kernel/trace/trace_syscalls.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <trace/syscall.h>
   3 #include <trace/events/syscalls.h>
   4 #include <linux/syscalls.h>
   5 #include <linux/slab.h>
   6 #include <linux/kernel.h>
   7 #include <linux/module.h>       /* for MODULE_NAME_LEN via KSYM_SYMBOL_LEN */
   8 #include <linux/ftrace.h>
   9 #include <linux/perf_event.h>
  10 #include <asm/syscall.h>
  11
  12 #include "trace_output.h"
  13 #include "trace.h"
  14
  15 static DEFINE_MUTEX(syscall_trace_lock);
  16
  17 static int syscall_enter_register(struct trace_event_call *event,
  18                                  enum trace_reg type, void *data);
  19 static int syscall_exit_register(struct trace_event_call *event,
  20                                  enum trace_reg type, void *data);
  21
  22 static struct list_head *
  23 syscall_get_enter_fields(struct trace_event_call *call)
  24 {
  25         struct syscall_metadata *entry = call->data;
  26
  27         return &entry->enter_fields;
  28 }
  29
  30 extern struct syscall_metadata *__start_syscalls_metadata[];
  31 extern struct syscall_metadata *__stop_syscalls_metadata[];
  32
  33 static struct syscall_metadata **syscalls_metadata;
  34
  35 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
  36 static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
  37 {
  38         /*
  39          * Only compare after the "sys" prefix. Archs that use
  40          * syscall wrappers may have syscalls symbols aliases prefixed
  41          * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
  42          * mismatch.
  43          */
  44         return !strcmp(sym + 3, name + 3);
  45 }
  46 #endif
  47
  48 #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
  49 /*
  50  * Some architectures that allow for 32bit applications
  51  * to run on a 64bit kernel, do not map the syscalls for
  52  * the 32bit tasks the same as they do for 64bit tasks.
  53  *
  54  *     *cough*x86*cough*
  55  *
  56  * In such a case, instead of reporting the wrong syscalls,
  57  * simply ignore them.
  58  *
  59  * For an arch to ignore the compat syscalls it needs to
  60  * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
  61  * define the function arch_trace_is_compat_syscall() to let
  62  * the tracing system know that it should ignore it.
  63  */
  64 static int
  65 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
  66 {
  67         if (unlikely(arch_trace_is_compat_syscall(regs)))
  68                 return -1;
  69
  70         return syscall_get_nr(task, regs);
  71 }
  72 #else
  73 static inline int
  74 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
  75 {
  76         return syscall_get_nr(task, regs);
  77 }
  78 #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
  79
  80 static __init struct syscall_metadata *
  81 find_syscall_meta(unsigned long syscall)
  82 {
  83         struct syscall_metadata **start;
  84         struct syscall_metadata **stop;
  85         char str[KSYM_SYMBOL_LEN];
  86
  87
  88         start = __start_syscalls_metadata;
  89         stop = __stop_syscalls_metadata;
  90         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
  91
  92         if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
  93                 return NULL;
  94
  95         for ( ; start < stop; start++) {
  96                 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
  97                         return *start;
  98         }
  99         return NULL;
 100 }
 101
 102 static struct syscall_metadata *syscall_nr_to_meta(int nr)
 103 {
 104         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
 105                 return NULL;
 106
 107         return syscalls_metadata[nr];
 108 }
 109
 110 const char *get_syscall_name(int syscall)
 111 {
 112         struct syscall_metadata *entry;
 113
 114         entry = syscall_nr_to_meta(syscall);
 115         if (!entry)
 116                 return NULL;
 117
 118         return entry->name;
 119 }
 120
 121 static enum print_line_t
 122 print_syscall_enter(struct trace_iterator *iter, int flags,
 123                     struct trace_event *event)
 124 {
 125         struct trace_array *tr = iter->tr;
 126         struct trace_seq *s = &iter->seq;
 127         struct trace_entry *ent = iter->ent;
 128         struct syscall_trace_enter *trace;
 129         struct syscall_metadata *entry;
 130         int i, syscall;
 131
 132         trace = (typeof(trace))ent;
 133         syscall = trace->nr;
 134         entry = syscall_nr_to_meta(syscall);
 135
 136         if (!entry)
 137                 goto end;
 138
 139         if (entry->enter_event->event.type != ent->type) {
 140                 WARN_ON_ONCE(1);
 141                 goto end;
 142         }
 143
 144         trace_seq_printf(s, "%s(", entry->name);
 145
 146         for (i = 0; i < entry->nb_args; i++) {
 147
 148                 if (trace_seq_has_overflowed(s))
 149                         goto end;
 150
 151                 /* parameter types */
 152                 if (tr->trace_flags & TRACE_ITER_VERBOSE)
 153                         trace_seq_printf(s, "%s ", entry->types[i]);
 154
 155                 /* parameter values */
 156                 trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 157                                  trace->args[i],
 158                                  i == entry->nb_args - 1 ? "" : ", ");
 159         }
 160
 161         trace_seq_putc(s, ')');
 162 end:
 163         trace_seq_putc(s, '\n');
 164
 165         return trace_handle_return(s);
 166 }
 167
 168 static enum print_line_t
 169 print_syscall_exit(struct trace_iterator *iter, int flags,
 170                    struct trace_event *event)
 171 {
 172         struct trace_seq *s = &iter->seq;
 173         struct trace_entry *ent = iter->ent;
 174         struct syscall_trace_exit *trace;
 175         int syscall;
 176         struct syscall_metadata *entry;
 177
 178         trace = (typeof(trace))ent;
 179         syscall = trace->nr;
 180         entry = syscall_nr_to_meta(syscall);
 181
 182         if (!entry) {
 183                 trace_seq_putc(s, '\n');
 184                 goto out;
 185         }
 186
 187         if (entry->exit_event->event.type != ent->type) {
 188                 WARN_ON_ONCE(1);
 189                 return TRACE_TYPE_UNHANDLED;
 190         }
 191
 192         trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 193                                 trace->ret);
 194
 195  out:
 196         return trace_handle_return(s);
 197 }
 198
 199 extern char *__bad_type_size(void);
 200
 201 #define SYSCALL_FIELD(_type, _name) {                                   \
 202         .type = #_type, .name = #_name,                                 \
 203         .size = sizeof(_type), .align = __alignof__(_type),             \
 204         .is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER }
 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 offset = offsetof(typeof(trace), args);
 272         int ret, i;
 273
 274         for (i = 0; i < meta->nb_args; i++) {
 275                 ret = trace_define_field(call, meta->types[i],
 276                                          meta->args[i], offset,
 277                                          sizeof(unsigned long), 0,
 278                                          FILTER_OTHER);
 279                 if (ret)
 280                         break;
 281                 offset += sizeof(unsigned long);
 282         }
 283
 284         return ret;
 285 }
 286
 287 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 288 {
 289         struct trace_array *tr = data;
 290         struct trace_event_file *trace_file;
 291         struct syscall_trace_enter *entry;
 292         struct syscall_metadata *sys_data;
 293         struct ring_buffer_event *event;
 294         struct ring_buffer *buffer;
 295         unsigned long irq_flags;
 296         unsigned long args[6];
 297         int pc;
 298         int syscall_nr;
 299         int size;
 300
 301         syscall_nr = trace_get_syscall_nr(current, regs);
 302         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 303                 return;
 304
 305         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
 306         trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
 307         if (!trace_file)
 308                 return;
 309
 310         if (trace_trigger_soft_disabled(trace_file))
 311                 return;
 312
 313         sys_data = syscall_nr_to_meta(syscall_nr);
 314         if (!sys_data)
 315                 return;
 316
 317         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 318
 319         local_save_flags(irq_flags);
 320         pc = preempt_count();
 321
 322         buffer = tr->trace_buffer.buffer;
 323         event = trace_buffer_lock_reserve(buffer,
 324                         sys_data->enter_event->event.type, size, irq_flags, pc);
 325         if (!event)
 326                 return;
 327
 328         entry = ring_buffer_event_data(event);
 329         entry->nr = syscall_nr;
 330         syscall_get_arguments(current, regs, args);
 331         memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
 332
 333         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 334                                     irq_flags, pc);
 335 }
 336
 337 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 338 {
 339         struct trace_array *tr = data;
 340         struct trace_event_file *trace_file;
 341         struct syscall_trace_exit *entry;
 342         struct syscall_metadata *sys_data;
 343         struct ring_buffer_event *event;
 344         struct ring_buffer *buffer;
 345         unsigned long irq_flags;
 346         int pc;
 347         int syscall_nr;
 348
 349         syscall_nr = trace_get_syscall_nr(current, regs);
 350         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 351                 return;
 352
 353         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 354         trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 355         if (!trace_file)
 356                 return;
 357
 358         if (trace_trigger_soft_disabled(trace_file))
 359                 return;
 360
 361         sys_data = syscall_nr_to_meta(syscall_nr);
 362         if (!sys_data)
 363                 return;
 364
 365         local_save_flags(irq_flags);
 366         pc = preempt_count();
 367
 368         buffer = tr->trace_buffer.buffer;
 369         event = trace_buffer_lock_reserve(buffer,
 370                         sys_data->exit_event->event.type, sizeof(*entry),
 371                         irq_flags, pc);
 372         if (!event)
 373                 return;
 374
 375         entry = ring_buffer_event_data(event);
 376         entry->nr = syscall_nr;
 377         entry->ret = syscall_get_return_value(current, regs);
 378
 379         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 380                                     irq_flags, pc);
 381 }
 382
 383 static int reg_event_syscall_enter(struct trace_event_file *file,
 384                                    struct trace_event_call *call)
 385 {
 386         struct trace_array *tr = file->tr;
 387         int ret = 0;
 388         int num;
 389
 390         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 391         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 392                 return -ENOSYS;
 393         mutex_lock(&syscall_trace_lock);
 394         if (!tr->sys_refcount_enter)
 395                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 396         if (!ret) {
 397                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 398                 tr->sys_refcount_enter++;
 399         }
 400         mutex_unlock(&syscall_trace_lock);
 401         return ret;
 402 }
 403
 404 static void unreg_event_syscall_enter(struct trace_event_file *file,
 405                                       struct trace_event_call *call)
 406 {
 407         struct trace_array *tr = file->tr;
 408         int num;
 409
 410         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 411         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 412                 return;
 413         mutex_lock(&syscall_trace_lock);
 414         tr->sys_refcount_enter--;
 415         RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
 416         if (!tr->sys_refcount_enter)
 417                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 418         mutex_unlock(&syscall_trace_lock);
 419 }
 420
 421 static int reg_event_syscall_exit(struct trace_event_file *file,
 422                                   struct trace_event_call *call)
 423 {
 424         struct trace_array *tr = file->tr;
 425         int ret = 0;
 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 -ENOSYS;
 431         mutex_lock(&syscall_trace_lock);
 432         if (!tr->sys_refcount_exit)
 433                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 434         if (!ret) {
 435                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 436                 tr->sys_refcount_exit++;
 437         }
 438         mutex_unlock(&syscall_trace_lock);
 439         return ret;
 440 }
 441
 442 static void unreg_event_syscall_exit(struct trace_event_file *file,
 443                                      struct trace_event_call *call)
 444 {
 445         struct trace_array *tr = file->tr;
 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;
 451         mutex_lock(&syscall_trace_lock);
 452         tr->sys_refcount_exit--;
 453         RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
 454         if (!tr->sys_refcount_exit)
 455                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 456         mutex_unlock(&syscall_trace_lock);
 457 }
 458
 459 static int __init init_syscall_trace(struct trace_event_call *call)
 460 {
 461         int id;
 462         int num;
 463
 464         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 465         if (num < 0 || num >= NR_syscalls) {
 466                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 467                                 ((struct syscall_metadata *)call->data)->name);
 468                 return -ENOSYS;
 469         }
 470
 471         if (set_syscall_print_fmt(call) < 0)
 472                 return -ENOMEM;
 473
 474         id = trace_event_raw_init(call);
 475
 476         if (id < 0) {
 477                 free_syscall_print_fmt(call);
 478                 return id;
 479         }
 480
 481         return id;
 482 }
 483
 484 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
 485         SYSCALL_FIELD(int, __syscall_nr),
 486         { .type = TRACE_FUNCTION_TYPE,
 487           .define_fields = syscall_enter_define_fields },
 488         {}
 489 };
 490
 491 struct trace_event_functions enter_syscall_print_funcs = {
 492         .trace          = print_syscall_enter,
 493 };
 494
 495 struct trace_event_functions exit_syscall_print_funcs = {
 496         .trace          = print_syscall_exit,
 497 };
 498
 499 struct trace_event_class __refdata event_class_syscall_enter = {
 500         .system         = "syscalls",
 501         .reg            = syscall_enter_register,
 502         .fields_array   = syscall_enter_fields_array,
 503         .get_fields     = syscall_get_enter_fields,
 504         .raw_init       = init_syscall_trace,
 505 };
 506
 507 struct trace_event_class __refdata event_class_syscall_exit = {
 508         .system         = "syscalls",
 509         .reg            = syscall_exit_register,
 510         .fields_array   = (struct trace_event_fields[]){
 511                 SYSCALL_FIELD(int, __syscall_nr),
 512                 SYSCALL_FIELD(long, ret),
 513                 {}
 514         },
 515         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 516         .raw_init       = init_syscall_trace,
 517 };
 518
 519 unsigned long __init __weak arch_syscall_addr(int nr)
 520 {
 521         return (unsigned long)sys_call_table[nr];
 522 }
 523
 524 void __init init_ftrace_syscalls(void)
 525 {
 526         struct syscall_metadata *meta;
 527         unsigned long addr;
 528         int i;
 529
 530         syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
 531                                     GFP_KERNEL);
 532         if (!syscalls_metadata) {
 533                 WARN_ON(1);
 534                 return;
 535         }
 536
 537         for (i = 0; i < NR_syscalls; i++) {
 538                 addr = arch_syscall_addr(i);
 539                 meta = find_syscall_meta(addr);
 540                 if (!meta)
 541                         continue;
 542
 543                 meta->syscall_nr = i;
 544                 syscalls_metadata[i] = meta;
 545         }
 546 }
 547
 548 #ifdef CONFIG_PERF_EVENTS
 549
 550 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 551 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 552 static int sys_perf_refcount_enter;
 553 static int sys_perf_refcount_exit;
 554
 555 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
 556                                struct syscall_metadata *sys_data,
 557                                struct syscall_trace_enter *rec)
 558 {
 559         struct syscall_tp_t {
 560                 unsigned long long regs;
 561                 unsigned long syscall_nr;
 562                 unsigned long args[SYSCALL_DEFINE_MAXARGS];
 563         } param;
 564         int i;
 565
 566         *(struct pt_regs **)&param = regs;
 567         param.syscall_nr = rec->nr;
 568         for (i = 0; i < sys_data->nb_args; i++)
 569                 param.args[i] = rec->args[i];
 570         return trace_call_bpf(call, &param);
 571 }
 572
 573 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 574 {
 575         struct syscall_metadata *sys_data;
 576         struct syscall_trace_enter *rec;
 577         struct hlist_head *head;
 578         unsigned long args[6];
 579         bool valid_prog_array;
 580         int syscall_nr;
 581         int rctx;
 582         int size;
 583
 584         syscall_nr = trace_get_syscall_nr(current, regs);
 585         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 586                 return;
 587         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 588                 return;
 589
 590         sys_data = syscall_nr_to_meta(syscall_nr);
 591         if (!sys_data)
 592                 return;
 593
 594         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 595         valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
 596         if (!valid_prog_array && hlist_empty(head))
 597                 return;
 598
 599         /* get the size after alignment with the u32 buffer size field */
 600         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 601         size = ALIGN(size + sizeof(u32), sizeof(u64));
 602         size -= sizeof(u32);
 603
 604         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 605         if (!rec)
 606                 return;
 607
 608         rec->nr = syscall_nr;
 609         syscall_get_arguments(current, regs, args);
 610         memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
 611
 612         if ((valid_prog_array &&
 613              !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
 614             hlist_empty(head)) {
 615                 perf_swevent_put_recursion_context(rctx);
 616                 return;
 617         }
 618
 619         perf_trace_buf_submit(rec, size, rctx,
 620                               sys_data->enter_event->event.type, 1, regs,
 621                               head, NULL);
 622 }
 623
 624 static int perf_sysenter_enable(struct trace_event_call *call)
 625 {
 626         int ret = 0;
 627         int num;
 628
 629         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 630
 631         mutex_lock(&syscall_trace_lock);
 632         if (!sys_perf_refcount_enter)
 633                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 634         if (ret) {
 635                 pr_info("event trace: Could not activate syscall entry trace point");
 636         } else {
 637                 set_bit(num, enabled_perf_enter_syscalls);
 638                 sys_perf_refcount_enter++;
 639         }
 640         mutex_unlock(&syscall_trace_lock);
 641         return ret;
 642 }
 643
 644 static void perf_sysenter_disable(struct trace_event_call *call)
 645 {
 646         int num;
 647
 648         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 649
 650         mutex_lock(&syscall_trace_lock);
 651         sys_perf_refcount_enter--;
 652         clear_bit(num, enabled_perf_enter_syscalls);
 653         if (!sys_perf_refcount_enter)
 654                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 655         mutex_unlock(&syscall_trace_lock);
 656 }
 657
 658 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
 659                               struct syscall_trace_exit *rec)
 660 {
 661         struct syscall_tp_t {
 662                 unsigned long long regs;
 663                 unsigned long syscall_nr;
 664                 unsigned long ret;
 665         } param;
 666
 667         *(struct pt_regs **)&param = regs;
 668         param.syscall_nr = rec->nr;
 669         param.ret = rec->ret;
 670         return trace_call_bpf(call, &param);
 671 }
 672
 673 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 674 {
 675         struct syscall_metadata *sys_data;
 676         struct syscall_trace_exit *rec;
 677         struct hlist_head *head;
 678         bool valid_prog_array;
 679         int syscall_nr;
 680         int rctx;
 681         int size;
 682
 683         syscall_nr = trace_get_syscall_nr(current, regs);
 684         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 685                 return;
 686         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 687                 return;
 688
 689         sys_data = syscall_nr_to_meta(syscall_nr);
 690         if (!sys_data)
 691                 return;
 692
 693         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 694         valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
 695         if (!valid_prog_array && hlist_empty(head))
 696                 return;
 697
 698         /* We can probably do that at build time */
 699         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 700         size -= sizeof(u32);
 701
 702         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 703         if (!rec)
 704                 return;
 705
 706         rec->nr = syscall_nr;
 707         rec->ret = syscall_get_return_value(current, regs);
 708
 709         if ((valid_prog_array &&
 710              !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
 711             hlist_empty(head)) {
 712                 perf_swevent_put_recursion_context(rctx);
 713                 return;
 714         }
 715
 716         perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
 717                               1, regs, head, NULL);
 718 }
 719
 720 static int perf_sysexit_enable(struct trace_event_call *call)
 721 {
 722         int ret = 0;
 723         int num;
 724
 725         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 726
 727         mutex_lock(&syscall_trace_lock);
 728         if (!sys_perf_refcount_exit)
 729                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 730         if (ret) {
 731                 pr_info("event trace: Could not activate syscall exit trace point");
 732         } else {
 733                 set_bit(num, enabled_perf_exit_syscalls);
 734                 sys_perf_refcount_exit++;
 735         }
 736         mutex_unlock(&syscall_trace_lock);
 737         return ret;
 738 }
 739
 740 static void perf_sysexit_disable(struct trace_event_call *call)
 741 {
 742         int num;
 743
 744         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 745
 746         mutex_lock(&syscall_trace_lock);
 747         sys_perf_refcount_exit--;
 748         clear_bit(num, enabled_perf_exit_syscalls);
 749         if (!sys_perf_refcount_exit)
 750                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 751         mutex_unlock(&syscall_trace_lock);
 752 }
 753
 754 #endif /* CONFIG_PERF_EVENTS */
 755
 756 static int syscall_enter_register(struct trace_event_call *event,
 757                                  enum trace_reg type, void *data)
 758 {
 759         struct trace_event_file *file = data;
 760
 761         switch (type) {
 762         case TRACE_REG_REGISTER:
 763                 return reg_event_syscall_enter(file, event);
 764         case TRACE_REG_UNREGISTER:
 765                 unreg_event_syscall_enter(file, event);
 766                 return 0;
 767
 768 #ifdef CONFIG_PERF_EVENTS
 769         case TRACE_REG_PERF_REGISTER:
 770                 return perf_sysenter_enable(event);
 771         case TRACE_REG_PERF_UNREGISTER:
 772                 perf_sysenter_disable(event);
 773                 return 0;
 774         case TRACE_REG_PERF_OPEN:
 775         case TRACE_REG_PERF_CLOSE:
 776         case TRACE_REG_PERF_ADD:
 777         case TRACE_REG_PERF_DEL:
 778                 return 0;
 779 #endif
 780         }
 781         return 0;
 782 }
 783
 784 static int syscall_exit_register(struct trace_event_call *event,
 785                                  enum trace_reg type, void *data)
 786 {
 787         struct trace_event_file *file = data;
 788
 789         switch (type) {
 790         case TRACE_REG_REGISTER:
 791                 return reg_event_syscall_exit(file, event);
 792         case TRACE_REG_UNREGISTER:
 793                 unreg_event_syscall_exit(file, event);
 794                 return 0;
 795
 796 #ifdef CONFIG_PERF_EVENTS
 797         case TRACE_REG_PERF_REGISTER:
 798                 return perf_sysexit_enable(event);
 799         case TRACE_REG_PERF_UNREGISTER:
 800                 perf_sysexit_disable(event);
 801                 return 0;
 802         case TRACE_REG_PERF_OPEN:
 803         case TRACE_REG_PERF_CLOSE:
 804         case TRACE_REG_PERF_ADD:
 805         case TRACE_REG_PERF_DEL:
 806                 return 0;
 807 #endif
 808         }
 809         return 0;
 810 }