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, field, name)                                \
 202         sizeof(type) != sizeof(trace.field) ?                           \
 203                 __bad_type_size() :                                     \
 204                 #type, #name, offsetof(typeof(trace), field),           \
 205                 sizeof(trace.field), is_signed_type(type)
 206
 207 static int __init
 208 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 209 {
 210         int i;
 211         int pos = 0;
 212
 213         /* When len=0, we just calculate the needed length */
 214 #define LEN_OR_ZERO (len ? len - pos : 0)
 215
 216         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 217         for (i = 0; i < entry->nb_args; i++) {
 218                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 219                                 entry->args[i], sizeof(unsigned long),
 220                                 i == entry->nb_args - 1 ? "" : ", ");
 221         }
 222         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 223
 224         for (i = 0; i < entry->nb_args; i++) {
 225                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 226                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 227         }
 228
 229 #undef LEN_OR_ZERO
 230
 231         /* return the length of print_fmt */
 232         return pos;
 233 }
 234
 235 static int __init set_syscall_print_fmt(struct trace_event_call *call)
 236 {
 237         char *print_fmt;
 238         int len;
 239         struct syscall_metadata *entry = call->data;
 240
 241         if (entry->enter_event != call) {
 242                 call->print_fmt = "\"0x%lx\", REC->ret";
 243                 return 0;
 244         }
 245
 246         /* First: called with 0 length to calculate the needed length */
 247         len = __set_enter_print_fmt(entry, NULL, 0);
 248
 249         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 250         if (!print_fmt)
 251                 return -ENOMEM;
 252
 253         /* Second: actually write the @print_fmt */
 254         __set_enter_print_fmt(entry, print_fmt, len + 1);
 255         call->print_fmt = print_fmt;
 256
 257         return 0;
 258 }
 259
 260 static void __init free_syscall_print_fmt(struct trace_event_call *call)
 261 {
 262         struct syscall_metadata *entry = call->data;
 263
 264         if (entry->enter_event == call)
 265                 kfree(call->print_fmt);
 266 }
 267
 268 static int __init syscall_enter_define_fields(struct trace_event_call *call)
 269 {
 270         struct syscall_trace_enter trace;
 271         struct syscall_metadata *meta = call->data;
 272         int ret;
 273         int i;
 274         int offset = offsetof(typeof(trace), args);
 275
 276         ret = trace_define_field(call, SYSCALL_FIELD(int, nr, __syscall_nr),
 277                                  FILTER_OTHER);
 278         if (ret)
 279                 return ret;
 280
 281         for (i = 0; i < meta->nb_args; i++) {
 282                 ret = trace_define_field(call, meta->types[i],
 283                                          meta->args[i], offset,
 284                                          sizeof(unsigned long), 0,
 285                                          FILTER_OTHER);
 286                 offset += sizeof(unsigned long);
 287         }
 288
 289         return ret;
 290 }
 291
 292 static int __init syscall_exit_define_fields(struct trace_event_call *call)
 293 {
 294         struct syscall_trace_exit trace;
 295         int ret;
 296
 297         ret = trace_define_field(call, SYSCALL_FIELD(int, nr, __syscall_nr),
 298                                  FILTER_OTHER);
 299         if (ret)
 300                 return ret;
 301
 302         ret = trace_define_field(call, SYSCALL_FIELD(long, ret, ret),
 303                                  FILTER_OTHER);
 304
 305         return ret;
 306 }
 307
 308 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 309 {
 310         struct trace_array *tr = data;
 311         struct trace_event_file *trace_file;
 312         struct syscall_trace_enter *entry;
 313         struct syscall_metadata *sys_data;
 314         struct ring_buffer_event *event;
 315         struct ring_buffer *buffer;
 316         unsigned long irq_flags;
 317         unsigned long args[6];
 318         int pc;
 319         int syscall_nr;
 320         int size;
 321
 322         syscall_nr = trace_get_syscall_nr(current, regs);
 323         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 324                 return;
 325
 326         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
 327         trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
 328         if (!trace_file)
 329                 return;
 330
 331         if (trace_trigger_soft_disabled(trace_file))
 332                 return;
 333
 334         sys_data = syscall_nr_to_meta(syscall_nr);
 335         if (!sys_data)
 336                 return;
 337
 338         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 339
 340         local_save_flags(irq_flags);
 341         pc = preempt_count();
 342
 343         buffer = tr->trace_buffer.buffer;
 344         event = trace_buffer_lock_reserve(buffer,
 345                         sys_data->enter_event->event.type, size, irq_flags, pc);
 346         if (!event)
 347                 return;
 348
 349         entry = ring_buffer_event_data(event);
 350         entry->nr = syscall_nr;
 351         syscall_get_arguments(current, regs, args);
 352         memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
 353
 354         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 355                                     irq_flags, pc);
 356 }
 357
 358 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 359 {
 360         struct trace_array *tr = data;
 361         struct trace_event_file *trace_file;
 362         struct syscall_trace_exit *entry;
 363         struct syscall_metadata *sys_data;
 364         struct ring_buffer_event *event;
 365         struct ring_buffer *buffer;
 366         unsigned long irq_flags;
 367         int pc;
 368         int syscall_nr;
 369
 370         syscall_nr = trace_get_syscall_nr(current, regs);
 371         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 372                 return;
 373
 374         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 375         trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 376         if (!trace_file)
 377                 return;
 378
 379         if (trace_trigger_soft_disabled(trace_file))
 380                 return;
 381
 382         sys_data = syscall_nr_to_meta(syscall_nr);
 383         if (!sys_data)
 384                 return;
 385
 386         local_save_flags(irq_flags);
 387         pc = preempt_count();
 388
 389         buffer = tr->trace_buffer.buffer;
 390         event = trace_buffer_lock_reserve(buffer,
 391                         sys_data->exit_event->event.type, sizeof(*entry),
 392                         irq_flags, pc);
 393         if (!event)
 394                 return;
 395
 396         entry = ring_buffer_event_data(event);
 397         entry->nr = syscall_nr;
 398         entry->ret = syscall_get_return_value(current, regs);
 399
 400         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 401                                     irq_flags, pc);
 402 }
 403
 404 static int reg_event_syscall_enter(struct trace_event_file *file,
 405                                    struct trace_event_call *call)
 406 {
 407         struct trace_array *tr = file->tr;
 408         int ret = 0;
 409         int num;
 410
 411         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 412         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 413                 return -ENOSYS;
 414         mutex_lock(&syscall_trace_lock);
 415         if (!tr->sys_refcount_enter)
 416                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 417         if (!ret) {
 418                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 419                 tr->sys_refcount_enter++;
 420         }
 421         mutex_unlock(&syscall_trace_lock);
 422         return ret;
 423 }
 424
 425 static void unreg_event_syscall_enter(struct trace_event_file *file,
 426                                       struct trace_event_call *call)
 427 {
 428         struct trace_array *tr = file->tr;
 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;
 434         mutex_lock(&syscall_trace_lock);
 435         tr->sys_refcount_enter--;
 436         RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
 437         if (!tr->sys_refcount_enter)
 438                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 439         mutex_unlock(&syscall_trace_lock);
 440 }
 441
 442 static int reg_event_syscall_exit(struct trace_event_file *file,
 443                                   struct trace_event_call *call)
 444 {
 445         struct trace_array *tr = file->tr;
 446         int ret = 0;
 447         int num;
 448
 449         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 450         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 451                 return -ENOSYS;
 452         mutex_lock(&syscall_trace_lock);
 453         if (!tr->sys_refcount_exit)
 454                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 455         if (!ret) {
 456                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 457                 tr->sys_refcount_exit++;
 458         }
 459         mutex_unlock(&syscall_trace_lock);
 460         return ret;
 461 }
 462
 463 static void unreg_event_syscall_exit(struct trace_event_file *file,
 464                                      struct trace_event_call *call)
 465 {
 466         struct trace_array *tr = file->tr;
 467         int num;
 468
 469         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 470         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 471                 return;
 472         mutex_lock(&syscall_trace_lock);
 473         tr->sys_refcount_exit--;
 474         RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
 475         if (!tr->sys_refcount_exit)
 476                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 477         mutex_unlock(&syscall_trace_lock);
 478 }
 479
 480 static int __init init_syscall_trace(struct trace_event_call *call)
 481 {
 482         int id;
 483         int num;
 484
 485         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 486         if (num < 0 || num >= NR_syscalls) {
 487                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 488                                 ((struct syscall_metadata *)call->data)->name);
 489                 return -ENOSYS;
 490         }
 491
 492         if (set_syscall_print_fmt(call) < 0)
 493                 return -ENOMEM;
 494
 495         id = trace_event_raw_init(call);
 496
 497         if (id < 0) {
 498                 free_syscall_print_fmt(call);
 499                 return id;
 500         }
 501
 502         return id;
 503 }
 504
 505 struct trace_event_functions enter_syscall_print_funcs = {
 506         .trace          = print_syscall_enter,
 507 };
 508
 509 struct trace_event_functions exit_syscall_print_funcs = {
 510         .trace          = print_syscall_exit,
 511 };
 512
 513 struct trace_event_class __refdata event_class_syscall_enter = {
 514         .system         = "syscalls",
 515         .reg            = syscall_enter_register,
 516         .define_fields  = syscall_enter_define_fields,
 517         .get_fields     = syscall_get_enter_fields,
 518         .raw_init       = init_syscall_trace,
 519 };
 520
 521 struct trace_event_class __refdata event_class_syscall_exit = {
 522         .system         = "syscalls",
 523         .reg            = syscall_exit_register,
 524         .define_fields  = syscall_exit_define_fields,
 525         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 526         .raw_init       = init_syscall_trace,
 527 };
 528
 529 unsigned long __init __weak arch_syscall_addr(int nr)
 530 {
 531         return (unsigned long)sys_call_table[nr];
 532 }
 533
 534 void __init init_ftrace_syscalls(void)
 535 {
 536         struct syscall_metadata *meta;
 537         unsigned long addr;
 538         int i;
 539
 540         syscalls_metadata = kcalloc(NR_syscalls, sizeof(*syscalls_metadata),
 541                                     GFP_KERNEL);
 542         if (!syscalls_metadata) {
 543                 WARN_ON(1);
 544                 return;
 545         }
 546
 547         for (i = 0; i < NR_syscalls; i++) {
 548                 addr = arch_syscall_addr(i);
 549                 meta = find_syscall_meta(addr);
 550                 if (!meta)
 551                         continue;
 552
 553                 meta->syscall_nr = i;
 554                 syscalls_metadata[i] = meta;
 555         }
 556 }
 557
 558 #ifdef CONFIG_PERF_EVENTS
 559
 560 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 561 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 562 static int sys_perf_refcount_enter;
 563 static int sys_perf_refcount_exit;
 564
 565 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
 566                                struct syscall_metadata *sys_data,
 567                                struct syscall_trace_enter *rec)
 568 {
 569         struct syscall_tp_t {
 570                 unsigned long long regs;
 571                 unsigned long syscall_nr;
 572                 unsigned long args[SYSCALL_DEFINE_MAXARGS];
 573         } param;
 574         int i;
 575
 576         *(struct pt_regs **)&param = regs;
 577         param.syscall_nr = rec->nr;
 578         for (i = 0; i < sys_data->nb_args; i++)
 579                 param.args[i] = rec->args[i];
 580         return trace_call_bpf(call, &param);
 581 }
 582
 583 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 584 {
 585         struct syscall_metadata *sys_data;
 586         struct syscall_trace_enter *rec;
 587         struct hlist_head *head;
 588         unsigned long args[6];
 589         bool valid_prog_array;
 590         int syscall_nr;
 591         int rctx;
 592         int size;
 593
 594         syscall_nr = trace_get_syscall_nr(current, regs);
 595         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 596                 return;
 597         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 598                 return;
 599
 600         sys_data = syscall_nr_to_meta(syscall_nr);
 601         if (!sys_data)
 602                 return;
 603
 604         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 605         valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
 606         if (!valid_prog_array && hlist_empty(head))
 607                 return;
 608
 609         /* get the size after alignment with the u32 buffer size field */
 610         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 611         size = ALIGN(size + sizeof(u32), sizeof(u64));
 612         size -= sizeof(u32);
 613
 614         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 615         if (!rec)
 616                 return;
 617
 618         rec->nr = syscall_nr;
 619         syscall_get_arguments(current, regs, args);
 620         memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
 621
 622         if ((valid_prog_array &&
 623              !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
 624             hlist_empty(head)) {
 625                 perf_swevent_put_recursion_context(rctx);
 626                 return;
 627         }
 628
 629         perf_trace_buf_submit(rec, size, rctx,
 630                               sys_data->enter_event->event.type, 1, regs,
 631                               head, NULL);
 632 }
 633
 634 static int perf_sysenter_enable(struct trace_event_call *call)
 635 {
 636         int ret = 0;
 637         int num;
 638
 639         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 640
 641         mutex_lock(&syscall_trace_lock);
 642         if (!sys_perf_refcount_enter)
 643                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 644         if (ret) {
 645                 pr_info("event trace: Could not activate syscall entry trace point");
 646         } else {
 647                 set_bit(num, enabled_perf_enter_syscalls);
 648                 sys_perf_refcount_enter++;
 649         }
 650         mutex_unlock(&syscall_trace_lock);
 651         return ret;
 652 }
 653
 654 static void perf_sysenter_disable(struct trace_event_call *call)
 655 {
 656         int num;
 657
 658         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 659
 660         mutex_lock(&syscall_trace_lock);
 661         sys_perf_refcount_enter--;
 662         clear_bit(num, enabled_perf_enter_syscalls);
 663         if (!sys_perf_refcount_enter)
 664                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 665         mutex_unlock(&syscall_trace_lock);
 666 }
 667
 668 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
 669                               struct syscall_trace_exit *rec)
 670 {
 671         struct syscall_tp_t {
 672                 unsigned long long regs;
 673                 unsigned long syscall_nr;
 674                 unsigned long ret;
 675         } param;
 676
 677         *(struct pt_regs **)&param = regs;
 678         param.syscall_nr = rec->nr;
 679         param.ret = rec->ret;
 680         return trace_call_bpf(call, &param);
 681 }
 682
 683 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 684 {
 685         struct syscall_metadata *sys_data;
 686         struct syscall_trace_exit *rec;
 687         struct hlist_head *head;
 688         bool valid_prog_array;
 689         int syscall_nr;
 690         int rctx;
 691         int size;
 692
 693         syscall_nr = trace_get_syscall_nr(current, regs);
 694         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 695                 return;
 696         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 697                 return;
 698
 699         sys_data = syscall_nr_to_meta(syscall_nr);
 700         if (!sys_data)
 701                 return;
 702
 703         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 704         valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
 705         if (!valid_prog_array && hlist_empty(head))
 706                 return;
 707
 708         /* We can probably do that at build time */
 709         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 710         size -= sizeof(u32);
 711
 712         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 713         if (!rec)
 714                 return;
 715
 716         rec->nr = syscall_nr;
 717         rec->ret = syscall_get_return_value(current, regs);
 718
 719         if ((valid_prog_array &&
 720              !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
 721             hlist_empty(head)) {
 722                 perf_swevent_put_recursion_context(rctx);
 723                 return;
 724         }
 725
 726         perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
 727                               1, regs, head, NULL);
 728 }
 729
 730 static int perf_sysexit_enable(struct trace_event_call *call)
 731 {
 732         int ret = 0;
 733         int num;
 734
 735         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 736
 737         mutex_lock(&syscall_trace_lock);
 738         if (!sys_perf_refcount_exit)
 739                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 740         if (ret) {
 741                 pr_info("event trace: Could not activate syscall exit trace point");
 742         } else {
 743                 set_bit(num, enabled_perf_exit_syscalls);
 744                 sys_perf_refcount_exit++;
 745         }
 746         mutex_unlock(&syscall_trace_lock);
 747         return ret;
 748 }
 749
 750 static void perf_sysexit_disable(struct trace_event_call *call)
 751 {
 752         int num;
 753
 754         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 755
 756         mutex_lock(&syscall_trace_lock);
 757         sys_perf_refcount_exit--;
 758         clear_bit(num, enabled_perf_exit_syscalls);
 759         if (!sys_perf_refcount_exit)
 760                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 761         mutex_unlock(&syscall_trace_lock);
 762 }
 763
 764 #endif /* CONFIG_PERF_EVENTS */
 765
 766 static int syscall_enter_register(struct trace_event_call *event,
 767                                  enum trace_reg type, void *data)
 768 {
 769         struct trace_event_file *file = data;
 770
 771         switch (type) {
 772         case TRACE_REG_REGISTER:
 773                 return reg_event_syscall_enter(file, event);
 774         case TRACE_REG_UNREGISTER:
 775                 unreg_event_syscall_enter(file, event);
 776                 return 0;
 777
 778 #ifdef CONFIG_PERF_EVENTS
 779         case TRACE_REG_PERF_REGISTER:
 780                 return perf_sysenter_enable(event);
 781         case TRACE_REG_PERF_UNREGISTER:
 782                 perf_sysenter_disable(event);
 783                 return 0;
 784         case TRACE_REG_PERF_OPEN:
 785         case TRACE_REG_PERF_CLOSE:
 786         case TRACE_REG_PERF_ADD:
 787         case TRACE_REG_PERF_DEL:
 788                 return 0;
 789 #endif
 790         }
 791         return 0;
 792 }
 793
 794 static int syscall_exit_register(struct trace_event_call *event,
 795                                  enum trace_reg type, void *data)
 796 {
 797         struct trace_event_file *file = data;
 798
 799         switch (type) {
 800         case TRACE_REG_REGISTER:
 801                 return reg_event_syscall_exit(file, event);
 802         case TRACE_REG_UNREGISTER:
 803                 unreg_event_syscall_exit(file, event);
 804                 return 0;
 805
 806 #ifdef CONFIG_PERF_EVENTS
 807         case TRACE_REG_PERF_REGISTER:
 808                 return perf_sysexit_enable(event);
 809         case TRACE_REG_PERF_UNREGISTER:
 810                 perf_sysexit_disable(event);
 811                 return 0;
 812         case TRACE_REG_PERF_OPEN:
 813         case TRACE_REG_PERF_CLOSE:
 814         case TRACE_REG_PERF_ADD:
 815         case TRACE_REG_PERF_DEL:
 816                 return 0;
 817 #endif
 818         }
 819         return 0;
 820 }