kernel/trace/trace_stack.c

   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
  21 #define STACK_TRACE_ENTRIES 500
  22
  23 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES];
  24 static unsigned stack_trace_index[STACK_TRACE_ENTRIES];
  25
  26 static unsigned int stack_trace_nr_entries;
  27 static unsigned long stack_trace_max_size;
  28 static arch_spinlock_t stack_trace_max_lock =
  29         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  30
  31 DEFINE_PER_CPU(int, disable_stack_tracer);
  32 static DEFINE_MUTEX(stack_sysctl_mutex);
  33
  34 int stack_tracer_enabled;
  35
  36 static void print_max_stack(void)
  37 {
  38         long i;
  39         int size;
  40
  41         pr_emerg("        Depth    Size   Location    (%d entries)\n"
  42                            "        -----    ----   --------\n",
  43                            stack_trace_nr_entries);
  44
  45         for (i = 0; i < stack_trace_nr_entries; i++) {
  46                 if (i + 1 == stack_trace_nr_entries)
  47                         size = stack_trace_index[i];
  48                 else
  49                         size = stack_trace_index[i] - stack_trace_index[i+1];
  50
  51                 pr_emerg("%3ld) %8d   %5d   %pS\n", i, stack_trace_index[i],
  52                                 size, (void *)stack_dump_trace[i]);
  53         }
  54 }
  55
  56 static void check_stack(unsigned long ip, unsigned long *stack)
  57 {
  58         unsigned long this_size, flags; unsigned long *p, *top, *start;
  59         static int tracer_frame;
  60         int frame_size = READ_ONCE(tracer_frame);
  61         int i, x;
  62
  63         this_size = ((unsigned long)stack) & (THREAD_SIZE-1);
  64         this_size = THREAD_SIZE - this_size;
  65         /* Remove the frame of the tracer */
  66         this_size -= frame_size;
  67
  68         if (this_size <= stack_trace_max_size)
  69                 return;
  70
  71         /* we do not handle interrupt stacks yet */
  72         if (!object_is_on_stack(stack))
  73                 return;
  74
  75         /* Can't do this from NMI context (can cause deadlocks) */
  76         if (in_nmi())
  77                 return;
  78
  79         local_irq_save(flags);
  80         arch_spin_lock(&stack_trace_max_lock);
  81
  82         /* In case another CPU set the tracer_frame on us */
  83         if (unlikely(!frame_size))
  84                 this_size -= tracer_frame;
  85
  86         /* a race could have already updated it */
  87         if (this_size <= stack_trace_max_size)
  88                 goto out;
  89
  90         stack_trace_max_size = this_size;
  91
  92         stack_trace_nr_entries = stack_trace_save(stack_dump_trace,
  93                                                ARRAY_SIZE(stack_dump_trace) - 1,
  94                                                0);
  95
  96         /* Skip over the overhead of the stack tracer itself */
  97         for (i = 0; i < stack_trace_nr_entries; i++) {
  98                 if (stack_dump_trace[i] == ip)
  99                         break;
 100         }
 101
 102         /*
 103          * Some archs may not have the passed in ip in the dump.
 104          * If that happens, we need to show everything.
 105          */
 106         if (i == stack_trace_nr_entries)
 107                 i = 0;
 108
 109         /*
 110          * Now find where in the stack these are.
 111          */
 112         x = 0;
 113         start = stack;
 114         top = (unsigned long *)
 115                 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
 116
 117         /*
 118          * Loop through all the entries. One of the entries may
 119          * for some reason be missed on the stack, so we may
 120          * have to account for them. If they are all there, this
 121          * loop will only happen once. This code only takes place
 122          * on a new max, so it is far from a fast path.
 123          */
 124         while (i < stack_trace_nr_entries) {
 125                 int found = 0;
 126
 127                 stack_trace_index[x] = this_size;
 128                 p = start;
 129
 130                 for (; p < top && i < stack_trace_nr_entries; p++) {
 131                         /*
 132                          * The READ_ONCE_NOCHECK is used to let KASAN know that
 133                          * this is not a stack-out-of-bounds error.
 134                          */
 135                         if ((READ_ONCE_NOCHECK(*p)) == stack_dump_trace[i]) {
 136                                 stack_dump_trace[x] = stack_dump_trace[i++];
 137                                 this_size = stack_trace_index[x++] =
 138                                         (top - p) * sizeof(unsigned long);
 139                                 found = 1;
 140                                 /* Start the search from here */
 141                                 start = p + 1;
 142                                 /*
 143                                  * We do not want to show the overhead
 144                                  * of the stack tracer stack in the
 145                                  * max stack. If we haven't figured
 146                                  * out what that is, then figure it out
 147                                  * now.
 148                                  */
 149                                 if (unlikely(!tracer_frame)) {
 150                                         tracer_frame = (p - stack) *
 151                                                 sizeof(unsigned long);
 152                                         stack_trace_max_size -= tracer_frame;
 153                                 }
 154                         }
 155                 }
 156
 157                 if (!found)
 158                         i++;
 159         }
 160
 161         stack_trace_nr_entries = x;
 162
 163         if (task_stack_end_corrupted(current)) {
 164                 print_max_stack();
 165                 BUG();
 166         }
 167
 168  out:
 169         arch_spin_unlock(&stack_trace_max_lock);
 170         local_irq_restore(flags);
 171 }
 172
 173 static void
 174 stack_trace_call(unsigned long ip, unsigned long parent_ip,
 175                  struct ftrace_ops *op, struct pt_regs *pt_regs)
 176 {
 177         unsigned long stack;
 178
 179         preempt_disable_notrace();
 180
 181         /* no atomic needed, we only modify this variable by this cpu */
 182         __this_cpu_inc(disable_stack_tracer);
 183         if (__this_cpu_read(disable_stack_tracer) != 1)
 184                 goto out;
 185
 186         /* If rcu is not watching, then save stack trace can fail */
 187         if (!rcu_is_watching())
 188                 goto out;
 189
 190         ip += MCOUNT_INSN_SIZE;
 191
 192         check_stack(ip, &stack);
 193
 194  out:
 195         __this_cpu_dec(disable_stack_tracer);
 196         /* prevent recursion in schedule */
 197         preempt_enable_notrace();
 198 }
 199
 200 static struct ftrace_ops trace_ops __read_mostly =
 201 {
 202         .func = stack_trace_call,
 203         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
 204 };
 205
 206 static ssize_t
 207 stack_max_size_read(struct file *filp, char __user *ubuf,
 208                     size_t count, loff_t *ppos)
 209 {
 210         unsigned long *ptr = filp->private_data;
 211         char buf[64];
 212         int r;
 213
 214         r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
 215         if (r > sizeof(buf))
 216                 r = sizeof(buf);
 217         return simple_read_from_buffer(ubuf, count, ppos, buf, r);
 218 }
 219
 220 static ssize_t
 221 stack_max_size_write(struct file *filp, const char __user *ubuf,
 222                      size_t count, loff_t *ppos)
 223 {
 224         long *ptr = filp->private_data;
 225         unsigned long val, flags;
 226         int ret;
 227
 228         ret = kstrtoul_from_user(ubuf, count, 10, &val);
 229         if (ret)
 230                 return ret;
 231
 232         local_irq_save(flags);
 233
 234         /*
 235          * In case we trace inside arch_spin_lock() or after (NMI),
 236          * we will cause circular lock, so we also need to increase
 237          * the percpu disable_stack_tracer here.
 238          */
 239         __this_cpu_inc(disable_stack_tracer);
 240
 241         arch_spin_lock(&stack_trace_max_lock);
 242         *ptr = val;
 243         arch_spin_unlock(&stack_trace_max_lock);
 244
 245         __this_cpu_dec(disable_stack_tracer);
 246         local_irq_restore(flags);
 247
 248         return count;
 249 }
 250
 251 static const struct file_operations stack_max_size_fops = {
 252         .open           = tracing_open_generic,
 253         .read           = stack_max_size_read,
 254         .write          = stack_max_size_write,
 255         .llseek         = default_llseek,
 256 };
 257
 258 static void *
 259 __next(struct seq_file *m, loff_t *pos)
 260 {
 261         long n = *pos - 1;
 262
 263         if (n >= stack_trace_nr_entries)
 264                 return NULL;
 265
 266         m->private = (void *)n;
 267         return &m->private;
 268 }
 269
 270 static void *
 271 t_next(struct seq_file *m, void *v, loff_t *pos)
 272 {
 273         (*pos)++;
 274         return __next(m, pos);
 275 }
 276
 277 static void *t_start(struct seq_file *m, loff_t *pos)
 278 {
 279         local_irq_disable();
 280
 281         __this_cpu_inc(disable_stack_tracer);
 282
 283         arch_spin_lock(&stack_trace_max_lock);
 284
 285         if (*pos == 0)
 286                 return SEQ_START_TOKEN;
 287
 288         return __next(m, pos);
 289 }
 290
 291 static void t_stop(struct seq_file *m, void *p)
 292 {
 293         arch_spin_unlock(&stack_trace_max_lock);
 294
 295         __this_cpu_dec(disable_stack_tracer);
 296
 297         local_irq_enable();
 298 }
 299
 300 static void trace_lookup_stack(struct seq_file *m, long i)
 301 {
 302         unsigned long addr = stack_dump_trace[i];
 303
 304         seq_printf(m, "%pS\n", (void *)addr);
 305 }
 306
 307 static void print_disabled(struct seq_file *m)
 308 {
 309         seq_puts(m, "#\n"
 310                  "#  Stack tracer disabled\n"
 311                  "#\n"
 312                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
 313                  "# kernel command line\n"
 314                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
 315                  "#\n");
 316 }
 317
 318 static int t_show(struct seq_file *m, void *v)
 319 {
 320         long i;
 321         int size;
 322
 323         if (v == SEQ_START_TOKEN) {
 324                 seq_printf(m, "        Depth    Size   Location"
 325                            "    (%d entries)\n"
 326                            "        -----    ----   --------\n",
 327                            stack_trace_nr_entries);
 328
 329                 if (!stack_tracer_enabled && !stack_trace_max_size)
 330                         print_disabled(m);
 331
 332                 return 0;
 333         }
 334
 335         i = *(long *)v;
 336
 337         if (i >= stack_trace_nr_entries)
 338                 return 0;
 339
 340         if (i + 1 == stack_trace_nr_entries)
 341                 size = stack_trace_index[i];
 342         else
 343                 size = stack_trace_index[i] - stack_trace_index[i+1];
 344
 345         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
 346
 347         trace_lookup_stack(m, i);
 348
 349         return 0;
 350 }
 351
 352 static const struct seq_operations stack_trace_seq_ops = {
 353         .start          = t_start,
 354         .next           = t_next,
 355         .stop           = t_stop,
 356         .show           = t_show,
 357 };
 358
 359 static int stack_trace_open(struct inode *inode, struct file *file)
 360 {
 361         return seq_open(file, &stack_trace_seq_ops);
 362 }
 363
 364 static const struct file_operations stack_trace_fops = {
 365         .open           = stack_trace_open,
 366         .read           = seq_read,
 367         .llseek         = seq_lseek,
 368         .release        = seq_release,
 369 };
 370
 371 #ifdef CONFIG_DYNAMIC_FTRACE
 372
 373 static int
 374 stack_trace_filter_open(struct inode *inode, struct file *file)
 375 {
 376         struct ftrace_ops *ops = inode->i_private;
 377
 378         return ftrace_regex_open(ops, FTRACE_ITER_FILTER,
 379                                  inode, file);
 380 }
 381
 382 static const struct file_operations stack_trace_filter_fops = {
 383         .open = stack_trace_filter_open,
 384         .read = seq_read,
 385         .write = ftrace_filter_write,
 386         .llseek = tracing_lseek,
 387         .release = ftrace_regex_release,
 388 };
 389
 390 #endif /* CONFIG_DYNAMIC_FTRACE */
 391
 392 int
 393 stack_trace_sysctl(struct ctl_table *table, int write,
 394                    void __user *buffer, size_t *lenp,
 395                    loff_t *ppos)
 396 {
 397         int was_enabled;
 398         int ret;
 399
 400         mutex_lock(&stack_sysctl_mutex);
 401         was_enabled = !!stack_tracer_enabled;
 402
 403         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 404
 405         if (ret || !write || (was_enabled == !!stack_tracer_enabled))
 406                 goto out;
 407
 408         if (stack_tracer_enabled)
 409                 register_ftrace_function(&trace_ops);
 410         else
 411                 unregister_ftrace_function(&trace_ops);
 412  out:
 413         mutex_unlock(&stack_sysctl_mutex);
 414         return ret;
 415 }
 416
 417 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
 418
 419 static __init int enable_stacktrace(char *str)
 420 {
 421         int len;
 422
 423         if ((len = str_has_prefix(str, "_filter=")))
 424                 strncpy(stack_trace_filter_buf, str + len, COMMAND_LINE_SIZE);
 425
 426         stack_tracer_enabled = 1;
 427         return 1;
 428 }
 429 __setup("stacktrace", enable_stacktrace);
 430
 431 static __init int stack_trace_init(void)
 432 {
 433         struct dentry *d_tracer;
 434
 435         d_tracer = tracing_init_dentry();
 436         if (IS_ERR(d_tracer))
 437                 return 0;
 438
 439         trace_create_file("stack_max_size", 0644, d_tracer,
 440                         &stack_trace_max_size, &stack_max_size_fops);
 441
 442         trace_create_file("stack_trace", 0444, d_tracer,
 443                         NULL, &stack_trace_fops);
 444
 445 #ifdef CONFIG_DYNAMIC_FTRACE
 446         trace_create_file("stack_trace_filter", 0644, d_tracer,
 447                           &trace_ops, &stack_trace_filter_fops);
 448 #endif
 449
 450         if (stack_trace_filter_buf[0])
 451                 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
 452
 453         if (stack_tracer_enabled)
 454                 register_ftrace_function(&trace_ops);
 455
 456         return 0;
 457 }
 458
 459 device_initcall(stack_trace_init);