trace_stack.c 10.8 KB
Newer Older
S
Steven Rostedt 已提交
1 2 3 4
/*
 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
 *
 */
5
#include <linux/sched/task_stack.h>
S
Steven Rostedt 已提交
6 7 8 9 10 11 12
#include <linux/stacktrace.h>
#include <linux/kallsyms.h>
#include <linux/seq_file.h>
#include <linux/spinlock.h>
#include <linux/uaccess.h>
#include <linux/ftrace.h>
#include <linux/module.h>
13
#include <linux/sysctl.h>
S
Steven Rostedt 已提交
14
#include <linux/init.h>
15 16 17

#include <asm/setup.h>

S
Steven Rostedt 已提交
18 19
#include "trace.h"

S
Steven Rostedt 已提交
20 21
static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
	 { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
22
unsigned stack_trace_index[STACK_TRACE_ENTRIES];
S
Steven Rostedt 已提交
23

24 25 26 27 28
/*
 * Reserve one entry for the passed in ip. This will allow
 * us to remove most or all of the stack size overhead
 * added by the stack tracer itself.
 */
29
struct stack_trace stack_trace_max = {
30
	.max_entries		= STACK_TRACE_ENTRIES - 1,
31
	.entries		= &stack_dump_trace[0],
S
Steven Rostedt 已提交
32 33
};

34
unsigned long stack_trace_max_size;
35
arch_spinlock_t stack_trace_max_lock =
36
	(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
S
Steven Rostedt 已提交
37

38
DEFINE_PER_CPU(int, disable_stack_tracer);
39 40 41 42
static DEFINE_MUTEX(stack_sysctl_mutex);

int stack_tracer_enabled;
static int last_stack_tracer_enabled;
S
Steven Rostedt 已提交
43

44
void stack_trace_print(void)
45 46 47 48 49 50
{
	long i;
	int size;

	pr_emerg("        Depth    Size   Location    (%d entries)\n"
			   "        -----    ----   --------\n",
51
			   stack_trace_max.nr_entries);
52

53
	for (i = 0; i < stack_trace_max.nr_entries; i++) {
54 55
		if (stack_dump_trace[i] == ULONG_MAX)
			break;
56
		if (i+1 == stack_trace_max.nr_entries ||
57
				stack_dump_trace[i+1] == ULONG_MAX)
58
			size = stack_trace_index[i];
59
		else
60
			size = stack_trace_index[i] - stack_trace_index[i+1];
61

62
		pr_emerg("%3ld) %8d   %5d   %pS\n", i, stack_trace_index[i],
63 64 65 66
				size, (void *)stack_dump_trace[i]);
	}
}

67
/*
68
 * When arch-specific code overrides this function, the following
69
 * data should be filled up, assuming stack_trace_max_lock is held to
70 71 72 73 74 75
 * prevent concurrent updates.
 *     stack_trace_index[]
 *     stack_trace_max
 *     stack_trace_max_size
 */
void __weak
76
check_stack(unsigned long ip, unsigned long *stack)
S
Steven Rostedt 已提交
77
{
78
	unsigned long this_size, flags; unsigned long *p, *top, *start;
79 80
	static int tracer_frame;
	int frame_size = ACCESS_ONCE(tracer_frame);
81
	int i, x;
S
Steven Rostedt 已提交
82

83
	this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
S
Steven Rostedt 已提交
84
	this_size = THREAD_SIZE - this_size;
85 86
	/* Remove the frame of the tracer */
	this_size -= frame_size;
S
Steven Rostedt 已提交
87

88
	if (this_size <= stack_trace_max_size)
S
Steven Rostedt 已提交
89 90
		return;

91
	/* we do not handle interrupt stacks yet */
92
	if (!object_is_on_stack(stack))
93 94
		return;

95 96 97 98
	/* Can't do this from NMI context (can cause deadlocks) */
	if (in_nmi())
		return;

99
	local_irq_save(flags);
100
	arch_spin_lock(&stack_trace_max_lock);
S
Steven Rostedt 已提交
101

102 103 104 105
	/* In case another CPU set the tracer_frame on us */
	if (unlikely(!frame_size))
		this_size -= tracer_frame;

S
Steven Rostedt 已提交
106
	/* a race could have already updated it */
107
	if (this_size <= stack_trace_max_size)
S
Steven Rostedt 已提交
108 109
		goto out;

110
	stack_trace_max_size = this_size;
S
Steven Rostedt 已提交
111

112 113
	stack_trace_max.nr_entries = 0;
	stack_trace_max.skip = 3;
S
Steven Rostedt 已提交
114

115
	save_stack_trace(&stack_trace_max);
S
Steven Rostedt 已提交
116

117
	/* Skip over the overhead of the stack tracer itself */
118
	for (i = 0; i < stack_trace_max.nr_entries; i++) {
119 120 121
		if (stack_dump_trace[i] == ip)
			break;
	}
122

123 124 125 126 127 128 129
	/*
	 * Some archs may not have the passed in ip in the dump.
	 * If that happens, we need to show everything.
	 */
	if (i == stack_trace_max.nr_entries)
		i = 0;

S
Steven Rostedt 已提交
130 131 132
	/*
	 * Now find where in the stack these are.
	 */
133
	x = 0;
134
	start = stack;
S
Steven Rostedt 已提交
135 136 137 138 139 140 141 142 143 144
	top = (unsigned long *)
		(((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);

	/*
	 * Loop through all the entries. One of the entries may
	 * for some reason be missed on the stack, so we may
	 * have to account for them. If they are all there, this
	 * loop will only happen once. This code only takes place
	 * on a new max, so it is far from a fast path.
	 */
145
	while (i < stack_trace_max.nr_entries) {
S
Steven Rostedt 已提交
146
		int found = 0;
S
Steven Rostedt 已提交
147

148
		stack_trace_index[x] = this_size;
S
Steven Rostedt 已提交
149 150
		p = start;

151
		for (; p < top && i < stack_trace_max.nr_entries; p++) {
152 153
			if (stack_dump_trace[i] == ULONG_MAX)
				break;
154 155 156 157 158
			/*
			 * The READ_ONCE_NOCHECK is used to let KASAN know that
			 * this is not a stack-out-of-bounds error.
			 */
			if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) {
159
				stack_dump_trace[x] = stack_dump_trace[i++];
160
				this_size = stack_trace_index[x++] =
S
Steven Rostedt 已提交
161
					(top - p) * sizeof(unsigned long);
S
Steven Rostedt 已提交
162
				found = 1;
S
Steven Rostedt 已提交
163 164
				/* Start the search from here */
				start = p + 1;
165 166 167 168 169 170 171
				/*
				 * We do not want to show the overhead
				 * of the stack tracer stack in the
				 * max stack. If we haven't figured
				 * out what that is, then figure it out
				 * now.
				 */
172
				if (unlikely(!tracer_frame)) {
173 174
					tracer_frame = (p - stack) *
						sizeof(unsigned long);
175
					stack_trace_max_size -= tracer_frame;
176
				}
S
Steven Rostedt 已提交
177 178 179
			}
		}

S
Steven Rostedt 已提交
180 181
		if (!found)
			i++;
S
Steven Rostedt 已提交
182 183
	}

184
	stack_trace_max.nr_entries = x;
185 186 187
	for (; x < i; x++)
		stack_dump_trace[x] = ULONG_MAX;

188
	if (task_stack_end_corrupted(current)) {
189
		stack_trace_print();
190 191 192
		BUG();
	}

S
Steven Rostedt 已提交
193
 out:
194
	arch_spin_unlock(&stack_trace_max_lock);
195
	local_irq_restore(flags);
S
Steven Rostedt 已提交
196 197 198
}

static void
199 200
stack_trace_call(unsigned long ip, unsigned long parent_ip,
		 struct ftrace_ops *op, struct pt_regs *pt_regs)
S
Steven Rostedt 已提交
201
{
202
	unsigned long stack;
S
Steven Rostedt 已提交
203

204
	preempt_disable_notrace();
S
Steven Rostedt 已提交
205 206

	/* no atomic needed, we only modify this variable by this cpu */
207 208
	__this_cpu_inc(disable_stack_tracer);
	if (__this_cpu_read(disable_stack_tracer) != 1)
S
Steven Rostedt 已提交
209 210
		goto out;

211
	ip += MCOUNT_INSN_SIZE;
212 213

	check_stack(ip, &stack);
S
Steven Rostedt 已提交
214 215

 out:
216
	__this_cpu_dec(disable_stack_tracer);
S
Steven Rostedt 已提交
217
	/* prevent recursion in schedule */
218
	preempt_enable_notrace();
S
Steven Rostedt 已提交
219 220 221 222 223
}

static struct ftrace_ops trace_ops __read_mostly =
{
	.func = stack_trace_call,
224
	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
S
Steven Rostedt 已提交
225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248
};

static ssize_t
stack_max_size_read(struct file *filp, char __user *ubuf,
		    size_t count, loff_t *ppos)
{
	unsigned long *ptr = filp->private_data;
	char buf[64];
	int r;

	r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
	if (r > sizeof(buf))
		r = sizeof(buf);
	return simple_read_from_buffer(ubuf, count, ppos, buf, r);
}

static ssize_t
stack_max_size_write(struct file *filp, const char __user *ubuf,
		     size_t count, loff_t *ppos)
{
	long *ptr = filp->private_data;
	unsigned long val, flags;
	int ret;

249 250
	ret = kstrtoul_from_user(ubuf, count, 10, &val);
	if (ret)
S
Steven Rostedt 已提交
251 252
		return ret;

253
	local_irq_save(flags);
254 255 256 257

	/*
	 * In case we trace inside arch_spin_lock() or after (NMI),
	 * we will cause circular lock, so we also need to increase
258
	 * the percpu disable_stack_tracer here.
259
	 */
260
	__this_cpu_inc(disable_stack_tracer);
261

262
	arch_spin_lock(&stack_trace_max_lock);
S
Steven Rostedt 已提交
263
	*ptr = val;
264
	arch_spin_unlock(&stack_trace_max_lock);
265

266
	__this_cpu_dec(disable_stack_tracer);
267
	local_irq_restore(flags);
S
Steven Rostedt 已提交
268 269 270 271

	return count;
}

272
static const struct file_operations stack_max_size_fops = {
S
Steven Rostedt 已提交
273 274 275
	.open		= tracing_open_generic,
	.read		= stack_max_size_read,
	.write		= stack_max_size_write,
276
	.llseek		= default_llseek,
S
Steven Rostedt 已提交
277 278 279
};

static void *
L
Li Zefan 已提交
280
__next(struct seq_file *m, loff_t *pos)
S
Steven Rostedt 已提交
281
{
L
Li Zefan 已提交
282
	long n = *pos - 1;
S
Steven Rostedt 已提交
283

284
	if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
S
Steven Rostedt 已提交
285 286
		return NULL;

L
Li Zefan 已提交
287
	m->private = (void *)n;
S
Steven Rostedt 已提交
288
	return &m->private;
S
Steven Rostedt 已提交
289 290
}

L
Li Zefan 已提交
291 292
static void *
t_next(struct seq_file *m, void *v, loff_t *pos)
S
Steven Rostedt 已提交
293
{
L
Li Zefan 已提交
294 295 296
	(*pos)++;
	return __next(m, pos);
}
S
Steven Rostedt 已提交
297

L
Li Zefan 已提交
298 299
static void *t_start(struct seq_file *m, loff_t *pos)
{
S
Steven Rostedt 已提交
300
	local_irq_disable();
301

302
	__this_cpu_inc(disable_stack_tracer);
303

304
	arch_spin_lock(&stack_trace_max_lock);
S
Steven Rostedt 已提交
305

306 307 308
	if (*pos == 0)
		return SEQ_START_TOKEN;

L
Li Zefan 已提交
309
	return __next(m, pos);
S
Steven Rostedt 已提交
310 311 312 313
}

static void t_stop(struct seq_file *m, void *p)
{
314
	arch_spin_unlock(&stack_trace_max_lock);
315

316
	__this_cpu_dec(disable_stack_tracer);
317

S
Steven Rostedt 已提交
318 319 320
	local_irq_enable();
}

321
static void trace_lookup_stack(struct seq_file *m, long i)
S
Steven Rostedt 已提交
322
{
S
Steven Rostedt 已提交
323
	unsigned long addr = stack_dump_trace[i];
S
Steven Rostedt 已提交
324

325
	seq_printf(m, "%pS\n", (void *)addr);
S
Steven Rostedt 已提交
326 327
}

328 329 330 331 332 333 334 335 336 337 338
static void print_disabled(struct seq_file *m)
{
	seq_puts(m, "#\n"
		 "#  Stack tracer disabled\n"
		 "#\n"
		 "# To enable the stack tracer, either add 'stacktrace' to the\n"
		 "# kernel command line\n"
		 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
		 "#\n");
}

S
Steven Rostedt 已提交
339 340
static int t_show(struct seq_file *m, void *v)
{
341
	long i;
S
Steven Rostedt 已提交
342 343
	int size;

344
	if (v == SEQ_START_TOKEN) {
345
		seq_printf(m, "        Depth    Size   Location"
S
Steven Rostedt 已提交
346
			   "    (%d entries)\n"
347
			   "        -----    ----   --------\n",
348
			   stack_trace_max.nr_entries);
349

350
		if (!stack_tracer_enabled && !stack_trace_max_size)
351 352
			print_disabled(m);

S
Steven Rostedt 已提交
353 354
		return 0;
	}
S
Steven Rostedt 已提交
355

356 357
	i = *(long *)v;

358
	if (i >= stack_trace_max.nr_entries ||
S
Steven Rostedt 已提交
359
	    stack_dump_trace[i] == ULONG_MAX)
S
Steven Rostedt 已提交
360 361
		return 0;

362
	if (i+1 == stack_trace_max.nr_entries ||
S
Steven Rostedt 已提交
363
	    stack_dump_trace[i+1] == ULONG_MAX)
364
		size = stack_trace_index[i];
S
Steven Rostedt 已提交
365
	else
366
		size = stack_trace_index[i] - stack_trace_index[i+1];
S
Steven Rostedt 已提交
367

368
	seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
S
Steven Rostedt 已提交
369 370

	trace_lookup_stack(m, i);
S
Steven Rostedt 已提交
371 372 373 374

	return 0;
}

375
static const struct seq_operations stack_trace_seq_ops = {
S
Steven Rostedt 已提交
376 377 378 379 380 381 382 383
	.start		= t_start,
	.next		= t_next,
	.stop		= t_stop,
	.show		= t_show,
};

static int stack_trace_open(struct inode *inode, struct file *file)
{
L
Li Zefan 已提交
384
	return seq_open(file, &stack_trace_seq_ops);
S
Steven Rostedt 已提交
385 386
}

387
static const struct file_operations stack_trace_fops = {
S
Steven Rostedt 已提交
388 389 390
	.open		= stack_trace_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
L
Li Zefan 已提交
391
	.release	= seq_release,
S
Steven Rostedt 已提交
392 393
};

394 395
#ifdef CONFIG_DYNAMIC_FTRACE

396 397 398
static int
stack_trace_filter_open(struct inode *inode, struct file *file)
{
399 400 401
	struct ftrace_ops *ops = inode->i_private;

	return ftrace_regex_open(ops, FTRACE_ITER_FILTER,
402 403 404 405 406 407 408
				 inode, file);
}

static const struct file_operations stack_trace_filter_fops = {
	.open = stack_trace_filter_open,
	.read = seq_read,
	.write = ftrace_filter_write,
409
	.llseek = tracing_lseek,
410 411 412
	.release = ftrace_regex_release,
};

413 414
#endif /* CONFIG_DYNAMIC_FTRACE */

415 416
int
stack_trace_sysctl(struct ctl_table *table, int write,
417
		   void __user *buffer, size_t *lenp,
418 419 420 421 422 423
		   loff_t *ppos)
{
	int ret;

	mutex_lock(&stack_sysctl_mutex);

424
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
425 426

	if (ret || !write ||
427
	    (last_stack_tracer_enabled == !!stack_tracer_enabled))
428 429
		goto out;

430
	last_stack_tracer_enabled = !!stack_tracer_enabled;
431 432 433 434 435 436 437 438 439 440 441

	if (stack_tracer_enabled)
		register_ftrace_function(&trace_ops);
	else
		unregister_ftrace_function(&trace_ops);

 out:
	mutex_unlock(&stack_sysctl_mutex);
	return ret;
}

442 443
static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;

444 445
static __init int enable_stacktrace(char *str)
{
446 447 448
	if (strncmp(str, "_filter=", 8) == 0)
		strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);

449 450
	stack_tracer_enabled = 1;
	last_stack_tracer_enabled = 1;
451 452 453 454
	return 1;
}
__setup("stacktrace", enable_stacktrace);

S
Steven Rostedt 已提交
455 456 457 458 459
static __init int stack_trace_init(void)
{
	struct dentry *d_tracer;

	d_tracer = tracing_init_dentry();
460
	if (IS_ERR(d_tracer))
461
		return 0;
S
Steven Rostedt 已提交
462

463
	trace_create_file("stack_max_size", 0644, d_tracer,
464
			&stack_trace_max_size, &stack_max_size_fops);
S
Steven Rostedt 已提交
465

466 467
	trace_create_file("stack_trace", 0444, d_tracer,
			NULL, &stack_trace_fops);
S
Steven Rostedt 已提交
468

469
#ifdef CONFIG_DYNAMIC_FTRACE
470
	trace_create_file("stack_trace_filter", 0444, d_tracer,
471
			  &trace_ops, &stack_trace_filter_fops);
472
#endif
473

474 475 476
	if (stack_trace_filter_buf[0])
		ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);

477
	if (stack_tracer_enabled)
478
		register_ftrace_function(&trace_ops);
S
Steven Rostedt 已提交
479 480 481 482 483

	return 0;
}

device_initcall(stack_trace_init);