kernel/trace/trace_stack.c

   1 /*
   2  * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
   3  *
   4  */
   5 #include <linux/stacktrace.h>
   6 #include <linux/kallsyms.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/uaccess.h>
  10 #include <linux/debugfs.h>
  11 #include <linux/ftrace.h>
  12 #include <linux/module.h>
  13 #include <linux/sysctl.h>
  14 #include <linux/init.h>
  15 #include <linux/fs.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+1] =
  24          { [0 ... (STACK_TRACE_ENTRIES)] = ULONG_MAX };
  25 static unsigned stack_dump_index[STACK_TRACE_ENTRIES];
  26
  27 static struct stack_trace max_stack_trace = {
  28         .max_entries            = STACK_TRACE_ENTRIES,
  29         .entries                = stack_dump_trace,
  30 };
  31
  32 static unsigned long max_stack_size;
  33 static arch_spinlock_t max_stack_lock =
  34         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
  35
  36 static int stack_trace_disabled __read_mostly;
  37 static DEFINE_PER_CPU(int, trace_active);
  38 static DEFINE_MUTEX(stack_sysctl_mutex);
  39
  40 int stack_tracer_enabled;
  41 static int last_stack_tracer_enabled;
  42
  43 static inline void check_stack(void)
  44 {
  45         unsigned long this_size, flags;
  46         unsigned long *p, *top, *start;
  47         int i;
  48
  49         this_size = ((unsigned long)&this_size) & (THREAD_SIZE-1);
  50         this_size = THREAD_SIZE - this_size;
  51
  52         if (this_size <= max_stack_size)
  53                 return;
  54
  55         /* we do not handle interrupt stacks yet */
  56         if (!object_is_on_stack(&this_size))
  57                 return;
  58
  59         local_irq_save(flags);
  60         arch_spin_lock(&max_stack_lock);
  61
  62         /* a race could have already updated it */
  63         if (this_size <= max_stack_size)
  64                 goto out;
  65
  66         max_stack_size = this_size;
  67
  68         max_stack_trace.nr_entries      = 0;
  69         max_stack_trace.skip            = 3;
  70
  71         save_stack_trace(&max_stack_trace);
  72
  73         /*
  74          * Now find where in the stack these are.
  75          */
  76         i = 0;
  77         start = &this_size;
  78         top = (unsigned long *)
  79                 (((unsigned long)start & ~(THREAD_SIZE-1)) + THREAD_SIZE);
  80
  81         /*
  82          * Loop through all the entries. One of the entries may
  83          * for some reason be missed on the stack, so we may
  84          * have to account for them. If they are all there, this
  85          * loop will only happen once. This code only takes place
  86          * on a new max, so it is far from a fast path.
  87          */
  88         while (i < max_stack_trace.nr_entries) {
  89                 int found = 0;
  90
  91                 stack_dump_index[i] = this_size;
  92                 p = start;
  93
  94                 for (; p < top && i < max_stack_trace.nr_entries; p++) {
  95                         if (*p == stack_dump_trace[i]) {
  96                                 this_size = stack_dump_index[i++] =
  97                                         (top - p) * sizeof(unsigned long);
  98                                 found = 1;
  99                                 /* Start the search from here */
 100                                 start = p + 1;
 101                         }
 102                 }
 103
 104                 if (!found)
 105                         i++;
 106         }
 107
 108  out:
 109         arch_spin_unlock(&max_stack_lock);
 110         local_irq_restore(flags);
 111 }
 112
 113 static void
 114 stack_trace_call(unsigned long ip, unsigned long parent_ip,
 115                  struct ftrace_ops *op, struct pt_regs *pt_regs)
 116 {
 117         int cpu;
 118
 119         if (unlikely(!ftrace_enabled || stack_trace_disabled))
 120                 return;
 121
 122         preempt_disable_notrace();
 123
 124         cpu = raw_smp_processor_id();
 125         /* no atomic needed, we only modify this variable by this cpu */
 126         if (per_cpu(trace_active, cpu)++ != 0)
 127                 goto out;
 128
 129         check_stack();
 130
 131  out:
 132         per_cpu(trace_active, cpu)--;
 133         /* prevent recursion in schedule */
 134         preempt_enable_notrace();
 135 }
 136
 137 static struct ftrace_ops trace_ops __read_mostly =
 138 {
 139         .func = stack_trace_call,
 140         .flags = FTRACE_OPS_FL_RECURSION_SAFE,
 141 };
 142
 143 static ssize_t
 144 stack_max_size_read(struct file *filp, char __user *ubuf,
 145                     size_t count, loff_t *ppos)
 146 {
 147         unsigned long *ptr = filp->private_data;
 148         char buf[64];
 149         int r;
 150
 151         r = snprintf(buf, sizeof(buf), "%ld\n", *ptr);
 152         if (r > sizeof(buf))
 153                 r = sizeof(buf);
 154         return simple_read_from_buffer(ubuf, count, ppos, buf, r);
 155 }
 156
 157 static ssize_t
 158 stack_max_size_write(struct file *filp, const char __user *ubuf,
 159                      size_t count, loff_t *ppos)
 160 {
 161         long *ptr = filp->private_data;
 162         unsigned long val, flags;
 163         int ret;
 164         int cpu;
 165
 166         ret = kstrtoul_from_user(ubuf, count, 10, &val);
 167         if (ret)
 168                 return ret;
 169
 170         local_irq_save(flags);
 171
 172         /*
 173          * In case we trace inside arch_spin_lock() or after (NMI),
 174          * we will cause circular lock, so we also need to increase
 175          * the percpu trace_active here.
 176          */
 177         cpu = smp_processor_id();
 178         per_cpu(trace_active, cpu)++;
 179
 180         arch_spin_lock(&max_stack_lock);
 181         *ptr = val;
 182         arch_spin_unlock(&max_stack_lock);
 183
 184         per_cpu(trace_active, cpu)--;
 185         local_irq_restore(flags);
 186
 187         return count;
 188 }
 189
 190 static const struct file_operations stack_max_size_fops = {
 191         .open           = tracing_open_generic,
 192         .read           = stack_max_size_read,
 193         .write          = stack_max_size_write,
 194         .llseek         = default_llseek,
 195 };
 196
 197 static void *
 198 __next(struct seq_file *m, loff_t *pos)
 199 {
 200         long n = *pos - 1;
 201
 202         if (n >= max_stack_trace.nr_entries || stack_dump_trace[n] == ULONG_MAX)
 203                 return NULL;
 204
 205         m->private = (void *)n;
 206         return &m->private;
 207 }
 208
 209 static void *
 210 t_next(struct seq_file *m, void *v, loff_t *pos)
 211 {
 212         (*pos)++;
 213         return __next(m, pos);
 214 }
 215
 216 static void *t_start(struct seq_file *m, loff_t *pos)
 217 {
 218         int cpu;
 219
 220         local_irq_disable();
 221
 222         cpu = smp_processor_id();
 223         per_cpu(trace_active, cpu)++;
 224
 225         arch_spin_lock(&max_stack_lock);
 226
 227         if (*pos == 0)
 228                 return SEQ_START_TOKEN;
 229
 230         return __next(m, pos);
 231 }
 232
 233 static void t_stop(struct seq_file *m, void *p)
 234 {
 235         int cpu;
 236
 237         arch_spin_unlock(&max_stack_lock);
 238
 239         cpu = smp_processor_id();
 240         per_cpu(trace_active, cpu)--;
 241
 242         local_irq_enable();
 243 }
 244
 245 static int trace_lookup_stack(struct seq_file *m, long i)
 246 {
 247         unsigned long addr = stack_dump_trace[i];
 248
 249         return seq_printf(m, "%pS\n", (void *)addr);
 250 }
 251
 252 static void print_disabled(struct seq_file *m)
 253 {
 254         seq_puts(m, "#\n"
 255                  "#  Stack tracer disabled\n"
 256                  "#\n"
 257                  "# To enable the stack tracer, either add 'stacktrace' to the\n"
 258                  "# kernel command line\n"
 259                  "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
 260                  "#\n");
 261 }
 262
 263 static int t_show(struct seq_file *m, void *v)
 264 {
 265         long i;
 266         int size;
 267
 268         if (v == SEQ_START_TOKEN) {
 269                 seq_printf(m, "        Depth    Size   Location"
 270                            "    (%d entries)\n"
 271                            "        -----    ----   --------\n",
 272                            max_stack_trace.nr_entries - 1);
 273
 274                 if (!stack_tracer_enabled && !max_stack_size)
 275                         print_disabled(m);
 276
 277                 return 0;
 278         }
 279
 280         i = *(long *)v;
 281
 282         if (i >= max_stack_trace.nr_entries ||
 283             stack_dump_trace[i] == ULONG_MAX)
 284                 return 0;
 285
 286         if (i+1 == max_stack_trace.nr_entries ||
 287             stack_dump_trace[i+1] == ULONG_MAX)
 288                 size = stack_dump_index[i];
 289         else
 290                 size = stack_dump_index[i] - stack_dump_index[i+1];
 291
 292         seq_printf(m, "%3ld) %8d   %5d   ", i, stack_dump_index[i], size);
 293
 294         trace_lookup_stack(m, i);
 295
 296         return 0;
 297 }
 298
 299 static const struct seq_operations stack_trace_seq_ops = {
 300         .start          = t_start,
 301         .next           = t_next,
 302         .stop           = t_stop,
 303         .show           = t_show,
 304 };
 305
 306 static int stack_trace_open(struct inode *inode, struct file *file)
 307 {
 308         return seq_open(file, &stack_trace_seq_ops);
 309 }
 310
 311 static const struct file_operations stack_trace_fops = {
 312         .open           = stack_trace_open,
 313         .read           = seq_read,
 314         .llseek         = seq_lseek,
 315         .release        = seq_release,
 316 };
 317
 318 static int
 319 stack_trace_filter_open(struct inode *inode, struct file *file)
 320 {
 321         return ftrace_regex_open(&trace_ops, FTRACE_ITER_FILTER,
 322                                  inode, file);
 323 }
 324
 325 static const struct file_operations stack_trace_filter_fops = {
 326         .open = stack_trace_filter_open,
 327         .read = seq_read,
 328         .write = ftrace_filter_write,
 329         .llseek = ftrace_regex_lseek,
 330         .release = ftrace_regex_release,
 331 };
 332
 333 int
 334 stack_trace_sysctl(struct ctl_table *table, int write,
 335                    void __user *buffer, size_t *lenp,
 336                    loff_t *ppos)
 337 {
 338         int ret;
 339
 340         mutex_lock(&stack_sysctl_mutex);
 341
 342         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 343
 344         if (ret || !write ||
 345             (last_stack_tracer_enabled == !!stack_tracer_enabled))
 346                 goto out;
 347
 348         last_stack_tracer_enabled = !!stack_tracer_enabled;
 349
 350         if (stack_tracer_enabled)
 351                 register_ftrace_function(&trace_ops);
 352         else
 353                 unregister_ftrace_function(&trace_ops);
 354
 355  out:
 356         mutex_unlock(&stack_sysctl_mutex);
 357         return ret;
 358 }
 359
 360 static char stack_trace_filter_buf[COMMAND_LINE_SIZE+1] __initdata;
 361
 362 static __init int enable_stacktrace(char *str)
 363 {
 364         if (strncmp(str, "_filter=", 8) == 0)
 365                 strncpy(stack_trace_filter_buf, str+8, COMMAND_LINE_SIZE);
 366
 367         stack_tracer_enabled = 1;
 368         last_stack_tracer_enabled = 1;
 369         return 1;
 370 }
 371 __setup("stacktrace", enable_stacktrace);
 372
 373 static __init int stack_trace_init(void)
 374 {
 375         struct dentry *d_tracer;
 376
 377         d_tracer = tracing_init_dentry();
 378
 379         trace_create_file("stack_max_size", 0644, d_tracer,
 380                         &max_stack_size, &stack_max_size_fops);
 381
 382         trace_create_file("stack_trace", 0444, d_tracer,
 383                         NULL, &stack_trace_fops);
 384
 385         trace_create_file("stack_trace_filter", 0444, d_tracer,
 386                         NULL, &stack_trace_filter_fops);
 387
 388         if (stack_trace_filter_buf[0])
 389                 ftrace_set_early_filter(&trace_ops, stack_trace_filter_buf, 1);
 390
 391         if (stack_tracer_enabled)
 392                 register_ftrace_function(&trace_ops);
 393
 394         return 0;
 395 }
 396
 397 device_initcall(stack_trace_init);