kernel/trace/trace_syscalls.c

   1 #include <trace/syscall.h>
   2 #include <trace/events/syscalls.h>
   3 #include <linux/syscalls.h>
   4 #include <linux/slab.h>
   5 #include <linux/kernel.h>
   6 #include <linux/module.h>       /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */
   7 #include <linux/ftrace.h>
   8 #include <linux/perf_event.h>
   9 #include <asm/syscall.h>
  10
  11 #include "trace_output.h"
  12 #include "trace.h"
  13
  14 static DEFINE_MUTEX(syscall_trace_lock);
  15
  16 static int syscall_enter_register(struct ftrace_event_call *event,
  17                                  enum trace_reg type, void *data);
  18 static int syscall_exit_register(struct ftrace_event_call *event,
  19                                  enum trace_reg type, void *data);
  20
  21 static struct list_head *
  22 syscall_get_enter_fields(struct ftrace_event_call *call)
  23 {
  24         struct syscall_metadata *entry = call->data;
  25
  26         return &entry->enter_fields;
  27 }
  28
  29 extern struct syscall_metadata *__start_syscalls_metadata[];
  30 extern struct syscall_metadata *__stop_syscalls_metadata[];
  31
  32 static struct syscall_metadata **syscalls_metadata;
  33
  34 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
  35 static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
  36 {
  37         /*
  38          * Only compare after the "sys" prefix. Archs that use
  39          * syscall wrappers may have syscalls symbols aliases prefixed
  40          * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
  41          * mismatch.
  42          */
  43         return !strcmp(sym + 3, name + 3);
  44 }
  45 #endif
  46
  47 #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
  48 /*
  49  * Some architectures that allow for 32bit applications
  50  * to run on a 64bit kernel, do not map the syscalls for
  51  * the 32bit tasks the same as they do for 64bit tasks.
  52  *
  53  *     *cough*x86*cough*
  54  *
  55  * In such a case, instead of reporting the wrong syscalls,
  56  * simply ignore them.
  57  *
  58  * For an arch to ignore the compat syscalls it needs to
  59  * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
  60  * define the function arch_trace_is_compat_syscall() to let
  61  * the tracing system know that it should ignore it.
  62  */
  63 static int
  64 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
  65 {
  66         if (unlikely(arch_trace_is_compat_syscall(regs)))
  67                 return -1;
  68
  69         return syscall_get_nr(task, regs);
  70 }
  71 #else
  72 static inline int
  73 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
  74 {
  75         return syscall_get_nr(task, regs);
  76 }
  77 #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
  78
  79 static __init struct syscall_metadata *
  80 find_syscall_meta(unsigned long syscall)
  81 {
  82         struct syscall_metadata **start;
  83         struct syscall_metadata **stop;
  84         char str[KSYM_SYMBOL_LEN];
  85
  86
  87         start = __start_syscalls_metadata;
  88         stop = __stop_syscalls_metadata;
  89         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
  90
  91         if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
  92                 return NULL;
  93
  94         for ( ; start < stop; start++) {
  95                 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
  96                         return *start;
  97         }
  98         return NULL;
  99 }
 100
 101 static struct syscall_metadata *syscall_nr_to_meta(int nr)
 102 {
 103         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
 104                 return NULL;
 105
 106         return syscalls_metadata[nr];
 107 }
 108
 109 static enum print_line_t
 110 print_syscall_enter(struct trace_iterator *iter, int flags,
 111                     struct trace_event *event)
 112 {
 113         struct trace_seq *s = &iter->seq;
 114         struct trace_entry *ent = iter->ent;
 115         struct syscall_trace_enter *trace;
 116         struct syscall_metadata *entry;
 117         int i, ret, syscall;
 118
 119         trace = (typeof(trace))ent;
 120         syscall = trace->nr;
 121         entry = syscall_nr_to_meta(syscall);
 122
 123         if (!entry)
 124                 goto end;
 125
 126         if (entry->enter_event->event.type != ent->type) {
 127                 WARN_ON_ONCE(1);
 128                 goto end;
 129         }
 130
 131         ret = trace_seq_printf(s, "%s(", entry->name);
 132         if (!ret)
 133                 return TRACE_TYPE_PARTIAL_LINE;
 134
 135         for (i = 0; i < entry->nb_args; i++) {
 136                 /* parameter types */
 137                 if (trace_flags & TRACE_ITER_VERBOSE) {
 138                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
 139                         if (!ret)
 140                                 return TRACE_TYPE_PARTIAL_LINE;
 141                 }
 142                 /* parameter values */
 143                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 144                                        trace->args[i],
 145                                        i == entry->nb_args - 1 ? "" : ", ");
 146                 if (!ret)
 147                         return TRACE_TYPE_PARTIAL_LINE;
 148         }
 149
 150         ret = trace_seq_putc(s, ')');
 151         if (!ret)
 152                 return TRACE_TYPE_PARTIAL_LINE;
 153
 154 end:
 155         ret =  trace_seq_putc(s, '\n');
 156         if (!ret)
 157                 return TRACE_TYPE_PARTIAL_LINE;
 158
 159         return TRACE_TYPE_HANDLED;
 160 }
 161
 162 static enum print_line_t
 163 print_syscall_exit(struct trace_iterator *iter, int flags,
 164                    struct trace_event *event)
 165 {
 166         struct trace_seq *s = &iter->seq;
 167         struct trace_entry *ent = iter->ent;
 168         struct syscall_trace_exit *trace;
 169         int syscall;
 170         struct syscall_metadata *entry;
 171         int ret;
 172
 173         trace = (typeof(trace))ent;
 174         syscall = trace->nr;
 175         entry = syscall_nr_to_meta(syscall);
 176
 177         if (!entry) {
 178                 trace_seq_putc(s, '\n');
 179                 return TRACE_TYPE_HANDLED;
 180         }
 181
 182         if (entry->exit_event->event.type != ent->type) {
 183                 WARN_ON_ONCE(1);
 184                 return TRACE_TYPE_UNHANDLED;
 185         }
 186
 187         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 188                                 trace->ret);
 189         if (!ret)
 190                 return TRACE_TYPE_PARTIAL_LINE;
 191
 192         return TRACE_TYPE_HANDLED;
 193 }
 194
 195 extern char *__bad_type_size(void);
 196
 197 #define SYSCALL_FIELD(type, name)                                       \
 198         sizeof(type) != sizeof(trace.name) ?                            \
 199                 __bad_type_size() :                                     \
 200                 #type, #name, offsetof(typeof(trace), name),            \
 201                 sizeof(trace.name), is_signed_type(type)
 202
 203 static int __init
 204 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 205 {
 206         int i;
 207         int pos = 0;
 208
 209         /* When len=0, we just calculate the needed length */
 210 #define LEN_OR_ZERO (len ? len - pos : 0)
 211
 212         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 213         for (i = 0; i < entry->nb_args; i++) {
 214                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 215                                 entry->args[i], sizeof(unsigned long),
 216                                 i == entry->nb_args - 1 ? "" : ", ");
 217         }
 218         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 219
 220         for (i = 0; i < entry->nb_args; i++) {
 221                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 222                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 223         }
 224
 225 #undef LEN_OR_ZERO
 226
 227         /* return the length of print_fmt */
 228         return pos;
 229 }
 230
 231 static int __init set_syscall_print_fmt(struct ftrace_event_call *call)
 232 {
 233         char *print_fmt;
 234         int len;
 235         struct syscall_metadata *entry = call->data;
 236
 237         if (entry->enter_event != call) {
 238                 call->print_fmt = "\"0x%lx\", REC->ret";
 239                 return 0;
 240         }
 241
 242         /* First: called with 0 length to calculate the needed length */
 243         len = __set_enter_print_fmt(entry, NULL, 0);
 244
 245         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 246         if (!print_fmt)
 247                 return -ENOMEM;
 248
 249         /* Second: actually write the @print_fmt */
 250         __set_enter_print_fmt(entry, print_fmt, len + 1);
 251         call->print_fmt = print_fmt;
 252
 253         return 0;
 254 }
 255
 256 static void __init free_syscall_print_fmt(struct ftrace_event_call *call)
 257 {
 258         struct syscall_metadata *entry = call->data;
 259
 260         if (entry->enter_event == call)
 261                 kfree(call->print_fmt);
 262 }
 263
 264 static int __init syscall_enter_define_fields(struct ftrace_event_call *call)
 265 {
 266         struct syscall_trace_enter trace;
 267         struct syscall_metadata *meta = call->data;
 268         int ret;
 269         int i;
 270         int offset = offsetof(typeof(trace), args);
 271
 272         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 273         if (ret)
 274                 return ret;
 275
 276         for (i = 0; i < meta->nb_args; i++) {
 277                 ret = trace_define_field(call, meta->types[i],
 278                                          meta->args[i], offset,
 279                                          sizeof(unsigned long), 0,
 280                                          FILTER_OTHER);
 281                 offset += sizeof(unsigned long);
 282         }
 283
 284         return ret;
 285 }
 286
 287 static int __init syscall_exit_define_fields(struct ftrace_event_call *call)
 288 {
 289         struct syscall_trace_exit trace;
 290         int ret;
 291
 292         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 293         if (ret)
 294                 return ret;
 295
 296         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 297                                  FILTER_OTHER);
 298
 299         return ret;
 300 }
 301
 302 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 303 {
 304         struct trace_array *tr = data;
 305         struct ftrace_event_file *ftrace_file;
 306         struct syscall_trace_enter *entry;
 307         struct syscall_metadata *sys_data;
 308         struct ring_buffer_event *event;
 309         struct ring_buffer *buffer;
 310         unsigned long irq_flags;
 311         int pc;
 312         int syscall_nr;
 313         int size;
 314
 315         syscall_nr = trace_get_syscall_nr(current, regs);
 316         if (syscall_nr < 0)
 317                 return;
 318
 319         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
 320         ftrace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
 321         if (!ftrace_file)
 322                 return;
 323
 324         if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
 325                 return;
 326
 327         sys_data = syscall_nr_to_meta(syscall_nr);
 328         if (!sys_data)
 329                 return;
 330
 331         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 332
 333         local_save_flags(irq_flags);
 334         pc = preempt_count();
 335
 336         buffer = tr->trace_buffer.buffer;
 337         event = trace_buffer_lock_reserve(buffer,
 338                         sys_data->enter_event->event.type, size, irq_flags, pc);
 339         if (!event)
 340                 return;
 341
 342         entry = ring_buffer_event_data(event);
 343         entry->nr = syscall_nr;
 344         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 345
 346         if (!filter_check_discard(ftrace_file, entry, buffer, event))
 347                 trace_current_buffer_unlock_commit(buffer, event,
 348                                                    irq_flags, pc);
 349 }
 350
 351 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 352 {
 353         struct trace_array *tr = data;
 354         struct ftrace_event_file *ftrace_file;
 355         struct syscall_trace_exit *entry;
 356         struct syscall_metadata *sys_data;
 357         struct ring_buffer_event *event;
 358         struct ring_buffer *buffer;
 359         unsigned long irq_flags;
 360         int pc;
 361         int syscall_nr;
 362
 363         syscall_nr = trace_get_syscall_nr(current, regs);
 364         if (syscall_nr < 0)
 365                 return;
 366
 367         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 368         ftrace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 369         if (!ftrace_file)
 370                 return;
 371
 372         if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &ftrace_file->flags))
 373                 return;
 374
 375         sys_data = syscall_nr_to_meta(syscall_nr);
 376         if (!sys_data)
 377                 return;
 378
 379         local_save_flags(irq_flags);
 380         pc = preempt_count();
 381
 382         buffer = tr->trace_buffer.buffer;
 383         event = trace_buffer_lock_reserve(buffer,
 384                         sys_data->exit_event->event.type, sizeof(*entry),
 385                         irq_flags, pc);
 386         if (!event)
 387                 return;
 388
 389         entry = ring_buffer_event_data(event);
 390         entry->nr = syscall_nr;
 391         entry->ret = syscall_get_return_value(current, regs);
 392
 393         if (!filter_check_discard(ftrace_file, entry, buffer, event))
 394                 trace_current_buffer_unlock_commit(buffer, event,
 395                                                    irq_flags, pc);
 396 }
 397
 398 static int reg_event_syscall_enter(struct ftrace_event_file *file,
 399                                    struct ftrace_event_call *call)
 400 {
 401         struct trace_array *tr = file->tr;
 402         int ret = 0;
 403         int num;
 404
 405         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 406         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 407                 return -ENOSYS;
 408         mutex_lock(&syscall_trace_lock);
 409         if (!tr->sys_refcount_enter)
 410                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 411         if (!ret) {
 412                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 413                 tr->sys_refcount_enter++;
 414         }
 415         mutex_unlock(&syscall_trace_lock);
 416         return ret;
 417 }
 418
 419 static void unreg_event_syscall_enter(struct ftrace_event_file *file,
 420                                       struct ftrace_event_call *call)
 421 {
 422         struct trace_array *tr = file->tr;
 423         int num;
 424
 425         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 426         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 427                 return;
 428         mutex_lock(&syscall_trace_lock);
 429         tr->sys_refcount_enter--;
 430         rcu_assign_pointer(tr->enter_syscall_files[num], NULL);
 431         if (!tr->sys_refcount_enter)
 432                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 433         mutex_unlock(&syscall_trace_lock);
 434         /*
 435          * Callers expect the event to be completely disabled on
 436          * return, so wait for current handlers to finish.
 437          */
 438         synchronize_sched();
 439 }
 440
 441 static int reg_event_syscall_exit(struct ftrace_event_file *file,
 442                                   struct ftrace_event_call *call)
 443 {
 444         struct trace_array *tr = file->tr;
 445         int ret = 0;
 446         int num;
 447
 448         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 449         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 450                 return -ENOSYS;
 451         mutex_lock(&syscall_trace_lock);
 452         if (!tr->sys_refcount_exit)
 453                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 454         if (!ret) {
 455                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 456                 tr->sys_refcount_exit++;
 457         }
 458         mutex_unlock(&syscall_trace_lock);
 459         return ret;
 460 }
 461
 462 static void unreg_event_syscall_exit(struct ftrace_event_file *file,
 463                                      struct ftrace_event_call *call)
 464 {
 465         struct trace_array *tr = file->tr;
 466         int num;
 467
 468         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 469         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 470                 return;
 471         mutex_lock(&syscall_trace_lock);
 472         tr->sys_refcount_exit--;
 473         rcu_assign_pointer(tr->exit_syscall_files[num], NULL);
 474         if (!tr->sys_refcount_exit)
 475                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 476         mutex_unlock(&syscall_trace_lock);
 477         /*
 478          * Callers expect the event to be completely disabled on
 479          * return, so wait for current handlers to finish.
 480          */
 481         synchronize_sched();
 482 }
 483
 484 static int __init init_syscall_trace(struct ftrace_event_call *call)
 485 {
 486         int id;
 487         int num;
 488
 489         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 490         if (num < 0 || num >= NR_syscalls) {
 491                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 492                                 ((struct syscall_metadata *)call->data)->name);
 493                 return -ENOSYS;
 494         }
 495
 496         if (set_syscall_print_fmt(call) < 0)
 497                 return -ENOMEM;
 498
 499         id = trace_event_raw_init(call);
 500
 501         if (id < 0) {
 502                 free_syscall_print_fmt(call);
 503                 return id;
 504         }
 505
 506         return id;
 507 }
 508
 509 struct trace_event_functions enter_syscall_print_funcs = {
 510         .trace          = print_syscall_enter,
 511 };
 512
 513 struct trace_event_functions exit_syscall_print_funcs = {
 514         .trace          = print_syscall_exit,
 515 };
 516
 517 struct ftrace_event_class __refdata event_class_syscall_enter = {
 518         .system         = "syscalls",
 519         .reg            = syscall_enter_register,
 520         .define_fields  = syscall_enter_define_fields,
 521         .get_fields     = syscall_get_enter_fields,
 522         .raw_init       = init_syscall_trace,
 523 };
 524
 525 struct ftrace_event_class __refdata event_class_syscall_exit = {
 526         .system         = "syscalls",
 527         .reg            = syscall_exit_register,
 528         .define_fields  = syscall_exit_define_fields,
 529         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 530         .raw_init       = init_syscall_trace,
 531 };
 532
 533 unsigned long __init __weak arch_syscall_addr(int nr)
 534 {
 535         return (unsigned long)sys_call_table[nr];
 536 }
 537
 538 static int __init init_ftrace_syscalls(void)
 539 {
 540         struct syscall_metadata *meta;
 541         unsigned long addr;
 542         int i;
 543
 544         syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
 545                                     GFP_KERNEL);
 546         if (!syscalls_metadata) {
 547                 WARN_ON(1);
 548                 return -ENOMEM;
 549         }
 550
 551         for (i = 0; i < NR_syscalls; i++) {
 552                 addr = arch_syscall_addr(i);
 553                 meta = find_syscall_meta(addr);
 554                 if (!meta)
 555                         continue;
 556
 557                 meta->syscall_nr = i;
 558                 syscalls_metadata[i] = meta;
 559         }
 560
 561         return 0;
 562 }
 563 early_initcall(init_ftrace_syscalls);
 564
 565 #ifdef CONFIG_PERF_EVENTS
 566
 567 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 568 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 569 static int sys_perf_refcount_enter;
 570 static int sys_perf_refcount_exit;
 571
 572 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 573 {
 574         struct syscall_metadata *sys_data;
 575         struct syscall_trace_enter *rec;
 576         struct hlist_head *head;
 577         int syscall_nr;
 578         int rctx;
 579         int size;
 580
 581         syscall_nr = trace_get_syscall_nr(current, regs);
 582         if (syscall_nr < 0)
 583                 return;
 584         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 585                 return;
 586
 587         sys_data = syscall_nr_to_meta(syscall_nr);
 588         if (!sys_data)
 589                 return;
 590
 591         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 592         if (hlist_empty(head))
 593                 return;
 594
 595         /* get the size after alignment with the u32 buffer size field */
 596         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 597         size = ALIGN(size + sizeof(u32), sizeof(u64));
 598         size -= sizeof(u32);
 599
 600         rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
 601                                 sys_data->enter_event->event.type, regs, &rctx);
 602         if (!rec)
 603                 return;
 604
 605         rec->nr = syscall_nr;
 606         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 607                                (unsigned long *)&rec->args);
 608         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 609 }
 610
 611 static int perf_sysenter_enable(struct ftrace_event_call *call)
 612 {
 613         int ret = 0;
 614         int num;
 615
 616         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 617
 618         mutex_lock(&syscall_trace_lock);
 619         if (!sys_perf_refcount_enter)
 620                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 621         if (ret) {
 622                 pr_info("event trace: Could not activate"
 623                                 "syscall entry trace point");
 624         } else {
 625                 set_bit(num, enabled_perf_enter_syscalls);
 626                 sys_perf_refcount_enter++;
 627         }
 628         mutex_unlock(&syscall_trace_lock);
 629         return ret;
 630 }
 631
 632 static void perf_sysenter_disable(struct ftrace_event_call *call)
 633 {
 634         int num;
 635
 636         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 637
 638         mutex_lock(&syscall_trace_lock);
 639         sys_perf_refcount_enter--;
 640         clear_bit(num, enabled_perf_enter_syscalls);
 641         if (!sys_perf_refcount_enter)
 642                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 643         mutex_unlock(&syscall_trace_lock);
 644 }
 645
 646 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 647 {
 648         struct syscall_metadata *sys_data;
 649         struct syscall_trace_exit *rec;
 650         struct hlist_head *head;
 651         int syscall_nr;
 652         int rctx;
 653         int size;
 654
 655         syscall_nr = trace_get_syscall_nr(current, regs);
 656         if (syscall_nr < 0)
 657                 return;
 658         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 659                 return;
 660
 661         sys_data = syscall_nr_to_meta(syscall_nr);
 662         if (!sys_data)
 663                 return;
 664
 665         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 666         if (hlist_empty(head))
 667                 return;
 668
 669         /* We can probably do that at build time */
 670         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 671         size -= sizeof(u32);
 672
 673         rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
 674                                 sys_data->exit_event->event.type, regs, &rctx);
 675         if (!rec)
 676                 return;
 677
 678         rec->nr = syscall_nr;
 679         rec->ret = syscall_get_return_value(current, regs);
 680         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 681 }
 682
 683 static int perf_sysexit_enable(struct ftrace_event_call *call)
 684 {
 685         int ret = 0;
 686         int num;
 687
 688         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 689
 690         mutex_lock(&syscall_trace_lock);
 691         if (!sys_perf_refcount_exit)
 692                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 693         if (ret) {
 694                 pr_info("event trace: Could not activate"
 695                                 "syscall exit trace point");
 696         } else {
 697                 set_bit(num, enabled_perf_exit_syscalls);
 698                 sys_perf_refcount_exit++;
 699         }
 700         mutex_unlock(&syscall_trace_lock);
 701         return ret;
 702 }
 703
 704 static void perf_sysexit_disable(struct ftrace_event_call *call)
 705 {
 706         int num;
 707
 708         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 709
 710         mutex_lock(&syscall_trace_lock);
 711         sys_perf_refcount_exit--;
 712         clear_bit(num, enabled_perf_exit_syscalls);
 713         if (!sys_perf_refcount_exit)
 714                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 715         mutex_unlock(&syscall_trace_lock);
 716 }
 717
 718 #endif /* CONFIG_PERF_EVENTS */
 719
 720 static int syscall_enter_register(struct ftrace_event_call *event,
 721                                  enum trace_reg type, void *data)
 722 {
 723         struct ftrace_event_file *file = data;
 724
 725         switch (type) {
 726         case TRACE_REG_REGISTER:
 727                 return reg_event_syscall_enter(file, event);
 728         case TRACE_REG_UNREGISTER:
 729                 unreg_event_syscall_enter(file, event);
 730                 return 0;
 731
 732 #ifdef CONFIG_PERF_EVENTS
 733         case TRACE_REG_PERF_REGISTER:
 734                 return perf_sysenter_enable(event);
 735         case TRACE_REG_PERF_UNREGISTER:
 736                 perf_sysenter_disable(event);
 737                 return 0;
 738         case TRACE_REG_PERF_OPEN:
 739         case TRACE_REG_PERF_CLOSE:
 740         case TRACE_REG_PERF_ADD:
 741         case TRACE_REG_PERF_DEL:
 742                 return 0;
 743 #endif
 744         }
 745         return 0;
 746 }
 747
 748 static int syscall_exit_register(struct ftrace_event_call *event,
 749                                  enum trace_reg type, void *data)
 750 {
 751         struct ftrace_event_file *file = data;
 752
 753         switch (type) {
 754         case TRACE_REG_REGISTER:
 755                 return reg_event_syscall_exit(file, event);
 756         case TRACE_REG_UNREGISTER:
 757                 unreg_event_syscall_exit(file, event);
 758                 return 0;
 759
 760 #ifdef CONFIG_PERF_EVENTS
 761         case TRACE_REG_PERF_REGISTER:
 762                 return perf_sysexit_enable(event);
 763         case TRACE_REG_PERF_UNREGISTER:
 764                 perf_sysexit_disable(event);
 765                 return 0;
 766         case TRACE_REG_PERF_OPEN:
 767         case TRACE_REG_PERF_CLOSE:
 768         case TRACE_REG_PERF_ADD:
 769         case TRACE_REG_PERF_DEL:
 770                 return 0;
 771 #endif
 772         }
 773         return 0;
 774 }