
 353	if (is_ret)
 354		tu->consumer.ret_handler = uretprobe_dispatcher;
 355	init_trace_uprobe_filter(tu->tp.event->filter);
 356	return tu;
 357
 358error:
 359	kfree(tu);
 360
 361	return ERR_PTR(ret);
 362}
 363
 364static void free_trace_uprobe(struct trace_uprobe *tu)
 365{
 366	if (!tu)
 367		return;
 368
 369	path_put(&tu->path);
 370	trace_probe_cleanup(&tu->tp);
 371	kfree(tu->filename);
 372	kfree(tu);
 373}
 374
 375static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 376{
 377	struct dyn_event *pos;
 378	struct trace_uprobe *tu;
 379
 380	for_each_trace_uprobe(tu, pos)
 381		if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 382		    strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 383			return tu;
 384
 385	return NULL;
 386}
 387
 388/* Unregister a trace_uprobe and probe_event */
 389static int unregister_trace_uprobe(struct trace_uprobe *tu)
 390{
 391	int ret;
 392
 393	if (trace_probe_has_sibling(&tu->tp))
 394		goto unreg;
 395
 396	/* If there's a reference to the dynamic event */
 397	if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 398		return -EBUSY;
 399
 400	ret = unregister_uprobe_event(tu);
 401	if (ret)
 402		return ret;
 403
 404unreg:
 405	dyn_event_remove(&tu->devent);
 406	trace_probe_unlink(&tu->tp);
 407	free_trace_uprobe(tu);
 408	return 0;
 409}
 410
 411static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 412					 struct trace_uprobe *comp)
 413{
 414	struct trace_probe_event *tpe = orig->tp.event;
 415	struct inode *comp_inode = d_real_inode(comp->path.dentry);
 416	int i;
 417
 418	list_for_each_entry(orig, &tpe->probes, tp.list) {
 419		if (comp_inode != d_real_inode(orig->path.dentry) ||
 420		    comp->offset != orig->offset)
 421			continue;
 422
 423		/*
 424		 * trace_probe_compare_arg_type() ensured that nr_args and
 425		 * each argument name and type are same. Let's compare comm.
 426		 */
 427		for (i = 0; i < orig->tp.nr_args; i++) {
 428			if (strcmp(orig->tp.args[i].comm,
 429				   comp->tp.args[i].comm))
 430				break;
 431		}
 432
 433		if (i == orig->tp.nr_args)
 434			return true;
 435	}
 436
 437	return false;
 438}
 439
 440static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 441{
 442	int ret;
 443
 444	ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 445	if (ret) {
 446		/* Note that argument starts index = 2 */
 447		trace_probe_log_set_index(ret + 1);
 448		trace_probe_log_err(0, DIFF_ARG_TYPE);
 449		return -EEXIST;
 450	}
 451	if (trace_uprobe_has_same_uprobe(to, tu)) {
 452		trace_probe_log_set_index(0);
 453		trace_probe_log_err(0, SAME_PROBE);
 454		return -EEXIST;
 455	}
 456
 457	/* Append to existing event */
 458	ret = trace_probe_append(&tu->tp, &to->tp);
 459	if (!ret)
 460		dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 461
 462	return ret;
 463}
 464
 465/*
 466 * Uprobe with multiple reference counter is not allowed. i.e.
 467 * If inode and offset matches, reference counter offset *must*
 468 * match as well. Though, there is one exception: If user is
 469 * replacing old trace_uprobe with new one(same group/event),
 470 * then we allow same uprobe with new reference counter as far
 471 * as the new one does not conflict with any other existing
 472 * ones.
 473 */
 474static int validate_ref_ctr_offset(struct trace_uprobe *new)
 475{
 476	struct dyn_event *pos;
 477	struct trace_uprobe *tmp;
 478	struct inode *new_inode = d_real_inode(new->path.dentry);
 479
 480	for_each_trace_uprobe(tmp, pos) {
 481		if (new_inode == d_real_inode(tmp->path.dentry) &&
 482		    new->offset == tmp->offset &&
 483		    new->ref_ctr_offset != tmp->ref_ctr_offset) {
 484			pr_warn("Reference counter offset mismatch.");
 485			return -EINVAL;
 486		}
 487	}
 488	return 0;
 489}
 490
 491/* Register a trace_uprobe and probe_event */
 492static int register_trace_uprobe(struct trace_uprobe *tu)
 493{
 494	struct trace_uprobe *old_tu;
 495	int ret;
 496
 497	mutex_lock(&event_mutex);
 498
 499	ret = validate_ref_ctr_offset(tu);
 500	if (ret)
 501		goto end;
 502
 503	/* register as an event */
 504	old_tu = find_probe_event(trace_probe_name(&tu->tp),
 505				  trace_probe_group_name(&tu->tp));
 506	if (old_tu) {
 507		if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 508			trace_probe_log_set_index(0);
 509			trace_probe_log_err(0, DIFF_PROBE_TYPE);
 510			ret = -EEXIST;
 511		} else {
 512			ret = append_trace_uprobe(tu, old_tu);
 513		}
 514		goto end;
 515	}
 516
 517	ret = register_uprobe_event(tu);
 518	if (ret) {
 519		if (ret == -EEXIST) {
 520			trace_probe_log_set_index(0);
 521			trace_probe_log_err(0, EVENT_EXIST);
 522		} else
 523			pr_warn("Failed to register probe event(%d)\n", ret);
 524		goto end;
 525	}
 526
 527	dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 528
 529end:
 530	mutex_unlock(&event_mutex);
 531
 532	return ret;
 533}
 534
 535/*
 536 * Argument syntax:
 537 *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 538 */
 539static int __trace_uprobe_create(int argc, const char **argv)
 540{
 541	struct trace_uprobe *tu;
 542	const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 543	char *arg, *filename, *rctr, *rctr_end, *tmp;
 544	char buf[MAX_EVENT_NAME_LEN];
 545	char gbuf[MAX_EVENT_NAME_LEN];
 546	enum probe_print_type ptype;
 547	struct path path;
 548	unsigned long offset, ref_ctr_offset;
 549	bool is_return = false;
 550	int i, ret;
 551
 552	ref_ctr_offset = 0;
 553
 554	switch (argv[0][0]) {
 555	case 'r':
 556		is_return = true;
 557		break;
 558	case 'p':
 559		break;
 560	default:
 561		return -ECANCELED;
 562	}
 563
 564	if (argc < 2)
 565		return -ECANCELED;
 566
 567	if (argv[0][1] == ':')
 568		event = &argv[0][2];
 569
 570	if (!strchr(argv[1], '/'))
 571		return -ECANCELED;
 572
 573	filename = kstrdup(argv[1], GFP_KERNEL);
 574	if (!filename)
 575		return -ENOMEM;
 576
 577	/* Find the last occurrence, in case the path contains ':' too. */
 578	arg = strrchr(filename, ':');
 579	if (!arg || !isdigit(arg[1])) {
 580		kfree(filename);
 581		return -ECANCELED;
 582	}
 583
 584	trace_probe_log_init("trace_uprobe", argc, argv);
 585	trace_probe_log_set_index(1);	/* filename is the 2nd argument */
 586
 587	*arg++ = '\0';
 588	ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 589	if (ret) {
 590		trace_probe_log_err(0, FILE_NOT_FOUND);
 591		kfree(filename);
 592		trace_probe_log_clear();
 593		return ret;
 594	}
 595	if (!d_is_reg(path.dentry)) {
 596		trace_probe_log_err(0, NO_REGULAR_FILE);
 597		ret = -EINVAL;
 598		goto fail_address_parse;
 599	}
 600
 601	/* Parse reference counter offset if specified. */
 602	rctr = strchr(arg, '(');
 603	if (rctr) {
 604		rctr_end = strchr(rctr, ')');
 605		if (!rctr_end) {
 606			ret = -EINVAL;
 607			rctr_end = rctr + strlen(rctr);
 608			trace_probe_log_err(rctr_end - filename,
 609					    REFCNT_OPEN_BRACE);
 610			goto fail_address_parse;
 611		} else if (rctr_end[1] != '\0') {
 612			ret = -EINVAL;
 613			trace_probe_log_err(rctr_end + 1 - filename,
 614					    BAD_REFCNT_SUFFIX);
 615			goto fail_address_parse;
 616		}
 617
 618		*rctr++ = '\0';
 619		*rctr_end = '\0';
 620		ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 621		if (ret) {
 622			trace_probe_log_err(rctr - filename, BAD_REFCNT);
 623			goto fail_address_parse;
 624		}
 625	}
 626
 627	/* Check if there is %return suffix */
 628	tmp = strchr(arg, '%');
 629	if (tmp) {
 630		if (!strcmp(tmp, "%return")) {
 631			*tmp = '\0';
 632			is_return = true;
 633		} else {
 634			trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 635			ret = -EINVAL;
 636			goto fail_address_parse;
 637		}
 638	}
 639
 640	/* Parse uprobe offset. */
 641	ret = kstrtoul(arg, 0, &offset);
 642	if (ret) {
 643		trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 644		goto fail_address_parse;
 645	}
 646
 647	/* setup a probe */
 648	trace_probe_log_set_index(0);
 649	if (event) {
 650		ret = traceprobe_parse_event_name(&event, &group, gbuf,
 651						  event - argv[0]);
 652		if (ret)
 653			goto fail_address_parse;
 654	}
 655
 656	if (!event) {
 657		char *tail;
 658		char *ptr;
 659
 660		tail = kstrdup(kbasename(filename), GFP_KERNEL);
 661		if (!tail) {
 662			ret = -ENOMEM;
 663			goto fail_address_parse;
 664		}
 665
 666		ptr = strpbrk(tail, ".-_");
 667		if (ptr)
 668			*ptr = '\0';
 669
 670		snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 671		event = buf;
 672		kfree(tail);
 673	}
 674
 675	argc -= 2;
 676	argv += 2;
 677
 678	tu = alloc_trace_uprobe(group, event, argc, is_return);
 679	if (IS_ERR(tu)) {
 680		ret = PTR_ERR(tu);
 681		/* This must return -ENOMEM otherwise there is a bug */
 682		WARN_ON_ONCE(ret != -ENOMEM);
 683		goto fail_address_parse;
 684	}
 685	tu->offset = offset;
 686	tu->ref_ctr_offset = ref_ctr_offset;
 687	tu->path = path;
 688	tu->filename = filename;
 689
 690	/* parse arguments */
 691	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 
 
 
 
 692		trace_probe_log_set_index(i + 2);
 693		ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
 694					(is_return ? TPARG_FL_RETURN : 0) |
 695					TPARG_FL_USER);
 696		if (ret)
 697			goto error;
 698	}
 699
 700	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 701	ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 702	if (ret < 0)
 703		goto error;
 704
 705	ret = register_trace_uprobe(tu);
 706	if (!ret)
 707		goto out;
 708
 709error:
 710	free_trace_uprobe(tu);
 711out:
 712	trace_probe_log_clear();
 713	return ret;
 714
 715fail_address_parse:
 716	trace_probe_log_clear();
 717	path_put(&path);
 718	kfree(filename);
 719
 720	return ret;
 721}
 722
 723int trace_uprobe_create(const char *raw_command)
 724{
 725	return trace_probe_create(raw_command, __trace_uprobe_create);
 726}
 727
 728static int create_or_delete_trace_uprobe(const char *raw_command)
 729{
 730	int ret;
 731
 732	if (raw_command[0] == '-')
 733		return dyn_event_release(raw_command, &trace_uprobe_ops);
 734
 735	ret = trace_uprobe_create(raw_command);
 736	return ret == -ECANCELED ? -EINVAL : ret;
 737}
 738
 739static int trace_uprobe_release(struct dyn_event *ev)
 740{
 741	struct trace_uprobe *tu = to_trace_uprobe(ev);
 742
 743	return unregister_trace_uprobe(tu);
 744}
 745
 746/* Probes listing interfaces */
 747static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 748{
 749	struct trace_uprobe *tu = to_trace_uprobe(ev);
 750	char c = is_ret_probe(tu) ? 'r' : 'p';
 751	int i;
 752
 753	seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 754			trace_probe_name(&tu->tp), tu->filename,
 755			(int)(sizeof(void *) * 2), tu->offset);
 756
 757	if (tu->ref_ctr_offset)
 758		seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 759
 760	for (i = 0; i < tu->tp.nr_args; i++)
 761		seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 762
 763	seq_putc(m, '\n');
 764	return 0;
 765}
 766
 767static int probes_seq_show(struct seq_file *m, void *v)
 768{
 769	struct dyn_event *ev = v;
 770
 771	if (!is_trace_uprobe(ev))
 772		return 0;
 773
 774	return trace_uprobe_show(m, ev);
 775}
 776
 777static const struct seq_operations probes_seq_op = {
 778	.start  = dyn_event_seq_start,
 779	.next   = dyn_event_seq_next,
 780	.stop   = dyn_event_seq_stop,
 781	.show   = probes_seq_show
 782};
 783
 784static int probes_open(struct inode *inode, struct file *file)
 785{
 786	int ret;
 787
 788	ret = security_locked_down(LOCKDOWN_TRACEFS);
 789	if (ret)
 790		return ret;
 791
 792	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 793		ret = dyn_events_release_all(&trace_uprobe_ops);
 794		if (ret)
 795			return ret;
 796	}
 797
 798	return seq_open(file, &probes_seq_op);
 799}
 800
 801static ssize_t probes_write(struct file *file, const char __user *buffer,
 802			    size_t count, loff_t *ppos)
 803{
 804	return trace_parse_run_command(file, buffer, count, ppos,
 805					create_or_delete_trace_uprobe);
 806}
 807
 808static const struct file_operations uprobe_events_ops = {
 809	.owner		= THIS_MODULE,
 810	.open		= probes_open,
 811	.read		= seq_read,
 812	.llseek		= seq_lseek,
 813	.release	= seq_release,
 814	.write		= probes_write,
 815};
 816
 817/* Probes profiling interfaces */
 818static int probes_profile_seq_show(struct seq_file *m, void *v)
 819{
 820	struct dyn_event *ev = v;
 821	struct trace_uprobe *tu;
 822
 823	if (!is_trace_uprobe(ev))
 824		return 0;
 825
 826	tu = to_trace_uprobe(ev);
 827	seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 828			trace_probe_name(&tu->tp), tu->nhit);
 829	return 0;
 830}
 831
 832static const struct seq_operations profile_seq_op = {
 833	.start  = dyn_event_seq_start,
 834	.next   = dyn_event_seq_next,
 835	.stop   = dyn_event_seq_stop,
 836	.show	= probes_profile_seq_show
 837};
 838
 839static int profile_open(struct inode *inode, struct file *file)
 840{
 841	int ret;
 842
 843	ret = security_locked_down(LOCKDOWN_TRACEFS);
 844	if (ret)
 845		return ret;
 846
 847	return seq_open(file, &profile_seq_op);
 848}
 849
 850static const struct file_operations uprobe_profile_ops = {
 851	.owner		= THIS_MODULE,
 852	.open		= profile_open,
 853	.read		= seq_read,
 854	.llseek		= seq_lseek,
 855	.release	= seq_release,
 856};
 857
 858struct uprobe_cpu_buffer {
 859	struct mutex mutex;
 860	void *buf;
 861};
 862static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 863static int uprobe_buffer_refcnt;
 864
 865static int uprobe_buffer_init(void)
 866{
 867	int cpu, err_cpu;
 868
 869	uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 870	if (uprobe_cpu_buffer == NULL)
 871		return -ENOMEM;
 872
 873	for_each_possible_cpu(cpu) {
 874		struct page *p = alloc_pages_node(cpu_to_node(cpu),
 875						  GFP_KERNEL, 0);
 876		if (p == NULL) {
 877			err_cpu = cpu;
 878			goto err;
 879		}
 880		per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 881		mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 882	}
 883
 884	return 0;
 885
 886err:
 887	for_each_possible_cpu(cpu) {
 888		if (cpu == err_cpu)
 889			break;
 890		free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 891	}
 892
 893	free_percpu(uprobe_cpu_buffer);
 894	return -ENOMEM;
 895}
 896
 897static int uprobe_buffer_enable(void)
 898{
 899	int ret = 0;
 900
 901	BUG_ON(!mutex_is_locked(&event_mutex));
 902
 903	if (uprobe_buffer_refcnt++ == 0) {
 904		ret = uprobe_buffer_init();
 905		if (ret < 0)
 906			uprobe_buffer_refcnt--;
 907	}
 908
 909	return ret;
 910}
 911
 912static void uprobe_buffer_disable(void)
 913{
 914	int cpu;
 915
 916	BUG_ON(!mutex_is_locked(&event_mutex));
 917
 918	if (--uprobe_buffer_refcnt == 0) {
 919		for_each_possible_cpu(cpu)
 920			free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 921							     cpu)->buf);
 922
 923		free_percpu(uprobe_cpu_buffer);
 924		uprobe_cpu_buffer = NULL;
 925	}
 926}
 927
 928static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 929{
 930	struct uprobe_cpu_buffer *ucb;
 931	int cpu;
 932
 933	cpu = raw_smp_processor_id();
 934	ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 935
 936	/*
 937	 * Use per-cpu buffers for fastest access, but we might migrate
 938	 * so the mutex makes sure we have sole access to it.
 939	 */
 940	mutex_lock(&ucb->mutex);
 941
 942	return ucb;
 943}
 944
 945static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 946{
 947	mutex_unlock(&ucb->mutex);
 948}
 949
 950static void __uprobe_trace_func(struct trace_uprobe *tu,
 951				unsigned long func, struct pt_regs *regs,
 952				struct uprobe_cpu_buffer *ucb, int dsize,
 953				struct trace_event_file *trace_file)
 954{
 955	struct uprobe_trace_entry_head *entry;
 956	struct trace_event_buffer fbuffer;
 957	void *data;
 958	int size, esize;
 959	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 960
 961	WARN_ON(call != trace_file->event_call);
 962
 963	if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 964		return;
 965
 966	if (trace_trigger_soft_disabled(trace_file))
 967		return;
 968
 969	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 970	size = esize + tu->tp.size + dsize;
 971	entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 972	if (!entry)
 973		return;
 974
 975	if (is_ret_probe(tu)) {
 976		entry->vaddr[0] = func;
 977		entry->vaddr[1] = instruction_pointer(regs);
 978		data = DATAOF_TRACE_ENTRY(entry, true);
 979	} else {
 980		entry->vaddr[0] = instruction_pointer(regs);
 981		data = DATAOF_TRACE_ENTRY(entry, false);
 982	}
 983
 984	memcpy(data, ucb->buf, tu->tp.size + dsize);
 985
 986	trace_event_buffer_commit(&fbuffer);
 987}
 988
 989/* uprobe handler */
 990static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 991			     struct uprobe_cpu_buffer *ucb, int dsize)
 992{
 993	struct event_file_link *link;
 994
 995	if (is_ret_probe(tu))
 996		return 0;
 997
 998	rcu_read_lock();
 999	trace_probe_for_each_link_rcu(link, &tu->tp)
1000		__uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1001	rcu_read_unlock();
1002
1003	return 0;
1004}
1005
1006static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1007				 struct pt_regs *regs,
1008				 struct uprobe_cpu_buffer *ucb, int dsize)
1009{
1010	struct event_file_link *link;
1011
1012	rcu_read_lock();
1013	trace_probe_for_each_link_rcu(link, &tu->tp)
1014		__uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1015	rcu_read_unlock();
1016}
1017
1018/* Event entry printers */
1019static enum print_line_t
1020print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1021{
1022	struct uprobe_trace_entry_head *entry;
1023	struct trace_seq *s = &iter->seq;
1024	struct trace_uprobe *tu;
1025	u8 *data;
1026
1027	entry = (struct uprobe_trace_entry_head *)iter->ent;
1028	tu = trace_uprobe_primary_from_call(
1029		container_of(event, struct trace_event_call, event));
1030	if (unlikely(!tu))
1031		goto out;
1032
1033	if (is_ret_probe(tu)) {
1034		trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1035				 trace_probe_name(&tu->tp),
1036				 entry->vaddr[1], entry->vaddr[0]);
1037		data = DATAOF_TRACE_ENTRY(entry, true);
1038	} else {
1039		trace_seq_printf(s, "%s: (0x%lx)",
1040				 trace_probe_name(&tu->tp),
1041				 entry->vaddr[0]);
1042		data = DATAOF_TRACE_ENTRY(entry, false);
1043	}
1044
1045	if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1046		goto out;
1047
1048	trace_seq_putc(s, '\n');
1049
1050 out:
1051	return trace_handle_return(s);
1052}
1053
1054typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1055				enum uprobe_filter_ctx ctx,
1056				struct mm_struct *mm);
1057
1058static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1059{
1060	int ret;
1061
1062	tu->consumer.filter = filter;
1063	tu->inode = d_real_inode(tu->path.dentry);
1064
1065	if (tu->ref_ctr_offset)
1066		ret = uprobe_register_refctr(tu->inode, tu->offset,
1067				tu->ref_ctr_offset, &tu->consumer);
1068	else
1069		ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1070
1071	if (ret)
1072		tu->inode = NULL;
1073
1074	return ret;
1075}
1076
1077static void __probe_event_disable(struct trace_probe *tp)
1078{
1079	struct trace_uprobe *tu;
1080
1081	tu = container_of(tp, struct trace_uprobe, tp);
1082	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1083
1084	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1085		if (!tu->inode)
1086			continue;
1087
1088		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1089		tu->inode = NULL;
1090	}
1091}
1092
1093static int probe_event_enable(struct trace_event_call *call,
1094			struct trace_event_file *file, filter_func_t filter)
1095{
1096	struct trace_probe *tp;
1097	struct trace_uprobe *tu;
1098	bool enabled;
1099	int ret;
1100
1101	tp = trace_probe_primary_from_call(call);
1102	if (WARN_ON_ONCE(!tp))
1103		return -ENODEV;
1104	enabled = trace_probe_is_enabled(tp);
1105
1106	/* This may also change "enabled" state */
1107	if (file) {
1108		if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1109			return -EINTR;
1110
1111		ret = trace_probe_add_file(tp, file);
1112		if (ret < 0)
1113			return ret;
1114	} else {
1115		if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1116			return -EINTR;
1117
1118		trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1119	}
1120
1121	tu = container_of(tp, struct trace_uprobe, tp);
1122	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1123
1124	if (enabled)
1125		return 0;
1126
1127	ret = uprobe_buffer_enable();
1128	if (ret)
1129		goto err_flags;
1130
1131	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1132		ret = trace_uprobe_enable(tu, filter);
1133		if (ret) {
1134			__probe_event_disable(tp);
1135			goto err_buffer;
1136		}
1137	}
1138
1139	return 0;
1140
1141 err_buffer:
1142	uprobe_buffer_disable();
1143
1144 err_flags:
1145	if (file)
1146		trace_probe_remove_file(tp, file);
1147	else
1148		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1149
1150	return ret;
1151}
1152
1153static void probe_event_disable(struct trace_event_call *call,
1154				struct trace_event_file *file)
1155{
1156	struct trace_probe *tp;
1157
1158	tp = trace_probe_primary_from_call(call);
1159	if (WARN_ON_ONCE(!tp))
1160		return;
1161
1162	if (!trace_probe_is_enabled(tp))
1163		return;
1164
1165	if (file) {
1166		if (trace_probe_remove_file(tp, file) < 0)
1167			return;
1168
1169		if (trace_probe_is_enabled(tp))
1170			return;
1171	} else
1172		trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1173
1174	__probe_event_disable(tp);
1175	uprobe_buffer_disable();
1176}
1177
1178static int uprobe_event_define_fields(struct trace_event_call *event_call)
1179{
1180	int ret, size;
1181	struct uprobe_trace_entry_head field;
1182	struct trace_uprobe *tu;
1183
1184	tu = trace_uprobe_primary_from_call(event_call);
1185	if (unlikely(!tu))
1186		return -ENODEV;
1187
1188	if (is_ret_probe(tu)) {
1189		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1190		DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1191		size = SIZEOF_TRACE_ENTRY(true);
1192	} else {
1193		DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1194		size = SIZEOF_TRACE_ENTRY(false);
1195	}
1196
1197	return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1198}
1199
1200#ifdef CONFIG_PERF_EVENTS
1201static bool
1202__uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1203{
1204	struct perf_event *event;
1205
1206	if (filter->nr_systemwide)
1207		return true;
1208
1209	list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1210		if (event->hw.target->mm == mm)
1211			return true;
1212	}
1213
1214	return false;
1215}
1216
1217static inline bool
1218trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1219			  struct perf_event *event)
1220{
1221	return __uprobe_perf_filter(filter, event->hw.target->mm);
1222}
1223
1224static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1225				       struct perf_event *event)
1226{
1227	bool done;
1228
1229	write_lock(&filter->rwlock);
1230	if (event->hw.target) {
1231		list_del(&event->hw.tp_list);
1232		done = filter->nr_systemwide ||
1233			(event->hw.target->flags & PF_EXITING) ||
1234			trace_uprobe_filter_event(filter, event);
1235	} else {
1236		filter->nr_systemwide--;
1237		done = filter->nr_systemwide;
1238	}
1239	write_unlock(&filter->rwlock);
1240
1241	return done;
1242}
1243
1244/* This returns true if the filter always covers target mm */
1245static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1246				    struct perf_event *event)
1247{
1248	bool done;
1249
1250	write_lock(&filter->rwlock);
1251	if (event->hw.target) {
1252		/*
1253		 * event->parent != NULL means copy_process(), we can avoid
1254		 * uprobe_apply(). current->mm must be probed and we can rely
1255		 * on dup_mmap() which preserves the already installed bp's.
1256		 *
1257		 * attr.enable_on_exec means that exec/mmap will install the
1258		 * breakpoints we need.
1259		 */
1260		done = filter->nr_systemwide ||
1261			event->parent || event->attr.enable_on_exec ||
1262			trace_uprobe_filter_event(filter, event);
1263		list_add(&event->hw.tp_list, &filter->perf_events);
1264	} else {
1265		done = filter->nr_systemwide;
1266		filter->nr_systemwide++;
1267	}
1268	write_unlock(&filter->rwlock);
1269
1270	return done;
1271}
1272
1273static int uprobe_perf_close(struct trace_event_call *call,
1274			     struct perf_event *event)
1275{
1276	struct trace_probe *tp;
1277	struct trace_uprobe *tu;
1278	int ret = 0;
1279
1280	tp = trace_probe_primary_from_call(call);
1281	if (WARN_ON_ONCE(!tp))
1282		return -ENODEV;
1283
1284	tu = container_of(tp, struct trace_uprobe, tp);
1285	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1286		return 0;
1287
1288	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1289		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1290		if (ret)
1291			break;
1292	}
1293
1294	return ret;
1295}
1296
1297static int uprobe_perf_open(struct trace_event_call *call,
1298			    struct perf_event *event)
1299{
1300	struct trace_probe *tp;
1301	struct trace_uprobe *tu;
1302	int err = 0;
1303
1304	tp = trace_probe_primary_from_call(call);
1305	if (WARN_ON_ONCE(!tp))
1306		return -ENODEV;
1307
1308	tu = container_of(tp, struct trace_uprobe, tp);
1309	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1310		return 0;
1311
1312	list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1313		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1314		if (err) {
1315			uprobe_perf_close(call, event);
1316			break;
1317		}
1318	}
1319
1320	return err;
1321}
1322
1323static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1324				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1325{
1326	struct trace_uprobe_filter *filter;
1327	struct trace_uprobe *tu;
1328	int ret;
1329
1330	tu = container_of(uc, struct trace_uprobe, consumer);
1331	filter = tu->tp.event->filter;
1332
1333	read_lock(&filter->rwlock);
1334	ret = __uprobe_perf_filter(filter, mm);
1335	read_unlock(&filter->rwlock);
1336
1337	return ret;
1338}
1339
1340static void __uprobe_perf_func(struct trace_uprobe *tu,
1341			       unsigned long func, struct pt_regs *regs,
1342			       struct uprobe_cpu_buffer *ucb, int dsize)
1343{
1344	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1345	struct uprobe_trace_entry_head *entry;
1346	struct hlist_head *head;
1347	void *data;
1348	int size, esize;
1349	int rctx;
1350
1351#ifdef CONFIG_BPF_EVENTS
1352	if (bpf_prog_array_valid(call)) {
1353		u32 ret;
1354
1355		ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
1356		if (!ret)
1357			return;
1358	}
1359#endif /* CONFIG_BPF_EVENTS */
1360
1361	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1362
1363	size = esize + tu->tp.size + dsize;
1364	size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1365	if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1366		return;
1367
1368	preempt_disable();
1369	head = this_cpu_ptr(call->perf_events);
1370	if (hlist_empty(head))
1371		goto out;
1372
1373	entry = perf_trace_buf_alloc(size, NULL, &rctx);
1374	if (!entry)
1375		goto out;
1376
1377	if (is_ret_probe(tu)) {
1378		entry->vaddr[0] = func;
1379		entry->vaddr[1] = instruction_pointer(regs);
1380		data = DATAOF_TRACE_ENTRY(entry, true);
1381	} else {
1382		entry->vaddr[0] = instruction_pointer(regs);
1383		data = DATAOF_TRACE_ENTRY(entry, false);
1384	}
1385
1386	memcpy(data, ucb->buf, tu->tp.size + dsize);
1387
1388	if (size - esize > tu->tp.size + dsize) {
1389		int len = tu->tp.size + dsize;
1390
1391		memset(data + len, 0, size - esize - len);
1392	}
1393
1394	perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1395			      head, NULL);
1396 out:
1397	preempt_enable();
1398}
1399
1400/* uprobe profile handler */
1401static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1402			    struct uprobe_cpu_buffer *ucb, int dsize)
1403{
1404	if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1405		return UPROBE_HANDLER_REMOVE;
1406
1407	if (!is_ret_probe(tu))
1408		__uprobe_perf_func(tu, 0, regs, ucb, dsize);
1409	return 0;
1410}
1411
1412static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1413				struct pt_regs *regs,
1414				struct uprobe_cpu_buffer *ucb, int dsize)
1415{
1416	__uprobe_perf_func(tu, func, regs, ucb, dsize);
1417}
1418
1419int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1420			const char **filename, u64 *probe_offset,
1421			bool perf_type_tracepoint)
1422{
1423	const char *pevent = trace_event_name(event->tp_event);
1424	const char *group = event->tp_event->class->system;
1425	struct trace_uprobe *tu;
1426
1427	if (perf_type_tracepoint)
1428		tu = find_probe_event(pevent, group);
1429	else
1430		tu = trace_uprobe_primary_from_call(event->tp_event);
1431	if (!tu)
1432		return -EINVAL;
1433
1434	*fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1435				    : BPF_FD_TYPE_UPROBE;
1436	*filename = tu->filename;
1437	*probe_offset = tu->offset;
 
1438	return 0;
1439}
1440#endif	/* CONFIG_PERF_EVENTS */
1441
1442static int
1443trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1444		      void *data)
1445{
1446	struct trace_event_file *file = data;
1447
1448	switch (type) {
1449	case TRACE_REG_REGISTER:
1450		return probe_event_enable(event, file, NULL);
1451
1452	case TRACE_REG_UNREGISTER:
1453		probe_event_disable(event, file);
1454		return 0;
1455
1456#ifdef CONFIG_PERF_EVENTS
1457	case TRACE_REG_PERF_REGISTER:
1458		return probe_event_enable(event, NULL, uprobe_perf_filter);
1459
1460	case TRACE_REG_PERF_UNREGISTER:
1461		probe_event_disable(event, NULL);
1462		return 0;
1463
1464	case TRACE_REG_PERF_OPEN:
1465		return uprobe_perf_open(event, data);
1466
1467	case TRACE_REG_PERF_CLOSE:
1468		return uprobe_perf_close(event, data);
1469
1470#endif
1471	default:
1472		return 0;
1473	}
1474}
1475
1476static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1477{
1478	struct trace_uprobe *tu;
1479	struct uprobe_dispatch_data udd;
1480	struct uprobe_cpu_buffer *ucb;
1481	int dsize, esize;
1482	int ret = 0;
1483
1484
1485	tu = container_of(con, struct trace_uprobe, consumer);
1486	tu->nhit++;
1487
1488	udd.tu = tu;
1489	udd.bp_addr = instruction_pointer(regs);
1490
1491	current->utask->vaddr = (unsigned long) &udd;
1492
1493	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1494		return 0;
1495
1496	dsize = __get_data_size(&tu->tp, regs);
1497	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1498
1499	ucb = uprobe_buffer_get();
1500	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1501
1502	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1503		ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1504
1505#ifdef CONFIG_PERF_EVENTS
1506	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1507		ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1508#endif
1509	uprobe_buffer_put(ucb);
1510	return ret;
1511}
1512
1513static int uretprobe_dispatcher(struct uprobe_consumer *con,
1514				unsigned long func, struct pt_regs *regs)
1515{
1516	struct trace_uprobe *tu;
1517	struct uprobe_dispatch_data udd;
1518	struct uprobe_cpu_buffer *ucb;
1519	int dsize, esize;
1520
1521	tu = container_of(con, struct trace_uprobe, consumer);
1522
1523	udd.tu = tu;
1524	udd.bp_addr = func;
1525
1526	current->utask->vaddr = (unsigned long) &udd;
1527
1528	if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1529		return 0;
1530
1531	dsize = __get_data_size(&tu->tp, regs);
1532	esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1533
1534	ucb = uprobe_buffer_get();
1535	store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1536
1537	if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1538		uretprobe_trace_func(tu, func, regs, ucb, dsize);
1539
1540#ifdef CONFIG_PERF_EVENTS
1541	if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1542		uretprobe_perf_func(tu, func, regs, ucb, dsize);
1543#endif
1544	uprobe_buffer_put(ucb);
1545	return 0;
1546}
1547
1548static struct trace_event_functions uprobe_funcs = {
1549	.trace		= print_uprobe_event
1550};
1551
1552static struct trace_event_fields uprobe_fields_array[] = {
1553	{ .type = TRACE_FUNCTION_TYPE,
1554	  .define_fields = uprobe_event_define_fields },
1555	{}
1556};
1557
1558static inline void init_trace_event_call(struct trace_uprobe *tu)
1559{
1560	struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1561	call->event.funcs = &uprobe_funcs;
1562	call->class->fields_array = uprobe_fields_array;
1563
1564	call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1565	call->class->reg = trace_uprobe_register;
1566}
1567
1568static int register_uprobe_event(struct trace_uprobe *tu)
1569{
1570	init_trace_event_call(tu);
1571
1572	return trace_probe_register_event_call(&tu->tp);
1573}
1574
1575static int unregister_uprobe_event(struct trace_uprobe *tu)
1576{
1577	return trace_probe_unregister_event_call(&tu->tp);
1578}
1579
1580#ifdef CONFIG_PERF_EVENTS
1581struct trace_event_call *
1582create_local_trace_uprobe(char *name, unsigned long offs,
1583			  unsigned long ref_ctr_offset, bool is_return)
1584{
1585	enum probe_print_type ptype;
1586	struct trace_uprobe *tu;
1587	struct path path;
1588	int ret;
1589
1590	ret = kern_path(name, LOOKUP_FOLLOW, &path);
1591	if (ret)
1592		return ERR_PTR(ret);
1593
1594	if (!d_is_reg(path.dentry)) {
1595		path_put(&path);
1596		return ERR_PTR(-EINVAL);
1597	}
1598
1599	/*
1600	 * local trace_kprobes are not added to dyn_event, so they are never
1601	 * searched in find_trace_kprobe(). Therefore, there is no concern of
1602	 * duplicated name "DUMMY_EVENT" here.
1603	 */
1604	tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1605				is_return);
1606
1607	if (IS_ERR(tu)) {
1608		pr_info("Failed to allocate trace_uprobe.(%d)\n",
1609			(int)PTR_ERR(tu));
1610		path_put(&path);
1611		return ERR_CAST(tu);
1612	}
1613
1614	tu->offset = offs;
1615	tu->path = path;
1616	tu->ref_ctr_offset = ref_ctr_offset;
1617	tu->filename = kstrdup(name, GFP_KERNEL);
1618	if (!tu->filename) {
1619		ret = -ENOMEM;
1620		goto error;
1621	}
1622
1623	init_trace_event_call(tu);
1624
1625	ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1626	if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1627		ret = -ENOMEM;
1628		goto error;
1629	}
1630
1631	return trace_probe_event_call(&tu->tp);
1632error:
1633	free_trace_uprobe(tu);
1634	return ERR_PTR(ret);
1635}
1636
1637void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1638{
1639	struct trace_uprobe *tu;
1640
1641	tu = trace_uprobe_primary_from_call(event_call);
1642
1643	free_trace_uprobe(tu);
1644}
1645#endif /* CONFIG_PERF_EVENTS */
1646
1647/* Make a trace interface for controlling probe points */
1648static __init int init_uprobe_trace(void)
1649{
1650	int ret;
1651
1652	ret = dyn_event_register(&trace_uprobe_ops);
1653	if (ret)
1654		return ret;
1655
1656	ret = tracing_init_dentry();
1657	if (ret)
1658		return 0;
1659
1660	trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1661				    NULL, &uprobe_events_ops);
1662	/* Profile interface */
1663	trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1664				    NULL, &uprobe_profile_ops);
1665	return 0;
1666}
1667
1668fs_initcall(init_uprobe_trace);