trace_stack.c 11.0 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;

99 100 101 102 103 104 105 106
	/*
	 * There's a slight chance that we are tracing inside the
	 * RCU infrastructure, and rcu_irq_enter() will not work
	 * as expected.
	 */
	if (unlikely(rcu_irq_enter_disabled()))
		return;

107
	local_irq_save(flags);
108
	arch_spin_lock(&stack_trace_max_lock);
S
Steven Rostedt 已提交
109

110 111 112 113 114 115
	/*
	 * RCU may not be watching, make it see us.
	 * The stack trace code uses rcu_sched.
	 */
	rcu_irq_enter();

116 117 118 119
	/* In case another CPU set the tracer_frame on us */
	if (unlikely(!frame_size))
		this_size -= tracer_frame;

S
Steven Rostedt 已提交
120
	/* a race could have already updated it */
121
	if (this_size <= stack_trace_max_size)
S
Steven Rostedt 已提交
122 123
		goto out;

124
	stack_trace_max_size = this_size;
S
Steven Rostedt 已提交
125

126 127
	stack_trace_max.nr_entries = 0;
	stack_trace_max.skip = 3;
S
Steven Rostedt 已提交
128

129
	save_stack_trace(&stack_trace_max);
S
Steven Rostedt 已提交
130

131
	/* Skip over the overhead of the stack tracer itself */
132
	for (i = 0; i < stack_trace_max.nr_entries; i++) {
133 134 135
		if (stack_dump_trace[i] == ip)
			break;
	}
136

137 138 139 140 141 142 143
	/*
	 * 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
144 145 146
	/*
	 * Now find where in the stack these are.
	 */
147
	x = 0;
148
	start = stack;
S
Steven Rostedt 已提交
149 150 151 152 153 154 155 156 157 158
	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.
	 */
159
	while (i < stack_trace_max.nr_entries) {
S
Steven Rostedt 已提交
160
		int found = 0;
S
Steven Rostedt 已提交
161

162
		stack_trace_index[x] = this_size;
S
Steven Rostedt 已提交
163 164
		p = start;

165
		for (; p < top && i < stack_trace_max.nr_entries; p++) {
166 167
			if (stack_dump_trace[i] == ULONG_MAX)
				break;
168 169 170 171 172
			/*
			 * 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]) {
173
				stack_dump_trace[x] = stack_dump_trace[i++];
174
				this_size = stack_trace_index[x++] =
S
Steven Rostedt 已提交
175
					(top - p) * sizeof(unsigned long);
S
Steven Rostedt 已提交
176
				found = 1;
S
Steven Rostedt 已提交
177 178
				/* Start the search from here */
				start = p + 1;
179 180 181 182 183 184 185
				/*
				 * 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.
				 */
186
				if (unlikely(!tracer_frame)) {
187 188
					tracer_frame = (p - stack) *
						sizeof(unsigned long);
189
					stack_trace_max_size -= tracer_frame;
190
				}
S
Steven Rostedt 已提交
191 192 193
			}
		}

S
Steven Rostedt 已提交
194 195
		if (!found)
			i++;
S
Steven Rostedt 已提交
196 197
	}

198
	stack_trace_max.nr_entries = x;
199 200 201
	for (; x < i; x++)
		stack_dump_trace[x] = ULONG_MAX;

202
	if (task_stack_end_corrupted(current)) {
203
		stack_trace_print();
204 205 206
		BUG();
	}

S
Steven Rostedt 已提交
207
 out:
208
	rcu_irq_exit();
209
	arch_spin_unlock(&stack_trace_max_lock);
210
	local_irq_restore(flags);
S
Steven Rostedt 已提交
211 212 213
}

static void
214 215
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
216
{
217
	unsigned long stack;
S
Steven Rostedt 已提交
218

219
	preempt_disable_notrace();
S
Steven Rostedt 已提交
220 221

	/* no atomic needed, we only modify this variable by this cpu */
222 223
	__this_cpu_inc(disable_stack_tracer);
	if (__this_cpu_read(disable_stack_tracer) != 1)
S
Steven Rostedt 已提交
224 225
		goto out;

226
	ip += MCOUNT_INSN_SIZE;
227 228

	check_stack(ip, &stack);
S
Steven Rostedt 已提交
229 230

 out:
231
	__this_cpu_dec(disable_stack_tracer);
S
Steven Rostedt 已提交
232
	/* prevent recursion in schedule */
233
	preempt_enable_notrace();
S
Steven Rostedt 已提交
234 235 236 237 238
}

static struct ftrace_ops trace_ops __read_mostly =
{
	.func = stack_trace_call,
239
	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
S
Steven Rostedt 已提交
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
};

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;

264 265
	ret = kstrtoul_from_user(ubuf, count, 10, &val);
	if (ret)
S
Steven Rostedt 已提交
266 267
		return ret;

268
	local_irq_save(flags);
269 270 271 272

	/*
	 * In case we trace inside arch_spin_lock() or after (NMI),
	 * we will cause circular lock, so we also need to increase
273
	 * the percpu disable_stack_tracer here.
274
	 */
275
	__this_cpu_inc(disable_stack_tracer);
276

277
	arch_spin_lock(&stack_trace_max_lock);
S
Steven Rostedt 已提交
278
	*ptr = val;
279
	arch_spin_unlock(&stack_trace_max_lock);
280

281
	__this_cpu_dec(disable_stack_tracer);
282
	local_irq_restore(flags);
S
Steven Rostedt 已提交
283 284 285 286

	return count;
}

287
static const struct file_operations stack_max_size_fops = {
S
Steven Rostedt 已提交
288 289 290
	.open		= tracing_open_generic,
	.read		= stack_max_size_read,
	.write		= stack_max_size_write,
291
	.llseek		= default_llseek,
S
Steven Rostedt 已提交
292 293 294
};

static void *
L
Li Zefan 已提交
295
__next(struct seq_file *m, loff_t *pos)
S
Steven Rostedt 已提交
296
{
L
Li Zefan 已提交
297
	long n = *pos - 1;
S
Steven Rostedt 已提交
298

299
	if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
S
Steven Rostedt 已提交
300 301
		return NULL;

L
Li Zefan 已提交
302
	m->private = (void *)n;
S
Steven Rostedt 已提交
303
	return &m->private;
S
Steven Rostedt 已提交
304 305
}

L
Li Zefan 已提交
306 307
static void *
t_next(struct seq_file *m, void *v, loff_t *pos)
S
Steven Rostedt 已提交
308
{
L
Li Zefan 已提交
309 310 311
	(*pos)++;
	return __next(m, pos);
}
S
Steven Rostedt 已提交
312

L
Li Zefan 已提交
313 314
static void *t_start(struct seq_file *m, loff_t *pos)
{
S
Steven Rostedt 已提交
315
	local_irq_disable();
316

317
	__this_cpu_inc(disable_stack_tracer);
318

319
	arch_spin_lock(&stack_trace_max_lock);
S
Steven Rostedt 已提交
320

321 322 323
	if (*pos == 0)
		return SEQ_START_TOKEN;

L
Li Zefan 已提交
324
	return __next(m, pos);
S
Steven Rostedt 已提交
325 326 327 328
}

static void t_stop(struct seq_file *m, void *p)
{
329
	arch_spin_unlock(&stack_trace_max_lock);
330

331
	__this_cpu_dec(disable_stack_tracer);
332

S
Steven Rostedt 已提交
333 334 335
	local_irq_enable();
}

336
static void trace_lookup_stack(struct seq_file *m, long i)
S
Steven Rostedt 已提交
337
{
S
Steven Rostedt 已提交
338
	unsigned long addr = stack_dump_trace[i];
S
Steven Rostedt 已提交
339

340
	seq_printf(m, "%pS\n", (void *)addr);
S
Steven Rostedt 已提交
341 342
}

343 344 345 346 347 348 349 350 351 352 353
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
354 355
static int t_show(struct seq_file *m, void *v)
{
356
	long i;
S
Steven Rostedt 已提交
357 358
	int size;

359
	if (v == SEQ_START_TOKEN) {
360
		seq_printf(m, "        Depth    Size   Location"
S
Steven Rostedt 已提交
361
			   "    (%d entries)\n"
362
			   "        -----    ----   --------\n",
363
			   stack_trace_max.nr_entries);
364

365
		if (!stack_tracer_enabled && !stack_trace_max_size)
366 367
			print_disabled(m);

S
Steven Rostedt 已提交
368 369
		return 0;
	}
S
Steven Rostedt 已提交
370

371 372
	i = *(long *)v;

373
	if (i >= stack_trace_max.nr_entries ||
S
Steven Rostedt 已提交
374
	    stack_dump_trace[i] == ULONG_MAX)
S
Steven Rostedt 已提交
375 376
		return 0;

377
	if (i+1 == stack_trace_max.nr_entries ||
S
Steven Rostedt 已提交
378
	    stack_dump_trace[i+1] == ULONG_MAX)
379
		size = stack_trace_index[i];
S
Steven Rostedt 已提交
380
	else
381
		size = stack_trace_index[i] - stack_trace_index[i+1];
S
Steven Rostedt 已提交
382

383
	seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
S
Steven Rostedt 已提交
384 385

	trace_lookup_stack(m, i);
S
Steven Rostedt 已提交
386 387 388 389

	return 0;
}

390
static const struct seq_operations stack_trace_seq_ops = {
S
Steven Rostedt 已提交
391 392 393 394 395 396 397 398
	.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
399
	return seq_open(file, &stack_trace_seq_ops);
S
Steven Rostedt 已提交
400 401
}

402
static const struct file_operations stack_trace_fops = {
S
Steven Rostedt 已提交
403 404 405
	.open		= stack_trace_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
L
Li Zefan 已提交
406
	.release	= seq_release,
S
Steven Rostedt 已提交
407 408
};

409 410 411 412 413 414 415 416 417 418 419
static int
stack_trace_filter_open(struct inode *inode, struct file *file)
{
	return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
				 inode, file);
}

static const struct file_operations stack_trace_filter_fops = {
	.open = stack_trace_filter_open,
	.read = seq_read,
	.write = ftrace_filter_write,
420
	.llseek = tracing_lseek,
421 422 423
	.release = ftrace_regex_release,
};

424 425
int
stack_trace_sysctl(struct ctl_table *table, int write,
426
		   void __user *buffer, size_t *lenp,
427 428 429 430 431 432
		   loff_t *ppos)
{
	int ret;

	mutex_lock(&stack_sysctl_mutex);

433
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
434 435

	if (ret || !write ||
436
	    (last_stack_tracer_enabled == !!stack_tracer_enabled))
437 438
		goto out;

439
	last_stack_tracer_enabled = !!stack_tracer_enabled;
440 441 442 443 444 445 446 447 448 449 450

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

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

451 452
static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;

453 454
static __init int enable_stacktrace(char *str)
{
455 456 457
	if (strncmp(str, "_filter=", 8) == 0)
		strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);

458 459
	stack_tracer_enabled = 1;
	last_stack_tracer_enabled = 1;
460 461 462 463
	return 1;
}
__setup("stacktrace", enable_stacktrace);

S
Steven Rostedt 已提交
464 465 466 467 468
static __init int stack_trace_init(void)
{
	struct dentry *d_tracer;

	d_tracer = tracing_init_dentry();
469
	if (IS_ERR(d_tracer))
470
		return 0;
S
Steven Rostedt 已提交
471

472
	trace_create_file("stack_max_size", 0644, d_tracer,
473
			&stack_trace_max_size, &stack_max_size_fops);
S
Steven Rostedt 已提交
474

475 476
	trace_create_file("stack_trace", 0444, d_tracer,
			NULL, &stack_trace_fops);
S
Steven Rostedt 已提交
477

478 479 480
	trace_create_file("stack_trace_filter", 0444, d_tracer,
			NULL, &stack_trace_filter_fops);

481 482 483
	if (stack_trace_filter_buf[0])
		ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);

484
	if (stack_tracer_enabled)
485
		register_ftrace_function(&trace_ops);
S
Steven Rostedt 已提交
486 487 488 489 490

	return 0;
}

device_initcall(stack_trace_init);