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 static unsigned long stack_dump_trace[STACK_TRACE_ENTRIES+1] =
  22          { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
  23 unsigned 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  */
  30 struct stack_trace stack_trace_max = {
  31         .max_entries            = STACK_TRACE_ENTRIES - 1,
  32         .entries                = &stack_dump_trace[0],
  33 };
  34
  35 unsigned long stack_trace_max_size;
  36 arch_spinlock_t stack_trace_max_lock =
  37         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  38
  39 DEFINE_PER_CPU(int, disable_stack_tracer);
  40 static DEFINE_MUTEX(stack_sysctl_mutex);
  41
  42 int stack_tracer_enabled;
  43 static int last_stack_tracer_enabled;
  44
  45 void 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  */
  76 void __weak
  77 check_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 = READ_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
 199 /* Some archs may not define MCOUNT_INSN_SIZE */
 200 #ifndef MCOUNT_INSN_SIZE
 201 # define MCOUNT_INSN_SIZE 0
 202 #endif
 203
 204 static void
 205 stack_trace_call(unsigned long ip, unsigned long parent_ip,
 206                  struct ftrace_ops *op, struct pt_regs *pt_regs)
 207 {
 208         unsigned long stack;
 209
 210         preempt_disable_notrace();
 211
 212         /* no atomic needed, we only modify this variable by this cpu */
 213         __this_cpu_inc(disable_stack_tracer);
 214         if (__this_cpu_read(disable_stack_tracer) != 1)
 215                 goto out;
 216
 217         /* If rcu is not watching, then save stack trace can fail */
 218         if (!rcu_is_watching())
 219                 goto out;
 220
 221         ip += MCOUNT_INSN_SIZE;
 222
 223         check_stack(ip, &stack);
 224
 225  out:
 226         __this_cpu_dec(disable_stack_tracer);
 227         /* prevent recursion in schedule */
 228         preempt_enable_notrace();
 229 }
 230
 231 static struct ftrace_ops trace_ops __read_mostly =
 232 {
 233         .func = stack_trace_call,
 234         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
 235 };
 236
 237 static ssize_t
 238 stack_max_size_read(struct file *filp, char __user *ubuf,
 239                     size_t count, loff_t *ppos)
 240 {
 241         unsigned long *ptr = filp->private_data;
 242         char buf[64];
 243         int r;
 244
 245         r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
 246         if (r > sizeof(buf))
 247                 r = sizeof(buf);
 248         return simple_read_from_buffer(ubuf, count, ppos, buf, r);
 249 }
 250
 251 static ssize_t
 252 stack_max_size_write(struct file *filp, const char __user *ubuf,
 253                      size_t count, loff_t *ppos)
 254 {
 255         long *ptr = filp->private_data;
 256         unsigned long val, flags;
 257         int ret;
 258
 259         ret = kstrtoul_from_user(ubuf, count, 10, &val);
 260         if (ret)
 261                 return ret;
 262
 263         local_irq_save(flags);
 264
 265         /*
 266          * In case we trace inside arch_spin_lock() or after (NMI),
 267          * we will cause circular lock, so we also need to increase
 268          * the percpu disable_stack_tracer here.
 269          */
 270         __this_cpu_inc(disable_stack_tracer);
 271
 272         arch_spin_lock(&stack_trace_max_lock);
 273         *ptr = val;
 274         arch_spin_unlock(&stack_trace_max_lock);
 275
 276         __this_cpu_dec(disable_stack_tracer);
 277         local_irq_restore(flags);
 278
 279         return count;
 280 }
 281
 282 static const struct file_operations stack_max_size_fops = {
 283         .open           = tracing_open_generic,
 284         .read           = stack_max_size_read,
 285         .write          = stack_max_size_write,
 286         .llseek         = default_llseek,
 287 };
 288
 289 static void *
 290 __next(struct seq_file *m, loff_t *pos)
 291 {
 292         long n = *pos - 1;
 293
 294         if (n > stack_trace_max.nr_entries || stack_dump_trace[n] == ULONG_MAX)
 295                 return NULL;
 296
 297         m->private = (void *)n;
 298         return &m->private;
 299 }
 300
 301 static void *
 302 t_next(struct seq_file *m, void *v, loff_t *pos)
 303 {
 304         (*pos)++;
 305         return __next(m, pos);
 306 }
 307
 308 static void *t_start(struct seq_file *m, loff_t *pos)
 309 {
 310         local_irq_disable();
 311
 312         __this_cpu_inc(disable_stack_tracer);
 313
 314         arch_spin_lock(&stack_trace_max_lock);
 315
 316         if (*pos == 0)
 317                 return SEQ_START_TOKEN;
 318
 319         return __next(m, pos);
 320 }
 321
 322 static void t_stop(struct seq_file *m, void *p)
 323 {
 324         arch_spin_unlock(&stack_trace_max_lock);
 325
 326         __this_cpu_dec(disable_stack_tracer);
 327
 328         local_irq_enable();
 329 }
 330
 331 static void trace_lookup_stack(struct seq_file *m, long i)
 332 {
 333         unsigned long addr = stack_dump_trace[i];
 334
 335         seq_printf(m, "%pS\n", (void *)addr);
 336 }
 337
 338 static void print_disabled(struct seq_file *m)
 339 {
 340         seq_puts(m, "#\n"
 341                  "#  Stack tracer disabled\n"
 342                  "#\n"
 343                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
 344                  "# kernel command line\n"
 345                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
 346                  "#\n");
 347 }
 348
 349 static int t_show(struct seq_file *m, void *v)
 350 {
 351         long i;
 352         int size;
 353
 354         if (v == SEQ_START_TOKEN) {
 355                 seq_printf(m, "        Depth    Size   Location"
 356                            "    (%d entries)\n"
 357                            "        -----    ----   --------\n",
 358                            stack_trace_max.nr_entries);
 359
 360                 if (!stack_tracer_enabled && !stack_trace_max_size)
 361                         print_disabled(m);
 362
 363                 return 0;
 364         }
 365
 366         i = *(long *)v;
 367
 368         if (i >= stack_trace_max.nr_entries ||
 369             stack_dump_trace[i] == ULONG_MAX)
 370                 return 0;
 371
 372         if (i+1 == stack_trace_max.nr_entries ||
 373             stack_dump_trace[i+1] == ULONG_MAX)
 374                 size = stack_trace_index[i];
 375         else
 376                 size = stack_trace_index[i] - stack_trace_index[i+1];
 377
 378         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_trace_index[i], size);
 379
 380         trace_lookup_stack(m, i);
 381
 382         return 0;
 383 }
 384
 385 static const struct seq_operations stack_trace_seq_ops = {
 386         .start          = t_start,
 387         .next           = t_next,
 388         .stop           = t_stop,
 389         .show           = t_show,
 390 };
 391
 392 static int stack_trace_open(struct inode *inode, struct file *file)
 393 {
 394         return seq_open(file, &stack_trace_seq_ops);
 395 }
 396
 397 static const struct file_operations stack_trace_fops = {
 398         .open           = stack_trace_open,
 399         .read           = seq_read,
 400         .llseek         = seq_lseek,
 401         .release        = seq_release,
 402 };
 403
 404 #ifdef CONFIG_DYNAMIC_FTRACE
 405
 406 static int
 407 stack_trace_filter_open(struct inode *inode, struct file *file)
 408 {
 409         struct ftrace_ops *ops = inode->i_private;
 410
 411         return ftrace_regex_open(ops, FTRACE_ITER_FILTER,
 412                                  inode, file);
 413 }
 414
 415 static const struct file_operations stack_trace_filter_fops = {
 416         .open = stack_trace_filter_open,
 417         .read = seq_read,
 418         .write = ftrace_filter_write,
 419         .llseek = tracing_lseek,
 420         .release = ftrace_regex_release,
 421 };
 422
 423 #endif /* CONFIG_DYNAMIC_FTRACE */
 424
 425 int
 426 stack_trace_sysctl(struct ctl_table *table, int write,
 427                    void __user *buffer, size_t *lenp,
 428                    loff_t *ppos)
 429 {
 430         int ret;
 431
 432         mutex_lock(&stack_sysctl_mutex);
 433
 434         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 435
 436         if (ret || !write ||
 437             (last_stack_tracer_enabled == !!stack_tracer_enabled))
 438                 goto out;
 439
 440         last_stack_tracer_enabled = !!stack_tracer_enabled;
 441
 442         if (stack_tracer_enabled)
 443                 register_ftrace_function(&trace_ops);
 444         else
 445                 unregister_ftrace_function(&trace_ops);
 446
 447  out:
 448         mutex_unlock(&stack_sysctl_mutex);
 449         return ret;
 450 }
 451
 452 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
 453
 454 static __init int enable_stacktrace(char *str)
 455 {
 456         if (strncmp(str, "_filter=", 8) == 0)
 457                 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
 458
 459         stack_tracer_enabled = 1;
 460         last_stack_tracer_enabled = 1;
 461         return 1;
 462 }
 463 __setup("stacktrace", enable_stacktrace);
 464
 465 static __init int stack_trace_init(void)
 466 {
 467         struct dentry *d_tracer;
 468
 469         d_tracer = tracing_init_dentry();
 470         if (IS_ERR(d_tracer))
 471                 return 0;
 472
 473         trace_create_file("stack_max_size", 0644, d_tracer,
 474                         &stack_trace_max_size, &stack_max_size_fops);
 475
 476         trace_create_file("stack_trace", 0444, d_tracer,
 477                         NULL, &stack_trace_fops);
 478
 479 #ifdef CONFIG_DYNAMIC_FTRACE
 480         trace_create_file("stack_trace_filter", 0644, d_tracer,
 481                           &trace_ops, &stack_trace_filter_fops);
 482 #endif
 483
 484         if (stack_trace_filter_buf[0])
 485                 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
 486
 487         if (stack_tracer_enabled)
 488                 register_ftrace_function(&trace_ops);
 489
 490         return 0;
 491 }
 492
 493 device_initcall(stack_trace_init);