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_printf(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
 204 int  __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 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 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 syscall_trace_enter *entry;
 306         struct syscall_metadata *sys_data;
 307         struct ring_buffer_event *event;
 308         struct ring_buffer *buffer;
 309         unsigned long irq_flags;
 310         int pc;
 311         int syscall_nr;
 312         int size;
 313
 314         syscall_nr = trace_get_syscall_nr(current, regs);
 315         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 316                 return;
 317         if (!test_bit(syscall_nr, tr->enabled_enter_syscalls))
 318                 return;
 319
 320         sys_data = syscall_nr_to_meta(syscall_nr);
 321         if (!sys_data)
 322                 return;
 323
 324         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 325
 326         local_save_flags(irq_flags);
 327         pc = preempt_count();
 328
 329         buffer = tr->trace_buffer.buffer;
 330         event = trace_buffer_lock_reserve(buffer,
 331                         sys_data->enter_event->event.type, size, irq_flags, pc);
 332         if (!event)
 333                 return;
 334
 335         entry = ring_buffer_event_data(event);
 336         entry->nr = syscall_nr;
 337         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 338
 339         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 340                                           entry, event))
 341                 trace_current_buffer_unlock_commit(buffer, event,
 342                                                    irq_flags, pc);
 343 }
 344
 345 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 346 {
 347         struct trace_array *tr = data;
 348         struct syscall_trace_exit *entry;
 349         struct syscall_metadata *sys_data;
 350         struct ring_buffer_event *event;
 351         struct ring_buffer *buffer;
 352         unsigned long irq_flags;
 353         int pc;
 354         int syscall_nr;
 355
 356         syscall_nr = trace_get_syscall_nr(current, regs);
 357         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 358                 return;
 359         if (!test_bit(syscall_nr, tr->enabled_exit_syscalls))
 360                 return;
 361
 362         sys_data = syscall_nr_to_meta(syscall_nr);
 363         if (!sys_data)
 364                 return;
 365
 366         local_save_flags(irq_flags);
 367         pc = preempt_count();
 368
 369         buffer = tr->trace_buffer.buffer;
 370         event = trace_buffer_lock_reserve(buffer,
 371                         sys_data->exit_event->event.type, sizeof(*entry),
 372                         irq_flags, pc);
 373         if (!event)
 374                 return;
 375
 376         entry = ring_buffer_event_data(event);
 377         entry->nr = syscall_nr;
 378         entry->ret = syscall_get_return_value(current, regs);
 379
 380         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 381                                           entry, event))
 382                 trace_current_buffer_unlock_commit(buffer, event,
 383                                                    irq_flags, pc);
 384 }
 385
 386 static int reg_event_syscall_enter(struct ftrace_event_file *file,
 387                                    struct ftrace_event_call *call)
 388 {
 389         struct trace_array *tr = file->tr;
 390         int ret = 0;
 391         int num;
 392
 393         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 394         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 395                 return -ENOSYS;
 396         mutex_lock(&syscall_trace_lock);
 397         if (!tr->sys_refcount_enter)
 398                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 399         if (!ret) {
 400                 set_bit(num, tr->enabled_enter_syscalls);
 401                 tr->sys_refcount_enter++;
 402         }
 403         mutex_unlock(&syscall_trace_lock);
 404         return ret;
 405 }
 406
 407 static void unreg_event_syscall_enter(struct ftrace_event_file *file,
 408                                       struct ftrace_event_call *call)
 409 {
 410         struct trace_array *tr = file->tr;
 411         int num;
 412
 413         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 414         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 415                 return;
 416         mutex_lock(&syscall_trace_lock);
 417         tr->sys_refcount_enter--;
 418         clear_bit(num, tr->enabled_enter_syscalls);
 419         if (!tr->sys_refcount_enter)
 420                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 421         mutex_unlock(&syscall_trace_lock);
 422 }
 423
 424 static int reg_event_syscall_exit(struct ftrace_event_file *file,
 425                                   struct ftrace_event_call *call)
 426 {
 427         struct trace_array *tr = file->tr;
 428         int ret = 0;
 429         int num;
 430
 431         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 432         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 433                 return -ENOSYS;
 434         mutex_lock(&syscall_trace_lock);
 435         if (!tr->sys_refcount_exit)
 436                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 437         if (!ret) {
 438                 set_bit(num, tr->enabled_exit_syscalls);
 439                 tr->sys_refcount_exit++;
 440         }
 441         mutex_unlock(&syscall_trace_lock);
 442         return ret;
 443 }
 444
 445 static void unreg_event_syscall_exit(struct ftrace_event_file *file,
 446                                      struct ftrace_event_call *call)
 447 {
 448         struct trace_array *tr = file->tr;
 449         int num;
 450
 451         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 452         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 453                 return;
 454         mutex_lock(&syscall_trace_lock);
 455         tr->sys_refcount_exit--;
 456         clear_bit(num, tr->enabled_exit_syscalls);
 457         if (!tr->sys_refcount_exit)
 458                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 459         mutex_unlock(&syscall_trace_lock);
 460 }
 461
 462 static int init_syscall_trace(struct ftrace_event_call *call)
 463 {
 464         int id;
 465         int num;
 466
 467         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 468         if (num < 0 || num >= NR_syscalls) {
 469                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 470                                 ((struct syscall_metadata *)call->data)->name);
 471                 return -ENOSYS;
 472         }
 473
 474         if (set_syscall_print_fmt(call) < 0)
 475                 return -ENOMEM;
 476
 477         id = trace_event_raw_init(call);
 478
 479         if (id < 0) {
 480                 free_syscall_print_fmt(call);
 481                 return id;
 482         }
 483
 484         return id;
 485 }
 486
 487 struct trace_event_functions enter_syscall_print_funcs = {
 488         .trace          = print_syscall_enter,
 489 };
 490
 491 struct trace_event_functions exit_syscall_print_funcs = {
 492         .trace          = print_syscall_exit,
 493 };
 494
 495 struct ftrace_event_class __refdata event_class_syscall_enter = {
 496         .system         = "syscalls",
 497         .reg            = syscall_enter_register,
 498         .define_fields  = syscall_enter_define_fields,
 499         .get_fields     = syscall_get_enter_fields,
 500         .raw_init       = init_syscall_trace,
 501 };
 502
 503 struct ftrace_event_class __refdata event_class_syscall_exit = {
 504         .system         = "syscalls",
 505         .reg            = syscall_exit_register,
 506         .define_fields  = syscall_exit_define_fields,
 507         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 508         .raw_init       = init_syscall_trace,
 509 };
 510
 511 unsigned long __init __weak arch_syscall_addr(int nr)
 512 {
 513         return (unsigned long)sys_call_table[nr];
 514 }
 515
 516 static int __init init_ftrace_syscalls(void)
 517 {
 518         struct syscall_metadata *meta;
 519         unsigned long addr;
 520         int i;
 521
 522         syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
 523                                     GFP_KERNEL);
 524         if (!syscalls_metadata) {
 525                 WARN_ON(1);
 526                 return -ENOMEM;
 527         }
 528
 529         for (i = 0; i < NR_syscalls; i++) {
 530                 addr = arch_syscall_addr(i);
 531                 meta = find_syscall_meta(addr);
 532                 if (!meta)
 533                         continue;
 534
 535                 meta->syscall_nr = i;
 536                 syscalls_metadata[i] = meta;
 537         }
 538
 539         return 0;
 540 }
 541 early_initcall(init_ftrace_syscalls);
 542
 543 #ifdef CONFIG_PERF_EVENTS
 544
 545 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 546 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 547 static int sys_perf_refcount_enter;
 548 static int sys_perf_refcount_exit;
 549
 550 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 551 {
 552         struct syscall_metadata *sys_data;
 553         struct syscall_trace_enter *rec;
 554         struct hlist_head *head;
 555         int syscall_nr;
 556         int rctx;
 557         int size;
 558
 559         syscall_nr = trace_get_syscall_nr(current, regs);
 560         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 561                 return;
 562         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 563                 return;
 564
 565         sys_data = syscall_nr_to_meta(syscall_nr);
 566         if (!sys_data)
 567                 return;
 568
 569         /* get the size after alignment with the u32 buffer size field */
 570         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 571         size = ALIGN(size + sizeof(u32), sizeof(u64));
 572         size -= sizeof(u32);
 573
 574         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 575                       "perf buffer not large enough"))
 576                 return;
 577
 578         rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
 579                                 sys_data->enter_event->event.type, regs, &rctx);
 580         if (!rec)
 581                 return;
 582
 583         rec->nr = syscall_nr;
 584         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 585                                (unsigned long *)&rec->args);
 586
 587         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 588         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 589 }
 590
 591 static int perf_sysenter_enable(struct ftrace_event_call *call)
 592 {
 593         int ret = 0;
 594         int num;
 595
 596         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 597
 598         mutex_lock(&syscall_trace_lock);
 599         if (!sys_perf_refcount_enter)
 600                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 601         if (ret) {
 602                 pr_info("event trace: Could not activate"
 603                                 "syscall entry trace point");
 604         } else {
 605                 set_bit(num, enabled_perf_enter_syscalls);
 606                 sys_perf_refcount_enter++;
 607         }
 608         mutex_unlock(&syscall_trace_lock);
 609         return ret;
 610 }
 611
 612 static void perf_sysenter_disable(struct ftrace_event_call *call)
 613 {
 614         int num;
 615
 616         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 617
 618         mutex_lock(&syscall_trace_lock);
 619         sys_perf_refcount_enter--;
 620         clear_bit(num, enabled_perf_enter_syscalls);
 621         if (!sys_perf_refcount_enter)
 622                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 623         mutex_unlock(&syscall_trace_lock);
 624 }
 625
 626 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 627 {
 628         struct syscall_metadata *sys_data;
 629         struct syscall_trace_exit *rec;
 630         struct hlist_head *head;
 631         int syscall_nr;
 632         int rctx;
 633         int size;
 634
 635         syscall_nr = trace_get_syscall_nr(current, regs);
 636         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 637                 return;
 638         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 639                 return;
 640
 641         sys_data = syscall_nr_to_meta(syscall_nr);
 642         if (!sys_data)
 643                 return;
 644
 645         /* We can probably do that at build time */
 646         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 647         size -= sizeof(u32);
 648
 649         /*
 650          * Impossible, but be paranoid with the future
 651          * How to put this check outside runtime?
 652          */
 653         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 654                 "exit event has grown above perf buffer size"))
 655                 return;
 656
 657         rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
 658                                 sys_data->exit_event->event.type, regs, &rctx);
 659         if (!rec)
 660                 return;
 661
 662         rec->nr = syscall_nr;
 663         rec->ret = syscall_get_return_value(current, regs);
 664
 665         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 666         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head, NULL);
 667 }
 668
 669 static int perf_sysexit_enable(struct ftrace_event_call *call)
 670 {
 671         int ret = 0;
 672         int num;
 673
 674         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 675
 676         mutex_lock(&syscall_trace_lock);
 677         if (!sys_perf_refcount_exit)
 678                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 679         if (ret) {
 680                 pr_info("event trace: Could not activate"
 681                                 "syscall exit trace point");
 682         } else {
 683                 set_bit(num, enabled_perf_exit_syscalls);
 684                 sys_perf_refcount_exit++;
 685         }
 686         mutex_unlock(&syscall_trace_lock);
 687         return ret;
 688 }
 689
 690 static void perf_sysexit_disable(struct ftrace_event_call *call)
 691 {
 692         int num;
 693
 694         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 695
 696         mutex_lock(&syscall_trace_lock);
 697         sys_perf_refcount_exit--;
 698         clear_bit(num, enabled_perf_exit_syscalls);
 699         if (!sys_perf_refcount_exit)
 700                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 701         mutex_unlock(&syscall_trace_lock);
 702 }
 703
 704 #endif /* CONFIG_PERF_EVENTS */
 705
 706 static int syscall_enter_register(struct ftrace_event_call *event,
 707                                  enum trace_reg type, void *data)
 708 {
 709         struct ftrace_event_file *file = data;
 710
 711         switch (type) {
 712         case TRACE_REG_REGISTER:
 713                 return reg_event_syscall_enter(file, event);
 714         case TRACE_REG_UNREGISTER:
 715                 unreg_event_syscall_enter(file, event);
 716                 return 0;
 717
 718 #ifdef CONFIG_PERF_EVENTS
 719         case TRACE_REG_PERF_REGISTER:
 720                 return perf_sysenter_enable(event);
 721         case TRACE_REG_PERF_UNREGISTER:
 722                 perf_sysenter_disable(event);
 723                 return 0;
 724         case TRACE_REG_PERF_OPEN:
 725         case TRACE_REG_PERF_CLOSE:
 726         case TRACE_REG_PERF_ADD:
 727         case TRACE_REG_PERF_DEL:
 728                 return 0;
 729 #endif
 730         }
 731         return 0;
 732 }
 733
 734 static int syscall_exit_register(struct ftrace_event_call *event,
 735                                  enum trace_reg type, void *data)
 736 {
 737         struct ftrace_event_file *file = data;
 738
 739         switch (type) {
 740         case TRACE_REG_REGISTER:
 741                 return reg_event_syscall_exit(file, event);
 742         case TRACE_REG_UNREGISTER:
 743                 unreg_event_syscall_exit(file, event);
 744                 return 0;
 745
 746 #ifdef CONFIG_PERF_EVENTS
 747         case TRACE_REG_PERF_REGISTER:
 748                 return perf_sysexit_enable(event);
 749         case TRACE_REG_PERF_UNREGISTER:
 750                 perf_sysexit_disable(event);
 751                 return 0;
 752         case TRACE_REG_PERF_OPEN:
 753         case TRACE_REG_PERF_CLOSE:
 754         case TRACE_REG_PERF_ADD:
 755         case TRACE_REG_PERF_DEL:
 756                 return 0;
 757 #endif
 758         }
 759         return 0;
 760 }