linux/kernel/trace/trace_stack.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
   4 *
   5 */
   6#include <linux/sched/task_stack.h>
   7#include <linux/stacktrace.h>
   8#include <linux/kallsyms.h>
   9#include <linux/seq_file.h>
  10#include <linux/spinlock.h>
  11#include <linux/uaccess.h>
  12#include <linux/ftrace.h>
  13#include <linux/module.h>
  14#include <linux/sysctl.h>
  15#include <linux/init.h>
  16
  17#include <asm/setup.h>
  18
  19#include "trace.h"
  20
  21static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
  22         { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
  23unsigned stack_trace_index[STACK_TRACE_ENTRIES];
  24
  25/*
  26 * Reserve one entry for the passed in ip. This will allow
  27 * us to remove most or all of the stack size overhead
  28 * added by the stack tracer itself.
  29 */
  30struct stack_trace stack_trace_max = {
  31        .max_entries            = STACK_TRACE_ENTRIES - 1,
  32        .entries                = &stack_dump_trace[0],
  33};
  34
  35unsigned long stack_trace_max_size;
  36arch_spinlock_t stack_trace_max_lock =
  37        (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  38
  39DEFINE_PER_CPU(int, disable_stack_tracer);
  40static DEFINE_MUTEX(stack_sysctl_mutex);
  41
  42int stack_tracer_enabled;
  43static int last_stack_tracer_enabled;
  44
  45void stack_trace_print(void)
  46{
  47        long i;
  48        int size;
  49
  50        pr_emerg("        Depth    Size   Location    (%d entries)\n"
  51                           "        -----    ----   --------\n",
  52                           stack_trace_max.nr_entries);
  53
  54        for (i = 0; i < stack_trace_max.nr_entries; i++) {
  55                if (stack_dump_trace[i] == ULONG_MAX)
  56                        break;
  57                if (i+1 == stack_trace_max.nr_entries ||
  58                                stack_dump_trace[i+1] == ULONG_MAX)
  59                        size = stack_trace_index[i];
  60                else
  61                        size = stack_trace_index[i] - stack_trace_index[i+1];
  62
  63                pr_emerg("%3ld) %8d   %5d   %pS\n", i, stack_trace_index[i],
  64                                size, (void *)stack_dump_trace[i]);
  65        }
  66}
  67
  68/*
  69 * When arch-specific code overrides this function, the following
  70 * data should be filled up, assuming stack_trace_max_lock is held to
  71 * prevent concurrent updates.
  72 *     stack_trace_index[]
  73 *     stack_trace_max
  74 *     stack_trace_max_size
  75 */
  76void __weak
  77check_stack(unsigned long ip, unsigned long *stack)
  78{
  79        unsigned long this_size, flags; unsigned long *p, *top, *start;
  80        static int tracer_frame;
  81        int frame_size = ACCESS_ONCE(tracer_frame);
  82        int i, x;
  83
  84        this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
  85        this_size = THREAD_SIZE - this_size;
  86        /* Remove the frame of the tracer */
  87        this_size -= frame_size;
  88
  89        if (this_size <= stack_trace_max_size)
  90                return;
  91
  92        /* we do not handle interrupt stacks yet */
  93        if (!object_is_on_stack(stack))
  94                return;
  95
  96        /* Can't do this from NMI context (can cause deadlocks) */
  97        if (in_nmi())
  98                return;
  99
 100        local_irq_save(flags);
 101        arch_spin_lock(&stack_trace_max_lock);
 102
 103        /* In case another CPU set the tracer_frame on us */
 104        if (unlikely(!frame_size))
 105                this_size -= tracer_frame;
 106
 107        /* a race could have already updated it */
 108        if (this_size <= stack_trace_max_size)
 109                goto out;
 110
 111        stack_trace_max_size = this_size;
 112
 113        stack_trace_max.nr_entries = 0;
 114        stack_trace_max.skip = 3;
 115
 116        save_stack_trace(&stack_trace_max);
 117
 118        /* Skip over the overhead of the stack tracer itself */
 119        for (i = 0; i < stack_trace_max.nr_entries; i++) {
 120                if (stack_dump_trace[i] == ip)
 121                        break;
 122        }
 123
 124        /*
 125         * Some archs may not have the passed in ip in the dump.
 126         * If that happens, we need to show everything.
 127         */
 128        if (i == stack_trace_max.nr_entries)
 129                i = 0;
 130
 131        /*
 132         * Now find where in the stack these are.
 133         */
 134        x = 0;
 135        start = stack;
 136        top = (unsigned long *)
 137                (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
 138
 139        /*
 140         * Loop through all the entries. One of the entries may
 141         * for some reason be missed on the stack, so we may
 142         * have to account for them. If they are all there, this
 143         * loop will only happen once. This code only takes place
 144         * on a new max, so it is far from a fast path.
 145         */
 146        while (i < stack_trace_max.nr_entries) {
 147                int found = 0;
 148
 149                stack_trace_index[x] = this_size;
 150                p = start;
 151
 152                for (; p < top && i < stack_trace_max.nr_entries; p++) {
 153                        if (stack_dump_trace[i] == ULONG_MAX)
 154                                break;
 155                        /*
 156                         * The READ_ONCE_NOCHECK is used to let KASAN know that
 157                         * this is not a stack-out-of-bounds error.
 158                         */
 159                        if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) {
 160                                stack_dump_trace[x] = stack_dump_trace[i++];
 161                                this_size = stack_trace_index[x++] =
 162                                        (top - p) * sizeof(unsigned long);
 163                                found = 1;
 164                                /* Start the search from here */
 165                                start = p + 1;
 166                                /*
 167                                 * We do not want to show the overhead
 168                                 * of the stack tracer stack in the
 169                                 * max stack. If we haven't figured
 170                                 * out what that is, then figure it out
 171                                 * now.
 172                                 */
 173                                if (unlikely(!tracer_frame)) {
 174                                        tracer_frame = (p - stack) *
 175                                                sizeof(unsigned long);
 176                                        stack_trace_max_size -= tracer_frame;
 177                                }
 178                        }
 179                }
 180
 181                if (!found)
 182                        i++;
 183        }
 184
 185        stack_trace_max.nr_entries = x;
 186        for (; x < i; x++)
 187                stack_dump_trace[x] = ULONG_MAX;
 188
 189        if (task_stack_end_corrupted(current)) {
 190                stack_trace_print();
 191                BUG();
 192        }
 193
 194 out:
 195        arch_spin_unlock(&stack_trace_max_lock);
 196        local_irq_restore(flags);
 197}
 198
 199static void
 200stack_trace_call(unsigned long ip, unsigned long parent_ip,
 201                 struct ftrace_ops *op, struct pt_regs *pt_regs)
 202{
 203        unsigned long stack;
 204
 205        preempt_disable_notrace();
 206
 207        /* no atomic needed, we only modify this variable by this cpu */
 208        __this_cpu_inc(disable_stack_tracer);
 209        if (__this_cpu_read(disable_stack_tracer) != 1)
 210                goto out;
 211
 212        ip += MCOUNT_INSN_SIZE;
 213
 214        check_stack(ip, &stack);
 215
 216 out:
 217        __this_cpu_dec(disable_stack_tracer);
 218        /* prevent recursion in schedule */
 219        preempt_enable_notrace();
 220}
 221
 222static struct ftrace_ops trace_ops __read_mostly =
 223{
 224        .func = stack_trace_call,
 225        .flags = FTRACE_OPS_FL_RECURSION_SAFE,
 226};
 227
 228static ssize_t
 229stack_max_size_read(struct file *filp, char __user *ubuf,
 230                    size_t count, loff_t *ppos)
 231{
 232        unsigned long *ptr = filp->private_data;
 233        char buf[64];
 234        int r;
 235
 236        r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
 237        if (r > sizeof(buf))
 238                r = sizeof(buf);
 239        return simple_read_from_buffer(ubuf, count, ppos, buf, r);
 240}
 241
 242static ssize_t
 243stack_max_size_write(struct file *filp, const char __user *ubuf,
 244                     size_t count, loff_t *ppos)
 245{
 246        long *ptr = filp->private_data;
 247        unsigned long val, flags;
 248        int ret;
 249
 250        ret = kstrtoul_from_user(ubuf, count, 10, &val);
 251        if (ret)
 252                return ret;
 253
 254        local_irq_save(flags);
 255
 256        /*
 257         * In case we trace inside arch_spin_lock() or after (NMI),
 258         * we will cause circular lock, so we also need to increase
 259         * the percpu disable_stack_tracer here.
 260         */
 261        __this_cpu_inc(disable_stack_tracer);
 262
 263        arch_spin_lock(&stack_trace_max_lock);
 264        *ptr = val;
 265        arch_spin_unlock(&stack_trace_max_lock);
 266
 267        __this_cpu_dec(disable_stack_tracer);
 268        local_irq_restore(flags);
 269
 270        return count;
 271}
 272
 273static const struct file_operations stack_max_size_fops = {
 274        .open           = tracing_open_generic,
 275        .read           = stack_max_size_read,
 276        .write          = stack_max_size_write,
 277        .llseek         = default_llseek,
 278};
 279
 280static void *
 281__next(struct seq_file *m, loff_t *pos)
 282{
 283        long n = *pos - 1;
 284
 285        if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
 286                return NULL;
 287
 288        m->private = (void *)n;
 289        return &m->private;
 290}
 291
 292static void *
 293t_next(struct seq_file *m, void *v, loff_t *pos)
 294{
 295        (*pos)++;
 296        return __next(m, pos);
 297}
 298
 299static void *t_start(struct seq_file *m, loff_t *pos)
 300{
 301        local_irq_disable();
 302
 303        __this_cpu_inc(disable_stack_tracer);
 304
 305        arch_spin_lock(&stack_trace_max_lock);
 306
 307        if (*pos == 0)
 308                return SEQ_START_TOKEN;
 309
 310        return __next(m, pos);
 311}
 312
 313static void t_stop(struct seq_file *m, void *p)
 314{
 315        arch_spin_unlock(&stack_trace_max_lock);
 316
 317        __this_cpu_dec(disable_stack_tracer);
 318
 319        local_irq_enable();
 320}
 321
 322static void trace_lookup_stack(struct seq_file *m, long i)
 323{
 324        unsigned long addr = stack_dump_trace[i];
 325
 326        seq_printf(m, "%pS\n", (void *)addr);
 327}
 328
 329static void print_disabled(struct seq_file *m)
 330{
 331        seq_puts(m, "#\n"
 332                 "#  Stack tracer disabled\n"
 333                 "#\n"
 334                 "# To enable the stack tracer, either add 'stacktrace' to the\n"
 335                 "# kernel command line\n"
 336                 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
 337                 "#\n");
 338}
 339
 340static int t_show(struct seq_file *m, void *v)
 341{
 342        long i;
 343        int size;
 344
 345        if (v == SEQ_START_TOKEN) {
 346                seq_printf(m, "        Depth    Size   Location"
 347                           "    (%d entries)\n"
 348                           "        -----    ----   --------\n",
 349                           stack_trace_max.nr_entries);
 350
 351                if (!stack_tracer_enabled && !stack_trace_max_size)
 352                        print_disabled(m);
 353
 354                return 0;
 355        }
 356
 357        i = *(long *)v;
 358
 359        if (i >= stack_trace_max.nr_entries ||
 360            stack_dump_trace[i] == ULONG_MAX)
 361                return 0;
 362
 363        if (i+1 == stack_trace_max.nr_entries ||
 364            stack_dump_trace[i+1] == ULONG_MAX)
 365                size = stack_trace_index[i];
 366        else
 367                size = stack_trace_index[i] - stack_trace_index[i+1];
 368
 369        seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
 370
 371        trace_lookup_stack(m, i);
 372
 373        return 0;
 374}
 375
 376static const struct seq_operations stack_trace_seq_ops = {
 377        .start          = t_start,
 378        .next           = t_next,
 379        .stop           = t_stop,
 380        .show           = t_show,
 381};
 382
 383static int stack_trace_open(struct inode *inode, struct file *file)
 384{
 385        return seq_open(file, &stack_trace_seq_ops);
 386}
 387
 388static const struct file_operations stack_trace_fops = {
 389        .open           = stack_trace_open,
 390        .read           = seq_read,
 391        .llseek         = seq_lseek,
 392        .release        = seq_release,
 393};
 394
 395#ifdef CONFIG_DYNAMIC_FTRACE
 396
 397static int
 398stack_trace_filter_open(struct inode *inode, struct file *file)
 399{
 400        struct ftrace_ops *ops = inode->i_private;
 401
 402        return ftrace_regex_open(ops, FTRACE_ITER_FILTER,
 403                                 inode, file);
 404}
 405
 406static const struct file_operations stack_trace_filter_fops = {
 407        .open = stack_trace_filter_open,
 408        .read = seq_read,
 409        .write = ftrace_filter_write,
 410        .llseek = tracing_lseek,
 411        .release = ftrace_regex_release,
 412};
 413
 414#endif /* CONFIG_DYNAMIC_FTRACE */
 415
 416int
 417stack_trace_sysctl(struct ctl_table *table, int write,
 418                   void __user *buffer, size_t *lenp,
 419                   loff_t *ppos)
 420{
 421        int ret;
 422
 423        mutex_lock(&stack_sysctl_mutex);
 424
 425        ret = proc_dointvec(table, write, buffer, lenp, ppos);
 426
 427        if (ret || !write ||
 428            (last_stack_tracer_enabled == !!stack_tracer_enabled))
 429                goto out;
 430
 431        last_stack_tracer_enabled = !!stack_tracer_enabled;
 432
 433        if (stack_tracer_enabled)
 434                register_ftrace_function(&trace_ops);
 435        else
 436                unregister_ftrace_function(&trace_ops);
 437
 438 out:
 439        mutex_unlock(&stack_sysctl_mutex);
 440        return ret;
 441}
 442
 443static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
 444
 445static __init int enable_stacktrace(char *str)
 446{
 447        if (strncmp(str, "_filter=", 8) == 0)
 448                strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
 449
 450        stack_tracer_enabled = 1;
 451        last_stack_tracer_enabled = 1;
 452        return 1;
 453}
 454__setup("stacktrace", enable_stacktrace);
 455
 456static __init int stack_trace_init(void)
 457{
 458        struct dentry *d_tracer;
 459
 460        d_tracer = tracing_init_dentry();
 461        if (IS_ERR(d_tracer))
 462                return 0;
 463
 464        trace_create_file("stack_max_size", 0644, d_tracer,
 465                        &stack_trace_max_size, &stack_max_size_fops);
 466
 467        trace_create_file("stack_trace", 0444, d_tracer,
 468                        NULL, &stack_trace_fops);
 469
 470#ifdef CONFIG_DYNAMIC_FTRACE
 471        trace_create_file("stack_trace_filter", 0444, d_tracer,
 472                          &trace_ops, &stack_trace_filter_fops);
 473#endif
 474
 475        if (stack_trace_filter_buf[0])
 476                ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
 477
 478        if (stack_tracer_enabled)
 479                register_ftrace_function(&trace_ops);
 480
 481        return 0;
 482}
 483
 484device_initcall(stack_trace_init);
 485