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 || syscall_nr >= NR_syscalls)
 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 (ftrace_trigger_soft_disabled(ftrace_file))
 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         event_trigger_unlock_commit(ftrace_file, buffer, event, entry,
 347                                     irq_flags, pc);
 348 }
 349
 350 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 351 {
 352         struct trace_array *tr = data;
 353         struct ftrace_event_file *ftrace_file;
 354         struct syscall_trace_exit *entry;
 355         struct syscall_metadata *sys_data;
 356         struct ring_buffer_event *event;
 357         struct ring_buffer *buffer;
 358         unsigned long irq_flags;
 359         int pc;
 360         int syscall_nr;
 361
 362         syscall_nr = trace_get_syscall_nr(current, regs);
 363         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 364                 return;
 365
 366         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 367         ftrace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 368         if (!ftrace_file)
 369                 return;
 370
 371         if (ftrace_trigger_soft_disabled(ftrace_file))
 372                 return;
 373
 374         sys_data = syscall_nr_to_meta(syscall_nr);
 375         if (!sys_data)
 376                 return;
 377
 378         local_save_flags(irq_flags);
 379         pc = preempt_count();
 380
 381         buffer = tr->trace_buffer.buffer;
 382         event = trace_buffer_lock_reserve(buffer,
 383                         sys_data->exit_event->event.type, sizeof(*entry),
 384                         irq_flags, pc);
 385         if (!event)
 386                 return;
 387
 388         entry = ring_buffer_event_data(event);
 389         entry->nr = syscall_nr;
 390         entry->ret = syscall_get_return_value(current, regs);
 391
 392         event_trigger_unlock_commit(ftrace_file, buffer, event, entry,
 393                                     irq_flags, pc);
 394 }
 395
 396 static int reg_event_syscall_enter(struct ftrace_event_file *file,
 397                                    struct ftrace_event_call *call)
 398 {
 399         struct trace_array *tr = file->tr;
 400         int ret = 0;
 401         int num;
 402
 403         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 404         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 405                 return -ENOSYS;
 406         mutex_lock(&syscall_trace_lock);
 407         if (!tr->sys_refcount_enter)
 408                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 409         if (!ret) {
 410                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 411                 tr->sys_refcount_enter++;
 412         }
 413         mutex_unlock(&syscall_trace_lock);
 414         return ret;
 415 }
 416
 417 static void unreg_event_syscall_enter(struct ftrace_event_file *file,
 418                                       struct ftrace_event_call *call)
 419 {
 420         struct trace_array *tr = file->tr;
 421         int num;
 422
 423         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 424         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 425                 return;
 426         mutex_lock(&syscall_trace_lock);
 427         tr->sys_refcount_enter--;
 428         RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
 429         if (!tr->sys_refcount_enter)
 430                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 431         mutex_unlock(&syscall_trace_lock);
 432 }
 433
 434 static int reg_event_syscall_exit(struct ftrace_event_file *file,
 435                                   struct ftrace_event_call *call)
 436 {
 437         struct trace_array *tr = file->tr;
 438         int ret = 0;
 439         int num;
 440
 441         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 442         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 443                 return -ENOSYS;
 444         mutex_lock(&syscall_trace_lock);
 445         if (!tr->sys_refcount_exit)
 446                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 447         if (!ret) {
 448                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 449                 tr->sys_refcount_exit++;
 450         }
 451         mutex_unlock(&syscall_trace_lock);
 452         return ret;
 453 }
 454
 455 static void unreg_event_syscall_exit(struct ftrace_event_file *file,
 456                                      struct ftrace_event_call *call)
 457 {
 458         struct trace_array *tr = file->tr;
 459         int num;
 460
 461         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 462         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 463                 return;
 464         mutex_lock(&syscall_trace_lock);
 465         tr->sys_refcount_exit--;
 466         RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
 467         if (!tr->sys_refcount_exit)
 468                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 469         mutex_unlock(&syscall_trace_lock);
 470 }
 471
 472 static int __init init_syscall_trace(struct ftrace_event_call *call)
 473 {
 474         int id;
 475         int num;
 476
 477         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 478         if (num < 0 || num >= NR_syscalls) {
 479                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 480                                 ((struct syscall_metadata *)call->data)->name);
 481                 return -ENOSYS;
 482         }
 483
 484         if (set_syscall_print_fmt(call) < 0)
 485                 return -ENOMEM;
 486
 487         id = trace_event_raw_init(call);
 488
 489         if (id < 0) {
 490                 free_syscall_print_fmt(call);
 491                 return id;
 492         }
 493
 494         return id;
 495 }
 496
 497 struct trace_event_functions enter_syscall_print_funcs = {
 498         .trace          = print_syscall_enter,
 499 };
 500
 501 struct trace_event_functions exit_syscall_print_funcs = {
 502         .trace          = print_syscall_exit,
 503 };
 504
 505 struct ftrace_event_class __refdata event_class_syscall_enter = {
 506         .system         = "syscalls",
 507         .reg            = syscall_enter_register,
 508         .define_fields  = syscall_enter_define_fields,
 509         .get_fields     = syscall_get_enter_fields,
 510         .raw_init       = init_syscall_trace,
 511 };
 512
 513 struct ftrace_event_class __refdata event_class_syscall_exit = {
 514         .system         = "syscalls",
 515         .reg            = syscall_exit_register,
 516         .define_fields  = syscall_exit_define_fields,
 517         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 518         .raw_init       = init_syscall_trace,
 519 };
 520
 521 unsigned long __init __weak arch_syscall_addr(int nr)
 522 {
 523         return (unsigned long)sys_call_table[nr];
 524 }
 525
 526 static int __init init_ftrace_syscalls(void)
 527 {
 528         struct syscall_metadata *meta;
 529         unsigned long addr;
 530         int i;
 531
 532         syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
 533                                     GFP_KERNEL);
 534         if (!syscalls_metadata) {
 535                 WARN_ON(1);
 536                 return -ENOMEM;
 537         }
 538
 539         for (i = 0; i < NR_syscalls; i++) {
 540                 addr = arch_syscall_addr(i);
 541                 meta = find_syscall_meta(addr);
 542                 if (!meta)
 543                         continue;
 544
 545                 meta->syscall_nr = i;
 546                 syscalls_metadata[i] = meta;
 547         }
 548
 549         return 0;
 550 }
 551 early_initcall(init_ftrace_syscalls);
 552
 553 #ifdef CONFIG_PERF_EVENTS
 554
 555 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 556 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 557 static int sys_perf_refcount_enter;
 558 static int sys_perf_refcount_exit;
 559
 560 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 561 {
 562         struct syscall_metadata *sys_data;
 563         struct syscall_trace_enter *rec;
 564         struct hlist_head *head;
 565         int syscall_nr;
 566         int rctx;
 567         int size;
 568
 569         syscall_nr = trace_get_syscall_nr(current, regs);
 570         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 571                 return;
 572         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 573                 return;
 574
 575         sys_data = syscall_nr_to_meta(syscall_nr);
 576         if (!sys_data)
 577                 return;
 578
 579         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 580         if (hlist_empty(head))
 581                 return;
 582
 583         /* get the size after alignment with the u32 buffer size field */
 584         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 585         size = ALIGN(size + sizeof(u32), sizeof(u64));
 586         size -= sizeof(u32);
 587
 588         rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
 589                                 sys_data->enter_event->event.type, regs, &rctx);
 590         if (!rec)
 591                 return;
 592
 593         rec->nr = syscall_nr;
 594         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 595                                (unsigned long *)&rec->args);
 596         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 597 }
 598
 599 static int perf_sysenter_enable(struct ftrace_event_call *call)
 600 {
 601         int ret = 0;
 602         int num;
 603
 604         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 605
 606         mutex_lock(&syscall_trace_lock);
 607         if (!sys_perf_refcount_enter)
 608                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 609         if (ret) {
 610                 pr_info("event trace: Could not activate"
 611                                 "syscall entry trace point");
 612         } else {
 613                 set_bit(num, enabled_perf_enter_syscalls);
 614                 sys_perf_refcount_enter++;
 615         }
 616         mutex_unlock(&syscall_trace_lock);
 617         return ret;
 618 }
 619
 620 static void perf_sysenter_disable(struct ftrace_event_call *call)
 621 {
 622         int num;
 623
 624         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 625
 626         mutex_lock(&syscall_trace_lock);
 627         sys_perf_refcount_enter--;
 628         clear_bit(num, enabled_perf_enter_syscalls);
 629         if (!sys_perf_refcount_enter)
 630                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 631         mutex_unlock(&syscall_trace_lock);
 632 }
 633
 634 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 635 {
 636         struct syscall_metadata *sys_data;
 637         struct syscall_trace_exit *rec;
 638         struct hlist_head *head;
 639         int syscall_nr;
 640         int rctx;
 641         int size;
 642
 643         syscall_nr = trace_get_syscall_nr(current, regs);
 644         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 645                 return;
 646         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 647                 return;
 648
 649         sys_data = syscall_nr_to_meta(syscall_nr);
 650         if (!sys_data)
 651                 return;
 652
 653         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 654         if (hlist_empty(head))
 655                 return;
 656
 657         /* We can probably do that at build time */
 658         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 659         size -= sizeof(u32);
 660
 661         rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
 662                                 sys_data->exit_event->event.type, regs, &rctx);
 663         if (!rec)
 664                 return;
 665
 666         rec->nr = syscall_nr;
 667         rec->ret = syscall_get_return_value(current, regs);
 668         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 669 }
 670
 671 static int perf_sysexit_enable(struct ftrace_event_call *call)
 672 {
 673         int ret = 0;
 674         int num;
 675
 676         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 677
 678         mutex_lock(&syscall_trace_lock);
 679         if (!sys_perf_refcount_exit)
 680                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 681         if (ret) {
 682                 pr_info("event trace: Could not activate"
 683                                 "syscall exit trace point");
 684         } else {
 685                 set_bit(num, enabled_perf_exit_syscalls);
 686                 sys_perf_refcount_exit++;
 687         }
 688         mutex_unlock(&syscall_trace_lock);
 689         return ret;
 690 }
 691
 692 static void perf_sysexit_disable(struct ftrace_event_call *call)
 693 {
 694         int num;
 695
 696         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 697
 698         mutex_lock(&syscall_trace_lock);
 699         sys_perf_refcount_exit--;
 700         clear_bit(num, enabled_perf_exit_syscalls);
 701         if (!sys_perf_refcount_exit)
 702                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 703         mutex_unlock(&syscall_trace_lock);
 704 }
 705
 706 #endif /* CONFIG_PERF_EVENTS */
 707
 708 static int syscall_enter_register(struct ftrace_event_call *event,
 709                                  enum trace_reg type, void *data)
 710 {
 711         struct ftrace_event_file *file = data;
 712
 713         switch (type) {
 714         case TRACE_REG_REGISTER:
 715                 return reg_event_syscall_enter(file, event);
 716         case TRACE_REG_UNREGISTER:
 717                 unreg_event_syscall_enter(file, event);
 718                 return 0;
 719
 720 #ifdef CONFIG_PERF_EVENTS
 721         case TRACE_REG_PERF_REGISTER:
 722                 return perf_sysenter_enable(event);
 723         case TRACE_REG_PERF_UNREGISTER:
 724                 perf_sysenter_disable(event);
 725                 return 0;
 726         case TRACE_REG_PERF_OPEN:
 727         case TRACE_REG_PERF_CLOSE:
 728         case TRACE_REG_PERF_ADD:
 729         case TRACE_REG_PERF_DEL:
 730                 return 0;
 731 #endif
 732         }
 733         return 0;
 734 }
 735
 736 static int syscall_exit_register(struct ftrace_event_call *event,
 737                                  enum trace_reg type, void *data)
 738 {
 739         struct ftrace_event_file *file = data;
 740
 741         switch (type) {
 742         case TRACE_REG_REGISTER:
 743                 return reg_event_syscall_exit(file, event);
 744         case TRACE_REG_UNREGISTER:
 745                 unreg_event_syscall_exit(file, event);
 746                 return 0;
 747
 748 #ifdef CONFIG_PERF_EVENTS
 749         case TRACE_REG_PERF_REGISTER:
 750                 return perf_sysexit_enable(event);
 751         case TRACE_REG_PERF_UNREGISTER:
 752                 perf_sysexit_disable(event);
 753                 return 0;
 754         case TRACE_REG_PERF_OPEN:
 755         case TRACE_REG_PERF_CLOSE:
 756         case TRACE_REG_PERF_ADD:
 757         case TRACE_REG_PERF_DEL:
 758                 return 0;
 759 #endif
 760         }
 761         return 0;
 762 }