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 trace_event_call *event,
  17                                  enum trace_reg type, void *data);
  18 static int syscall_exit_register(struct trace_event_call *event,
  19                                  enum trace_reg type, void *data);
  20
  21 static struct list_head *
  22 syscall_get_enter_fields(struct trace_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 const char *get_syscall_name(int syscall)
 110 {
 111         struct syscall_metadata *entry;
 112
 113         entry = syscall_nr_to_meta(syscall);
 114         if (!entry)
 115                 return NULL;
 116
 117         return entry->name;
 118 }
 119
 120 static enum print_line_t
 121 print_syscall_enter(struct trace_iterator *iter, int flags,
 122                     struct trace_event *event)
 123 {
 124         struct trace_array *tr = iter->tr;
 125         struct trace_seq *s = &iter->seq;
 126         struct trace_entry *ent = iter->ent;
 127         struct syscall_trace_enter *trace;
 128         struct syscall_metadata *entry;
 129         int i, syscall;
 130
 131         trace = (typeof(trace))ent;
 132         syscall = trace->nr;
 133         entry = syscall_nr_to_meta(syscall);
 134
 135         if (!entry)
 136                 goto end;
 137
 138         if (entry->enter_event->event.type != ent->type) {
 139                 WARN_ON_ONCE(1);
 140                 goto end;
 141         }
 142
 143         trace_seq_printf(s, "%s(", entry->name);
 144
 145         for (i = 0; i < entry->nb_args; i++) {
 146
 147                 if (trace_seq_has_overflowed(s))
 148                         goto end;
 149
 150                 /* parameter types */
 151                 if (tr->trace_flags & TRACE_ITER_VERBOSE)
 152                         trace_seq_printf(s, "%s ", entry->types[i]);
 153
 154                 /* parameter values */
 155                 trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 156                                  trace->args[i],
 157                                  i == entry->nb_args - 1 ? "" : ", ");
 158         }
 159
 160         trace_seq_putc(s, ')');
 161 end:
 162         trace_seq_putc(s, '\n');
 163
 164         return trace_handle_return(s);
 165 }
 166
 167 static enum print_line_t
 168 print_syscall_exit(struct trace_iterator *iter, int flags,
 169                    struct trace_event *event)
 170 {
 171         struct trace_seq *s = &iter->seq;
 172         struct trace_entry *ent = iter->ent;
 173         struct syscall_trace_exit *trace;
 174         int syscall;
 175         struct syscall_metadata *entry;
 176
 177         trace = (typeof(trace))ent;
 178         syscall = trace->nr;
 179         entry = syscall_nr_to_meta(syscall);
 180
 181         if (!entry) {
 182                 trace_seq_putc(s, '\n');
 183                 goto out;
 184         }
 185
 186         if (entry->exit_event->event.type != ent->type) {
 187                 WARN_ON_ONCE(1);
 188                 return TRACE_TYPE_UNHANDLED;
 189         }
 190
 191         trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 192                                 trace->ret);
 193
 194  out:
 195         return trace_handle_return(s);
 196 }
 197
 198 extern char *__bad_type_size(void);
 199
 200 #define SYSCALL_FIELD(type, field, name)                                \
 201         sizeof(type) != sizeof(trace.field) ?                           \
 202                 __bad_type_size() :                                     \
 203                 #type, #name, offsetof(typeof(trace), field),           \
 204                 sizeof(trace.field), is_signed_type(type)
 205
 206 static int __init
 207 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 208 {
 209         int i;
 210         int pos = 0;
 211
 212         /* When len=0, we just calculate the needed length */
 213 #define LEN_OR_ZERO (len ? len - pos : 0)
 214
 215         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 216         for (i = 0; i < entry->nb_args; i++) {
 217                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 218                                 entry->args[i], sizeof(unsigned long),
 219                                 i == entry->nb_args - 1 ? "" : ", ");
 220         }
 221         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 222
 223         for (i = 0; i < entry->nb_args; i++) {
 224                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 225                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 226         }
 227
 228 #undef LEN_OR_ZERO
 229
 230         /* return the length of print_fmt */
 231         return pos;
 232 }
 233
 234 static int __init set_syscall_print_fmt(struct trace_event_call *call)
 235 {
 236         char *print_fmt;
 237         int len;
 238         struct syscall_metadata *entry = call->data;
 239
 240         if (entry->enter_event != call) {
 241                 call->print_fmt = "\"0x%lx\", REC->ret";
 242                 return 0;
 243         }
 244
 245         /* First: called with 0 length to calculate the needed length */
 246         len = __set_enter_print_fmt(entry, NULL, 0);
 247
 248         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 249         if (!print_fmt)
 250                 return -ENOMEM;
 251
 252         /* Second: actually write the @print_fmt */
 253         __set_enter_print_fmt(entry, print_fmt, len + 1);
 254         call->print_fmt = print_fmt;
 255
 256         return 0;
 257 }
 258
 259 static void __init free_syscall_print_fmt(struct trace_event_call *call)
 260 {
 261         struct syscall_metadata *entry = call->data;
 262
 263         if (entry->enter_event == call)
 264                 kfree(call->print_fmt);
 265 }
 266
 267 static int __init syscall_enter_define_fields(struct trace_event_call *call)
 268 {
 269         struct syscall_trace_enter trace;
 270         struct syscall_metadata *meta = call->data;
 271         int ret;
 272         int i;
 273         int offset = offsetof(typeof(trace), args);
 274
 275         ret = trace_define_field(call, SYSCALL_FIELD(int, nr, __syscall_nr),
 276                                  FILTER_OTHER);
 277         if (ret)
 278                 return ret;
 279
 280         for (i = 0; i < meta->nb_args; i++) {
 281                 ret = trace_define_field(call, meta->types[i],
 282                                          meta->args[i], offset,
 283                                          sizeof(unsigned long), 0,
 284                                          FILTER_OTHER);
 285                 offset += sizeof(unsigned long);
 286         }
 287
 288         return ret;
 289 }
 290
 291 static int __init syscall_exit_define_fields(struct trace_event_call *call)
 292 {
 293         struct syscall_trace_exit trace;
 294         int ret;
 295
 296         ret = trace_define_field(call, SYSCALL_FIELD(int, nr, __syscall_nr),
 297                                  FILTER_OTHER);
 298         if (ret)
 299                 return ret;
 300
 301         ret = trace_define_field(call, SYSCALL_FIELD(long, ret, ret),
 302                                  FILTER_OTHER);
 303
 304         return ret;
 305 }
 306
 307 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 308 {
 309         struct trace_array *tr = data;
 310         struct trace_event_file *trace_file;
 311         struct syscall_trace_enter *entry;
 312         struct syscall_metadata *sys_data;
 313         struct ring_buffer_event *event;
 314         struct ring_buffer *buffer;
 315         unsigned long irq_flags;
 316         int pc;
 317         int syscall_nr;
 318         int size;
 319
 320         syscall_nr = trace_get_syscall_nr(current, regs);
 321         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 322                 return;
 323
 324         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
 325         trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
 326         if (!trace_file)
 327                 return;
 328
 329         if (trace_trigger_soft_disabled(trace_file))
 330                 return;
 331
 332         sys_data = syscall_nr_to_meta(syscall_nr);
 333         if (!sys_data)
 334                 return;
 335
 336         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 337
 338         local_save_flags(irq_flags);
 339         pc = preempt_count();
 340
 341         buffer = tr->trace_buffer.buffer;
 342         event = trace_buffer_lock_reserve(buffer,
 343                         sys_data->enter_event->event.type, size, irq_flags, pc);
 344         if (!event)
 345                 return;
 346
 347         entry = ring_buffer_event_data(event);
 348         entry->nr = syscall_nr;
 349         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 350
 351         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 352                                     irq_flags, pc);
 353 }
 354
 355 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 356 {
 357         struct trace_array *tr = data;
 358         struct trace_event_file *trace_file;
 359         struct syscall_trace_exit *entry;
 360         struct syscall_metadata *sys_data;
 361         struct ring_buffer_event *event;
 362         struct ring_buffer *buffer;
 363         unsigned long irq_flags;
 364         int pc;
 365         int syscall_nr;
 366
 367         syscall_nr = trace_get_syscall_nr(current, regs);
 368         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 369                 return;
 370
 371         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 372         trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 373         if (!trace_file)
 374                 return;
 375
 376         if (trace_trigger_soft_disabled(trace_file))
 377                 return;
 378
 379         sys_data = syscall_nr_to_meta(syscall_nr);
 380         if (!sys_data)
 381                 return;
 382
 383         local_save_flags(irq_flags);
 384         pc = preempt_count();
 385
 386         buffer = tr->trace_buffer.buffer;
 387         event = trace_buffer_lock_reserve(buffer,
 388                         sys_data->exit_event->event.type, sizeof(*entry),
 389                         irq_flags, pc);
 390         if (!event)
 391                 return;
 392
 393         entry = ring_buffer_event_data(event);
 394         entry->nr = syscall_nr;
 395         entry->ret = syscall_get_return_value(current, regs);
 396
 397         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 398                                     irq_flags, pc);
 399 }
 400
 401 static int reg_event_syscall_enter(struct trace_event_file *file,
 402                                    struct trace_event_call *call)
 403 {
 404         struct trace_array *tr = file->tr;
 405         int ret = 0;
 406         int num;
 407
 408         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 409         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 410                 return -ENOSYS;
 411         mutex_lock(&syscall_trace_lock);
 412         if (!tr->sys_refcount_enter)
 413                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 414         if (!ret) {
 415                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 416                 tr->sys_refcount_enter++;
 417         }
 418         mutex_unlock(&syscall_trace_lock);
 419         return ret;
 420 }
 421
 422 static void unreg_event_syscall_enter(struct trace_event_file *file,
 423                                       struct trace_event_call *call)
 424 {
 425         struct trace_array *tr = file->tr;
 426         int num;
 427
 428         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 429         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 430                 return;
 431         mutex_lock(&syscall_trace_lock);
 432         tr->sys_refcount_enter--;
 433         RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
 434         if (!tr->sys_refcount_enter)
 435                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 436         mutex_unlock(&syscall_trace_lock);
 437 }
 438
 439 static int reg_event_syscall_exit(struct trace_event_file *file,
 440                                   struct trace_event_call *call)
 441 {
 442         struct trace_array *tr = file->tr;
 443         int ret = 0;
 444         int num;
 445
 446         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 447         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 448                 return -ENOSYS;
 449         mutex_lock(&syscall_trace_lock);
 450         if (!tr->sys_refcount_exit)
 451                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 452         if (!ret) {
 453                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 454                 tr->sys_refcount_exit++;
 455         }
 456         mutex_unlock(&syscall_trace_lock);
 457         return ret;
 458 }
 459
 460 static void unreg_event_syscall_exit(struct trace_event_file *file,
 461                                      struct trace_event_call *call)
 462 {
 463         struct trace_array *tr = file->tr;
 464         int num;
 465
 466         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 467         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 468                 return;
 469         mutex_lock(&syscall_trace_lock);
 470         tr->sys_refcount_exit--;
 471         RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
 472         if (!tr->sys_refcount_exit)
 473                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 474         mutex_unlock(&syscall_trace_lock);
 475 }
 476
 477 static int __init init_syscall_trace(struct trace_event_call *call)
 478 {
 479         int id;
 480         int num;
 481
 482         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 483         if (num < 0 || num >= NR_syscalls) {
 484                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 485                                 ((struct syscall_metadata *)call->data)->name);
 486                 return -ENOSYS;
 487         }
 488
 489         if (set_syscall_print_fmt(call) < 0)
 490                 return -ENOMEM;
 491
 492         id = trace_event_raw_init(call);
 493
 494         if (id < 0) {
 495                 free_syscall_print_fmt(call);
 496                 return id;
 497         }
 498
 499         return id;
 500 }
 501
 502 struct trace_event_functions enter_syscall_print_funcs = {
 503         .trace          = print_syscall_enter,
 504 };
 505
 506 struct trace_event_functions exit_syscall_print_funcs = {
 507         .trace          = print_syscall_exit,
 508 };
 509
 510 struct trace_event_class __refdata event_class_syscall_enter = {
 511         .system         = "syscalls",
 512         .reg            = syscall_enter_register,
 513         .define_fields  = syscall_enter_define_fields,
 514         .get_fields     = syscall_get_enter_fields,
 515         .raw_init       = init_syscall_trace,
 516 };
 517
 518 struct trace_event_class __refdata event_class_syscall_exit = {
 519         .system         = "syscalls",
 520         .reg            = syscall_exit_register,
 521         .define_fields  = syscall_exit_define_fields,
 522         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 523         .raw_init       = init_syscall_trace,
 524 };
 525
 526 unsigned long __init __weak arch_syscall_addr(int nr)
 527 {
 528         return (unsigned long)sys_call_table[nr];
 529 }
 530
 531 void __init init_ftrace_syscalls(void)
 532 {
 533         struct syscall_metadata *meta;
 534         unsigned long addr;
 535         int i;
 536
 537         syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
 538                                     GFP_KERNEL);
 539         if (!syscalls_metadata) {
 540                 WARN_ON(1);
 541                 return;
 542         }
 543
 544         for (i = 0; i < NR_syscalls; i++) {
 545                 addr = arch_syscall_addr(i);
 546                 meta = find_syscall_meta(addr);
 547                 if (!meta)
 548                         continue;
 549
 550                 meta->syscall_nr = i;
 551                 syscalls_metadata[i] = meta;
 552         }
 553 }
 554
 555 #ifdef CONFIG_PERF_EVENTS
 556
 557 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 558 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 559 static int sys_perf_refcount_enter;
 560 static int sys_perf_refcount_exit;
 561
 562 static int perf_call_bpf_enter(struct bpf_prog *prog, struct pt_regs *regs,
 563                               struct syscall_metadata *sys_data,
 564                               struct syscall_trace_enter *rec) {
 565         struct syscall_tp_t {
 566                 unsigned long long regs;
 567                 unsigned long syscall_nr;
 568                 unsigned long args[SYSCALL_DEFINE_MAXARGS];
 569         } param;
 570         int i;
 571
 572         *(struct pt_regs **)&param = regs;
 573         param.syscall_nr = rec->nr;
 574         for (i = 0; i < sys_data->nb_args; i++)
 575                 param.args[i] = rec->args[i];
 576         return trace_call_bpf(prog, &param);
 577 }
 578
 579 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 580 {
 581         struct syscall_metadata *sys_data;
 582         struct syscall_trace_enter *rec;
 583         struct hlist_head *head;
 584         struct bpf_prog *prog;
 585         int syscall_nr;
 586         int rctx;
 587         int size;
 588
 589         syscall_nr = trace_get_syscall_nr(current, regs);
 590         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 591                 return;
 592         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 593                 return;
 594
 595         sys_data = syscall_nr_to_meta(syscall_nr);
 596         if (!sys_data)
 597                 return;
 598
 599         prog = READ_ONCE(sys_data->enter_event->prog);
 600         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 601         if (!prog && hlist_empty(head))
 602                 return;
 603
 604         /* get the size after alignment with the u32 buffer size field */
 605         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 606         size = ALIGN(size + sizeof(u32), sizeof(u64));
 607         size -= sizeof(u32);
 608
 609         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 610         if (!rec)
 611                 return;
 612
 613         rec->nr = syscall_nr;
 614         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 615                                (unsigned long *)&rec->args);
 616
 617         if ((prog && !perf_call_bpf_enter(prog, regs, sys_data, rec)) ||
 618             hlist_empty(head)) {
 619                 perf_swevent_put_recursion_context(rctx);
 620                 return;
 621         }
 622
 623         perf_trace_buf_submit(rec, size, rctx,
 624                               sys_data->enter_event->event.type, 1, regs,
 625                               head, NULL, NULL);
 626 }
 627
 628 static int perf_sysenter_enable(struct trace_event_call *call)
 629 {
 630         int ret = 0;
 631         int num;
 632
 633         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 634
 635         mutex_lock(&syscall_trace_lock);
 636         if (!sys_perf_refcount_enter)
 637                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 638         if (ret) {
 639                 pr_info("event trace: Could not activate syscall entry trace point");
 640         } else {
 641                 set_bit(num, enabled_perf_enter_syscalls);
 642                 sys_perf_refcount_enter++;
 643         }
 644         mutex_unlock(&syscall_trace_lock);
 645         return ret;
 646 }
 647
 648 static void perf_sysenter_disable(struct trace_event_call *call)
 649 {
 650         int num;
 651
 652         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 653
 654         mutex_lock(&syscall_trace_lock);
 655         sys_perf_refcount_enter--;
 656         clear_bit(num, enabled_perf_enter_syscalls);
 657         if (!sys_perf_refcount_enter)
 658                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 659         mutex_unlock(&syscall_trace_lock);
 660 }
 661
 662 static int perf_call_bpf_exit(struct bpf_prog *prog, struct pt_regs *regs,
 663                               struct syscall_trace_exit *rec) {
 664         struct syscall_tp_t {
 665                 unsigned long long regs;
 666                 unsigned long syscall_nr;
 667                 unsigned long ret;
 668         } param;
 669
 670         *(struct pt_regs **)&param = regs;
 671         param.syscall_nr = rec->nr;
 672         param.ret = rec->ret;
 673         return trace_call_bpf(prog, &param);
 674 }
 675
 676 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 677 {
 678         struct syscall_metadata *sys_data;
 679         struct syscall_trace_exit *rec;
 680         struct hlist_head *head;
 681         struct bpf_prog *prog;
 682         int syscall_nr;
 683         int rctx;
 684         int size;
 685
 686         syscall_nr = trace_get_syscall_nr(current, regs);
 687         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 688                 return;
 689         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 690                 return;
 691
 692         sys_data = syscall_nr_to_meta(syscall_nr);
 693         if (!sys_data)
 694                 return;
 695
 696         prog = READ_ONCE(sys_data->exit_event->prog);
 697         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 698         if (!prog && hlist_empty(head))
 699                 return;
 700
 701         /* We can probably do that at build time */
 702         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 703         size -= sizeof(u32);
 704
 705         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 706         if (!rec)
 707                 return;
 708
 709         rec->nr = syscall_nr;
 710         rec->ret = syscall_get_return_value(current, regs);
 711
 712         if ((prog && !perf_call_bpf_exit(prog, regs, rec)) ||
 713             hlist_empty(head)) {
 714                 perf_swevent_put_recursion_context(rctx);
 715                 return;
 716         }
 717
 718         perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
 719                               1, regs, head, NULL, NULL);
 720 }
 721
 722 static int perf_sysexit_enable(struct trace_event_call *call)
 723 {
 724         int ret = 0;
 725         int num;
 726
 727         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 728
 729         mutex_lock(&syscall_trace_lock);
 730         if (!sys_perf_refcount_exit)
 731                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 732         if (ret) {
 733                 pr_info("event trace: Could not activate syscall exit trace point");
 734         } else {
 735                 set_bit(num, enabled_perf_exit_syscalls);
 736                 sys_perf_refcount_exit++;
 737         }
 738         mutex_unlock(&syscall_trace_lock);
 739         return ret;
 740 }
 741
 742 static void perf_sysexit_disable(struct trace_event_call *call)
 743 {
 744         int num;
 745
 746         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 747
 748         mutex_lock(&syscall_trace_lock);
 749         sys_perf_refcount_exit--;
 750         clear_bit(num, enabled_perf_exit_syscalls);
 751         if (!sys_perf_refcount_exit)
 752                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 753         mutex_unlock(&syscall_trace_lock);
 754 }
 755
 756 #endif /* CONFIG_PERF_EVENTS */
 757
 758 static int syscall_enter_register(struct trace_event_call *event,
 759                                  enum trace_reg type, void *data)
 760 {
 761         struct trace_event_file *file = data;
 762
 763         switch (type) {
 764         case TRACE_REG_REGISTER:
 765                 return reg_event_syscall_enter(file, event);
 766         case TRACE_REG_UNREGISTER:
 767                 unreg_event_syscall_enter(file, event);
 768                 return 0;
 769
 770 #ifdef CONFIG_PERF_EVENTS
 771         case TRACE_REG_PERF_REGISTER:
 772                 return perf_sysenter_enable(event);
 773         case TRACE_REG_PERF_UNREGISTER:
 774                 perf_sysenter_disable(event);
 775                 return 0;
 776         case TRACE_REG_PERF_OPEN:
 777         case TRACE_REG_PERF_CLOSE:
 778         case TRACE_REG_PERF_ADD:
 779         case TRACE_REG_PERF_DEL:
 780                 return 0;
 781 #endif
 782         }
 783         return 0;
 784 }
 785
 786 static int syscall_exit_register(struct trace_event_call *event,
 787                                  enum trace_reg type, void *data)
 788 {
 789         struct trace_event_file *file = data;
 790
 791         switch (type) {
 792         case TRACE_REG_REGISTER:
 793                 return reg_event_syscall_exit(file, event);
 794         case TRACE_REG_UNREGISTER:
 795                 unreg_event_syscall_exit(file, event);
 796                 return 0;
 797
 798 #ifdef CONFIG_PERF_EVENTS
 799         case TRACE_REG_PERF_REGISTER:
 800                 return perf_sysexit_enable(event);
 801         case TRACE_REG_PERF_UNREGISTER:
 802                 perf_sysexit_disable(event);
 803                 return 0;
 804         case TRACE_REG_PERF_OPEN:
 805         case TRACE_REG_PERF_CLOSE:
 806         case TRACE_REG_PERF_ADD:
 807         case TRACE_REG_PERF_DEL:
 808                 return 0;
 809 #endif
 810         }
 811         return 0;
 812 }