kernel/trace/trace_syscalls.c

   1 #include <trace/syscall.h>
   2 #include <trace/events/syscalls.h>
   3 #include <linux/kernel.h>
   4 #include <linux/ftrace.h>
   5 #include <linux/perf_event.h>
   6 #include <asm/syscall.h>
   7
   8 #include "trace_output.h"
   9 #include "trace.h"
  10
  11 static DEFINE_MUTEX(syscall_trace_lock);
  12 static int sys_refcount_enter;
  13 static int sys_refcount_exit;
  14 static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
  15 static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
  16
  17 enum print_line_t
  18 print_syscall_enter(struct trace_iterator *iter, int flags)
  19 {
  20         struct trace_seq *s = &iter->seq;
  21         struct trace_entry *ent = iter->ent;
  22         struct syscall_trace_enter *trace;
  23         struct syscall_metadata *entry;
  24         int i, ret, syscall;
  25
  26         trace = (typeof(trace))ent;
  27         syscall = trace->nr;
  28         entry = syscall_nr_to_meta(syscall);
  29
  30         if (!entry)
  31                 goto end;
  32
  33         if (entry->enter_id != ent->type) {
  34                 WARN_ON_ONCE(1);
  35                 goto end;
  36         }
  37
  38         ret = trace_seq_printf(s, "%s(", entry->name);
  39         if (!ret)
  40                 return TRACE_TYPE_PARTIAL_LINE;
  41
  42         for (i = 0; i < entry->nb_args; i++) {
  43                 /* parameter types */
  44                 if (trace_flags & TRACE_ITER_VERBOSE) {
  45                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
  46                         if (!ret)
  47                                 return TRACE_TYPE_PARTIAL_LINE;
  48                 }
  49                 /* parameter values */
  50                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
  51                                        trace->args[i],
  52                                        i == entry->nb_args - 1 ? "" : ", ");
  53                 if (!ret)
  54                         return TRACE_TYPE_PARTIAL_LINE;
  55         }
  56
  57         ret = trace_seq_putc(s, ')');
  58         if (!ret)
  59                 return TRACE_TYPE_PARTIAL_LINE;
  60
  61 end:
  62         ret =  trace_seq_putc(s, '\n');
  63         if (!ret)
  64                 return TRACE_TYPE_PARTIAL_LINE;
  65
  66         return TRACE_TYPE_HANDLED;
  67 }
  68
  69 enum print_line_t
  70 print_syscall_exit(struct trace_iterator *iter, int flags)
  71 {
  72         struct trace_seq *s = &iter->seq;
  73         struct trace_entry *ent = iter->ent;
  74         struct syscall_trace_exit *trace;
  75         int syscall;
  76         struct syscall_metadata *entry;
  77         int ret;
  78
  79         trace = (typeof(trace))ent;
  80         syscall = trace->nr;
  81         entry = syscall_nr_to_meta(syscall);
  82
  83         if (!entry) {
  84                 trace_seq_printf(s, "\n");
  85                 return TRACE_TYPE_HANDLED;
  86         }
  87
  88         if (entry->exit_id != ent->type) {
  89                 WARN_ON_ONCE(1);
  90                 return TRACE_TYPE_UNHANDLED;
  91         }
  92
  93         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
  94                                 trace->ret);
  95         if (!ret)
  96                 return TRACE_TYPE_PARTIAL_LINE;
  97
  98         return TRACE_TYPE_HANDLED;
  99 }
 100
 101 extern char *__bad_type_size(void);
 102
 103 #define SYSCALL_FIELD(type, name)                                       \
 104         sizeof(type) != sizeof(trace.name) ?                            \
 105                 __bad_type_size() :                                     \
 106                 #type, #name, offsetof(typeof(trace), name), sizeof(trace.name)
 107
 108 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
 109 {
 110         int i;
 111         int nr;
 112         int ret;
 113         struct syscall_metadata *entry;
 114         struct syscall_trace_enter trace;
 115         int offset = offsetof(struct syscall_trace_enter, args);
 116
 117         nr = syscall_name_to_nr(call->data);
 118         entry = syscall_nr_to_meta(nr);
 119
 120         if (!entry)
 121                 return 0;
 122
 123         ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n",
 124                                SYSCALL_FIELD(int, nr));
 125         if (!ret)
 126                 return 0;
 127
 128         for (i = 0; i < entry->nb_args; i++) {
 129                 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
 130                                         entry->args[i]);
 131                 if (!ret)
 132                         return 0;
 133                 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;\n", offset,
 134                                        sizeof(unsigned long));
 135                 if (!ret)
 136                         return 0;
 137                 offset += sizeof(unsigned long);
 138         }
 139
 140         trace_seq_puts(s, "\nprint fmt: \"");
 141         for (i = 0; i < entry->nb_args; i++) {
 142                 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
 143                                         sizeof(unsigned long),
 144                                         i == entry->nb_args - 1 ? "" : ", ");
 145                 if (!ret)
 146                         return 0;
 147         }
 148         trace_seq_putc(s, '"');
 149
 150         for (i = 0; i < entry->nb_args; i++) {
 151                 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
 152                                        entry->args[i]);
 153                 if (!ret)
 154                         return 0;
 155         }
 156
 157         return trace_seq_putc(s, '\n');
 158 }
 159
 160 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
 161 {
 162         int ret;
 163         struct syscall_trace_exit trace;
 164
 165         ret = trace_seq_printf(s,
 166                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 167                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n",
 168                                SYSCALL_FIELD(int, nr),
 169                                SYSCALL_FIELD(unsigned long, ret));
 170         if (!ret)
 171                 return 0;
 172
 173         return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
 174 }
 175
 176 int syscall_enter_define_fields(struct ftrace_event_call *call)
 177 {
 178         struct syscall_trace_enter trace;
 179         struct syscall_metadata *meta;
 180         int ret;
 181         int nr;
 182         int i;
 183         int offset = offsetof(typeof(trace), args);
 184
 185         nr = syscall_name_to_nr(call->data);
 186         meta = syscall_nr_to_meta(nr);
 187
 188         if (!meta)
 189                 return 0;
 190
 191         ret = trace_define_common_fields(call);
 192         if (ret)
 193                 return ret;
 194
 195         for (i = 0; i < meta->nb_args; i++) {
 196                 ret = trace_define_field(call, meta->types[i],
 197                                          meta->args[i], offset,
 198                                          sizeof(unsigned long), 0,
 199                                          FILTER_OTHER);
 200                 offset += sizeof(unsigned long);
 201         }
 202
 203         return ret;
 204 }
 205
 206 int syscall_exit_define_fields(struct ftrace_event_call *call)
 207 {
 208         struct syscall_trace_exit trace;
 209         int ret;
 210
 211         ret = trace_define_common_fields(call);
 212         if (ret)
 213                 return ret;
 214
 215         ret = trace_define_field(call, SYSCALL_FIELD(unsigned long, ret), 0,
 216                                  FILTER_OTHER);
 217
 218         return ret;
 219 }
 220
 221 void ftrace_syscall_enter(struct pt_regs *regs, long id)
 222 {
 223         struct syscall_trace_enter *entry;
 224         struct syscall_metadata *sys_data;
 225         struct ring_buffer_event *event;
 226         struct ring_buffer *buffer;
 227         int size;
 228         int syscall_nr;
 229
 230         syscall_nr = syscall_get_nr(current, regs);
 231         if (syscall_nr < 0)
 232                 return;
 233         if (!test_bit(syscall_nr, enabled_enter_syscalls))
 234                 return;
 235
 236         sys_data = syscall_nr_to_meta(syscall_nr);
 237         if (!sys_data)
 238                 return;
 239
 240         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 241
 242         event = trace_current_buffer_lock_reserve(&buffer, sys_data->enter_id,
 243                                                   size, 0, 0);
 244         if (!event)
 245                 return;
 246
 247         entry = ring_buffer_event_data(event);
 248         entry->nr = syscall_nr;
 249         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 250
 251         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 252                                           entry, event))
 253                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 254 }
 255
 256 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
 257 {
 258         struct syscall_trace_exit *entry;
 259         struct syscall_metadata *sys_data;
 260         struct ring_buffer_event *event;
 261         struct ring_buffer *buffer;
 262         int syscall_nr;
 263
 264         syscall_nr = syscall_get_nr(current, regs);
 265         if (syscall_nr < 0)
 266                 return;
 267         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 268                 return;
 269
 270         sys_data = syscall_nr_to_meta(syscall_nr);
 271         if (!sys_data)
 272                 return;
 273
 274         event = trace_current_buffer_lock_reserve(&buffer, sys_data->exit_id,
 275                                 sizeof(*entry), 0, 0);
 276         if (!event)
 277                 return;
 278
 279         entry = ring_buffer_event_data(event);
 280         entry->nr = syscall_nr;
 281         entry->ret = syscall_get_return_value(current, regs);
 282
 283         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 284                                           entry, event))
 285                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 286 }
 287
 288 int reg_event_syscall_enter(void *ptr)
 289 {
 290         int ret = 0;
 291         int num;
 292         char *name;
 293
 294         name = (char *)ptr;
 295         num = syscall_name_to_nr(name);
 296         if (num < 0 || num >= NR_syscalls)
 297                 return -ENOSYS;
 298         mutex_lock(&syscall_trace_lock);
 299         if (!sys_refcount_enter)
 300                 ret = register_trace_sys_enter(ftrace_syscall_enter);
 301         if (ret) {
 302                 pr_info("event trace: Could not activate"
 303                                 "syscall entry trace point");
 304         } else {
 305                 set_bit(num, enabled_enter_syscalls);
 306                 sys_refcount_enter++;
 307         }
 308         mutex_unlock(&syscall_trace_lock);
 309         return ret;
 310 }
 311
 312 void unreg_event_syscall_enter(void *ptr)
 313 {
 314         int num;
 315         char *name;
 316
 317         name = (char *)ptr;
 318         num = syscall_name_to_nr(name);
 319         if (num < 0 || num >= NR_syscalls)
 320                 return;
 321         mutex_lock(&syscall_trace_lock);
 322         sys_refcount_enter--;
 323         clear_bit(num, enabled_enter_syscalls);
 324         if (!sys_refcount_enter)
 325                 unregister_trace_sys_enter(ftrace_syscall_enter);
 326         mutex_unlock(&syscall_trace_lock);
 327 }
 328
 329 int reg_event_syscall_exit(void *ptr)
 330 {
 331         int ret = 0;
 332         int num;
 333         char *name;
 334
 335         name = (char *)ptr;
 336         num = syscall_name_to_nr(name);
 337         if (num < 0 || num >= NR_syscalls)
 338                 return -ENOSYS;
 339         mutex_lock(&syscall_trace_lock);
 340         if (!sys_refcount_exit)
 341                 ret = register_trace_sys_exit(ftrace_syscall_exit);
 342         if (ret) {
 343                 pr_info("event trace: Could not activate"
 344                                 "syscall exit trace point");
 345         } else {
 346                 set_bit(num, enabled_exit_syscalls);
 347                 sys_refcount_exit++;
 348         }
 349         mutex_unlock(&syscall_trace_lock);
 350         return ret;
 351 }
 352
 353 void unreg_event_syscall_exit(void *ptr)
 354 {
 355         int num;
 356         char *name;
 357
 358         name = (char *)ptr;
 359         num = syscall_name_to_nr(name);
 360         if (num < 0 || num >= NR_syscalls)
 361                 return;
 362         mutex_lock(&syscall_trace_lock);
 363         sys_refcount_exit--;
 364         clear_bit(num, enabled_exit_syscalls);
 365         if (!sys_refcount_exit)
 366                 unregister_trace_sys_exit(ftrace_syscall_exit);
 367         mutex_unlock(&syscall_trace_lock);
 368 }
 369
 370 struct trace_event event_syscall_enter = {
 371         .trace                  = print_syscall_enter,
 372 };
 373
 374 struct trace_event event_syscall_exit = {
 375         .trace                  = print_syscall_exit,
 376 };
 377
 378 #ifdef CONFIG_EVENT_PROFILE
 379
 380 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 381 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 382 static int sys_prof_refcount_enter;
 383 static int sys_prof_refcount_exit;
 384
 385 static void prof_syscall_enter(struct pt_regs *regs, long id)
 386 {
 387         struct syscall_metadata *sys_data;
 388         struct syscall_trace_enter *rec;
 389         unsigned long flags;
 390         char *raw_data;
 391         int syscall_nr;
 392         int size;
 393         int cpu;
 394
 395         syscall_nr = syscall_get_nr(current, regs);
 396         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 397                 return;
 398
 399         sys_data = syscall_nr_to_meta(syscall_nr);
 400         if (!sys_data)
 401                 return;
 402
 403         /* get the size after alignment with the u32 buffer size field */
 404         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 405         size = ALIGN(size + sizeof(u32), sizeof(u64));
 406         size -= sizeof(u32);
 407
 408         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 409                       "profile buffer not large enough"))
 410                 return;
 411
 412         /* Protect the per cpu buffer, begin the rcu read side */
 413         local_irq_save(flags);
 414
 415         cpu = smp_processor_id();
 416
 417         if (in_nmi())
 418                 raw_data = rcu_dereference(trace_profile_buf_nmi);
 419         else
 420                 raw_data = rcu_dereference(trace_profile_buf);
 421
 422         if (!raw_data)
 423                 goto end;
 424
 425         raw_data = per_cpu_ptr(raw_data, cpu);
 426
 427         /* zero the dead bytes from align to not leak stack to user */
 428         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 429
 430         rec = (struct syscall_trace_enter *) raw_data;
 431         tracing_generic_entry_update(&rec->ent, 0, 0);
 432         rec->ent.type = sys_data->enter_id;
 433         rec->nr = syscall_nr;
 434         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 435                                (unsigned long *)&rec->args);
 436         perf_tp_event(sys_data->enter_id, 0, 1, rec, size);
 437
 438 end:
 439         local_irq_restore(flags);
 440 }
 441
 442 int reg_prof_syscall_enter(char *name)
 443 {
 444         int ret = 0;
 445         int num;
 446
 447         num = syscall_name_to_nr(name);
 448         if (num < 0 || num >= NR_syscalls)
 449                 return -ENOSYS;
 450
 451         mutex_lock(&syscall_trace_lock);
 452         if (!sys_prof_refcount_enter)
 453                 ret = register_trace_sys_enter(prof_syscall_enter);
 454         if (ret) {
 455                 pr_info("event trace: Could not activate"
 456                                 "syscall entry trace point");
 457         } else {
 458                 set_bit(num, enabled_prof_enter_syscalls);
 459                 sys_prof_refcount_enter++;
 460         }
 461         mutex_unlock(&syscall_trace_lock);
 462         return ret;
 463 }
 464
 465 void unreg_prof_syscall_enter(char *name)
 466 {
 467         int num;
 468
 469         num = syscall_name_to_nr(name);
 470         if (num < 0 || num >= NR_syscalls)
 471                 return;
 472
 473         mutex_lock(&syscall_trace_lock);
 474         sys_prof_refcount_enter--;
 475         clear_bit(num, enabled_prof_enter_syscalls);
 476         if (!sys_prof_refcount_enter)
 477                 unregister_trace_sys_enter(prof_syscall_enter);
 478         mutex_unlock(&syscall_trace_lock);
 479 }
 480
 481 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 482 {
 483         struct syscall_metadata *sys_data;
 484         struct syscall_trace_exit *rec;
 485         unsigned long flags;
 486         int syscall_nr;
 487         char *raw_data;
 488         int size;
 489         int cpu;
 490
 491         syscall_nr = syscall_get_nr(current, regs);
 492         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 493                 return;
 494
 495         sys_data = syscall_nr_to_meta(syscall_nr);
 496         if (!sys_data)
 497                 return;
 498
 499         /* We can probably do that at build time */
 500         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 501         size -= sizeof(u32);
 502
 503         /*
 504          * Impossible, but be paranoid with the future
 505          * How to put this check outside runtime?
 506          */
 507         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 508                 "exit event has grown above profile buffer size"))
 509                 return;
 510
 511         /* Protect the per cpu buffer, begin the rcu read side */
 512         local_irq_save(flags);
 513         cpu = smp_processor_id();
 514
 515         if (in_nmi())
 516                 raw_data = rcu_dereference(trace_profile_buf_nmi);
 517         else
 518                 raw_data = rcu_dereference(trace_profile_buf);
 519
 520         if (!raw_data)
 521                 goto end;
 522
 523         raw_data = per_cpu_ptr(raw_data, cpu);
 524
 525         /* zero the dead bytes from align to not leak stack to user */
 526         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 527
 528         rec = (struct syscall_trace_exit *)raw_data;
 529
 530         tracing_generic_entry_update(&rec->ent, 0, 0);
 531         rec->ent.type = sys_data->exit_id;
 532         rec->nr = syscall_nr;
 533         rec->ret = syscall_get_return_value(current, regs);
 534
 535         perf_tp_event(sys_data->exit_id, 0, 1, rec, size);
 536
 537 end:
 538         local_irq_restore(flags);
 539 }
 540
 541 int reg_prof_syscall_exit(char *name)
 542 {
 543         int ret = 0;
 544         int num;
 545
 546         num = syscall_name_to_nr(name);
 547         if (num < 0 || num >= NR_syscalls)
 548                 return -ENOSYS;
 549
 550         mutex_lock(&syscall_trace_lock);
 551         if (!sys_prof_refcount_exit)
 552                 ret = register_trace_sys_exit(prof_syscall_exit);
 553         if (ret) {
 554                 pr_info("event trace: Could not activate"
 555                                 "syscall entry trace point");
 556         } else {
 557                 set_bit(num, enabled_prof_exit_syscalls);
 558                 sys_prof_refcount_exit++;
 559         }
 560         mutex_unlock(&syscall_trace_lock);
 561         return ret;
 562 }
 563
 564 void unreg_prof_syscall_exit(char *name)
 565 {
 566         int num;
 567
 568         num = syscall_name_to_nr(name);
 569         if (num < 0 || num >= NR_syscalls)
 570                 return;
 571
 572         mutex_lock(&syscall_trace_lock);
 573         sys_prof_refcount_exit--;
 574         clear_bit(num, enabled_prof_exit_syscalls);
 575         if (!sys_prof_refcount_exit)
 576                 unregister_trace_sys_exit(prof_syscall_exit);
 577         mutex_unlock(&syscall_trace_lock);
 578 }
 579
 580 #endif
 581
 582