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