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 <linux/xarray.h>
  11 #include <asm/syscall.h>
  12
  13 #include "trace_output.h"
  14 #include "trace.h"
  15
  16 static DEFINE_MUTEX(syscall_trace_lock);
  17
  18 static int syscall_enter_register(struct trace_event_call *event,
  19                                  enum trace_reg type, void *data);
  20 static int syscall_exit_register(struct trace_event_call *event,
  21                                  enum trace_reg type, void *data);
  22
  23 static struct list_head *
  24 syscall_get_enter_fields(struct trace_event_call *call)
  25 {
  26         struct syscall_metadata *entry = call->data;
  27
  28         return &entry->enter_fields;
  29 }
  30
  31 extern struct syscall_metadata *__start_syscalls_metadata[];
  32 extern struct syscall_metadata *__stop_syscalls_metadata[];
  33
  34 static DEFINE_XARRAY(syscalls_metadata_sparse);
  35 static struct syscall_metadata **syscalls_metadata;
  36
  37 #ifndef ARCH_HAS_SYSCALL_MATCH_SYM_NAME
  38 static inline bool arch_syscall_match_sym_name(const char *sym, const char *name)
  39 {
  40         /*
  41          * Only compare after the "sys" prefix. Archs that use
  42          * syscall wrappers may have syscalls symbols aliases prefixed
  43          * with ".SyS" or ".sys" instead of "sys", leading to an unwanted
  44          * mismatch.
  45          */
  46         return !strcmp(sym + 3, name + 3);
  47 }
  48 #endif
  49
  50 #ifdef ARCH_TRACE_IGNORE_COMPAT_SYSCALLS
  51 /*
  52  * Some architectures that allow for 32bit applications
  53  * to run on a 64bit kernel, do not map the syscalls for
  54  * the 32bit tasks the same as they do for 64bit tasks.
  55  *
  56  *     *cough*x86*cough*
  57  *
  58  * In such a case, instead of reporting the wrong syscalls,
  59  * simply ignore them.
  60  *
  61  * For an arch to ignore the compat syscalls it needs to
  62  * define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS as well as
  63  * define the function arch_trace_is_compat_syscall() to let
  64  * the tracing system know that it should ignore it.
  65  */
  66 static int
  67 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
  68 {
  69         if (unlikely(arch_trace_is_compat_syscall(regs)))
  70                 return -1;
  71
  72         return syscall_get_nr(task, regs);
  73 }
  74 #else
  75 static inline int
  76 trace_get_syscall_nr(struct task_struct *task, struct pt_regs *regs)
  77 {
  78         return syscall_get_nr(task, regs);
  79 }
  80 #endif /* ARCH_TRACE_IGNORE_COMPAT_SYSCALLS */
  81
  82 static __init struct syscall_metadata *
  83 find_syscall_meta(unsigned long syscall)
  84 {
  85         struct syscall_metadata **start;
  86         struct syscall_metadata **stop;
  87         char str[KSYM_SYMBOL_LEN];
  88
  89
  90         start = __start_syscalls_metadata;
  91         stop = __stop_syscalls_metadata;
  92         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
  93
  94         if (arch_syscall_match_sym_name(str, "sys_ni_syscall"))
  95                 return NULL;
  96
  97         for ( ; start < stop; start++) {
  98                 if ((*start)->name && arch_syscall_match_sym_name(str, (*start)->name))
  99                         return *start;
 100         }
 101         return NULL;
 102 }
 103
 104 static struct syscall_metadata *syscall_nr_to_meta(int nr)
 105 {
 106         if (IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR))
 107                 return xa_load(&syscalls_metadata_sparse, (unsigned long)nr);
 108
 109         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
 110                 return NULL;
 111
 112         return syscalls_metadata[nr];
 113 }
 114
 115 const char *get_syscall_name(int syscall)
 116 {
 117         struct syscall_metadata *entry;
 118
 119         entry = syscall_nr_to_meta(syscall);
 120         if (!entry)
 121                 return NULL;
 122
 123         return entry->name;
 124 }
 125
 126 static enum print_line_t
 127 print_syscall_enter(struct trace_iterator *iter, int flags,
 128                     struct trace_event *event)
 129 {
 130         struct trace_array *tr = iter->tr;
 131         struct trace_seq *s = &iter->seq;
 132         struct trace_entry *ent = iter->ent;
 133         struct syscall_trace_enter *trace;
 134         struct syscall_metadata *entry;
 135         int i, syscall;
 136
 137         trace = (typeof(trace))ent;
 138         syscall = trace->nr;
 139         entry = syscall_nr_to_meta(syscall);
 140
 141         if (!entry)
 142                 goto end;
 143
 144         if (entry->enter_event->event.type != ent->type) {
 145                 WARN_ON_ONCE(1);
 146                 goto end;
 147         }
 148
 149         trace_seq_printf(s, "%s(", entry->name);
 150
 151         for (i = 0; i < entry->nb_args; i++) {
 152
 153                 if (trace_seq_has_overflowed(s))
 154                         goto end;
 155
 156                 /* parameter types */
 157                 if (tr && tr->trace_flags & TRACE_ITER_VERBOSE)
 158                         trace_seq_printf(s, "%s ", entry->types[i]);
 159
 160                 /* parameter values */
 161                 trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 162                                  trace->args[i],
 163                                  i == entry->nb_args - 1 ? "" : ", ");
 164         }
 165
 166         trace_seq_putc(s, ')');
 167 end:
 168         trace_seq_putc(s, '\n');
 169
 170         return trace_handle_return(s);
 171 }
 172
 173 static enum print_line_t
 174 print_syscall_exit(struct trace_iterator *iter, int flags,
 175                    struct trace_event *event)
 176 {
 177         struct trace_seq *s = &iter->seq;
 178         struct trace_entry *ent = iter->ent;
 179         struct syscall_trace_exit *trace;
 180         int syscall;
 181         struct syscall_metadata *entry;
 182
 183         trace = (typeof(trace))ent;
 184         syscall = trace->nr;
 185         entry = syscall_nr_to_meta(syscall);
 186
 187         if (!entry) {
 188                 trace_seq_putc(s, '\n');
 189                 goto out;
 190         }
 191
 192         if (entry->exit_event->event.type != ent->type) {
 193                 WARN_ON_ONCE(1);
 194                 return TRACE_TYPE_UNHANDLED;
 195         }
 196
 197         trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 198                                 trace->ret);
 199
 200  out:
 201         return trace_handle_return(s);
 202 }
 203
 204 #define SYSCALL_FIELD(_type, _name) {                                   \
 205         .type = #_type, .name = #_name,                                 \
 206         .size = sizeof(_type), .align = __alignof__(_type),             \
 207         .is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER }
 208
 209 static int __init
 210 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 211 {
 212         int i;
 213         int pos = 0;
 214
 215         /* When len=0, we just calculate the needed length */
 216 #define LEN_OR_ZERO (len ? len - pos : 0)
 217
 218         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 219         for (i = 0; i < entry->nb_args; i++) {
 220                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 221                                 entry->args[i], sizeof(unsigned long),
 222                                 i == entry->nb_args - 1 ? "" : ", ");
 223         }
 224         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 225
 226         for (i = 0; i < entry->nb_args; i++) {
 227                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 228                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 229         }
 230
 231 #undef LEN_OR_ZERO
 232
 233         /* return the length of print_fmt */
 234         return pos;
 235 }
 236
 237 static int __init set_syscall_print_fmt(struct trace_event_call *call)
 238 {
 239         char *print_fmt;
 240         int len;
 241         struct syscall_metadata *entry = call->data;
 242
 243         if (entry->enter_event != call) {
 244                 call->print_fmt = "\"0x%lx\", REC->ret";
 245                 return 0;
 246         }
 247
 248         /* First: called with 0 length to calculate the needed length */
 249         len = __set_enter_print_fmt(entry, NULL, 0);
 250
 251         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 252         if (!print_fmt)
 253                 return -ENOMEM;
 254
 255         /* Second: actually write the @print_fmt */
 256         __set_enter_print_fmt(entry, print_fmt, len + 1);
 257         call->print_fmt = print_fmt;
 258
 259         return 0;
 260 }
 261
 262 static void __init free_syscall_print_fmt(struct trace_event_call *call)
 263 {
 264         struct syscall_metadata *entry = call->data;
 265
 266         if (entry->enter_event == call)
 267                 kfree(call->print_fmt);
 268 }
 269
 270 static int __init syscall_enter_define_fields(struct trace_event_call *call)
 271 {
 272         struct syscall_trace_enter trace;
 273         struct syscall_metadata *meta = call->data;
 274         int offset = offsetof(typeof(trace), args);
 275         int ret = 0;
 276         int i;
 277
 278         for (i = 0; i < meta->nb_args; i++) {
 279                 ret = trace_define_field(call, meta->types[i],
 280                                          meta->args[i], offset,
 281                                          sizeof(unsigned long), 0,
 282                                          FILTER_OTHER);
 283                 if (ret)
 284                         break;
 285                 offset += sizeof(unsigned long);
 286         }
 287
 288         return ret;
 289 }
 290
 291 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 292 {
 293         struct trace_array *tr = data;
 294         struct trace_event_file *trace_file;
 295         struct syscall_trace_enter *entry;
 296         struct syscall_metadata *sys_data;
 297         struct trace_event_buffer fbuffer;
 298         unsigned long args[6];
 299         int syscall_nr;
 300         int size;
 301
 302         syscall_nr = trace_get_syscall_nr(current, regs);
 303         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 304                 return;
 305
 306         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
 307         trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
 308         if (!trace_file)
 309                 return;
 310
 311         if (trace_trigger_soft_disabled(trace_file))
 312                 return;
 313
 314         sys_data = syscall_nr_to_meta(syscall_nr);
 315         if (!sys_data)
 316                 return;
 317
 318         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 319
 320         entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 321         if (!entry)
 322                 return;
 323
 324         entry = ring_buffer_event_data(fbuffer.event);
 325         entry->nr = syscall_nr;
 326         syscall_get_arguments(current, regs, args);
 327         memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
 328
 329         trace_event_buffer_commit(&fbuffer);
 330 }
 331
 332 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 333 {
 334         struct trace_array *tr = data;
 335         struct trace_event_file *trace_file;
 336         struct syscall_trace_exit *entry;
 337         struct syscall_metadata *sys_data;
 338         struct trace_event_buffer fbuffer;
 339         int syscall_nr;
 340
 341         syscall_nr = trace_get_syscall_nr(current, regs);
 342         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 343                 return;
 344
 345         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 346         trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 347         if (!trace_file)
 348                 return;
 349
 350         if (trace_trigger_soft_disabled(trace_file))
 351                 return;
 352
 353         sys_data = syscall_nr_to_meta(syscall_nr);
 354         if (!sys_data)
 355                 return;
 356
 357         entry = trace_event_buffer_reserve(&fbuffer, trace_file, sizeof(*entry));
 358         if (!entry)
 359                 return;
 360
 361         entry = ring_buffer_event_data(fbuffer.event);
 362         entry->nr = syscall_nr;
 363         entry->ret = syscall_get_return_value(current, regs);
 364
 365         trace_event_buffer_commit(&fbuffer);
 366 }
 367
 368 static int reg_event_syscall_enter(struct trace_event_file *file,
 369                                    struct trace_event_call *call)
 370 {
 371         struct trace_array *tr = file->tr;
 372         int ret = 0;
 373         int num;
 374
 375         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 376         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 377                 return -ENOSYS;
 378         mutex_lock(&syscall_trace_lock);
 379         if (!tr->sys_refcount_enter)
 380                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 381         if (!ret) {
 382                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 383                 tr->sys_refcount_enter++;
 384         }
 385         mutex_unlock(&syscall_trace_lock);
 386         return ret;
 387 }
 388
 389 static void unreg_event_syscall_enter(struct trace_event_file *file,
 390                                       struct trace_event_call *call)
 391 {
 392         struct trace_array *tr = file->tr;
 393         int num;
 394
 395         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 396         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 397                 return;
 398         mutex_lock(&syscall_trace_lock);
 399         tr->sys_refcount_enter--;
 400         RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
 401         if (!tr->sys_refcount_enter)
 402                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 403         mutex_unlock(&syscall_trace_lock);
 404 }
 405
 406 static int reg_event_syscall_exit(struct trace_event_file *file,
 407                                   struct trace_event_call *call)
 408 {
 409         struct trace_array *tr = file->tr;
 410         int ret = 0;
 411         int num;
 412
 413         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 414         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 415                 return -ENOSYS;
 416         mutex_lock(&syscall_trace_lock);
 417         if (!tr->sys_refcount_exit)
 418                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 419         if (!ret) {
 420                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 421                 tr->sys_refcount_exit++;
 422         }
 423         mutex_unlock(&syscall_trace_lock);
 424         return ret;
 425 }
 426
 427 static void unreg_event_syscall_exit(struct trace_event_file *file,
 428                                      struct trace_event_call *call)
 429 {
 430         struct trace_array *tr = file->tr;
 431         int num;
 432
 433         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 434         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 435                 return;
 436         mutex_lock(&syscall_trace_lock);
 437         tr->sys_refcount_exit--;
 438         RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
 439         if (!tr->sys_refcount_exit)
 440                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 441         mutex_unlock(&syscall_trace_lock);
 442 }
 443
 444 static int __init init_syscall_trace(struct trace_event_call *call)
 445 {
 446         int id;
 447         int num;
 448
 449         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 450         if (num < 0 || num >= NR_syscalls) {
 451                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 452                                 ((struct syscall_metadata *)call->data)->name);
 453                 return -ENOSYS;
 454         }
 455
 456         if (set_syscall_print_fmt(call) < 0)
 457                 return -ENOMEM;
 458
 459         id = trace_event_raw_init(call);
 460
 461         if (id < 0) {
 462                 free_syscall_print_fmt(call);
 463                 return id;
 464         }
 465
 466         return id;
 467 }
 468
 469 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
 470         SYSCALL_FIELD(int, __syscall_nr),
 471         { .type = TRACE_FUNCTION_TYPE,
 472           .define_fields = syscall_enter_define_fields },
 473         {}
 474 };
 475
 476 struct trace_event_functions enter_syscall_print_funcs = {
 477         .trace          = print_syscall_enter,
 478 };
 479
 480 struct trace_event_functions exit_syscall_print_funcs = {
 481         .trace          = print_syscall_exit,
 482 };
 483
 484 struct trace_event_class __refdata event_class_syscall_enter = {
 485         .system         = "syscalls",
 486         .reg            = syscall_enter_register,
 487         .fields_array   = syscall_enter_fields_array,
 488         .get_fields     = syscall_get_enter_fields,
 489         .raw_init       = init_syscall_trace,
 490 };
 491
 492 struct trace_event_class __refdata event_class_syscall_exit = {
 493         .system         = "syscalls",
 494         .reg            = syscall_exit_register,
 495         .fields_array   = (struct trace_event_fields[]){
 496                 SYSCALL_FIELD(int, __syscall_nr),
 497                 SYSCALL_FIELD(long, ret),
 498                 {}
 499         },
 500         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 501         .raw_init       = init_syscall_trace,
 502 };
 503
 504 unsigned long __init __weak arch_syscall_addr(int nr)
 505 {
 506         return (unsigned long)sys_call_table[nr];
 507 }
 508
 509 void __init init_ftrace_syscalls(void)
 510 {
 511         struct syscall_metadata *meta;
 512         unsigned long addr;
 513         int i;
 514         void *ret;
 515
 516         if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
 517                 syscalls_metadata = kcalloc(NR_syscalls,
 518                                         sizeof(*syscalls_metadata),
 519                                         GFP_KERNEL);
 520                 if (!syscalls_metadata) {
 521                         WARN_ON(1);
 522                         return;
 523                 }
 524         }
 525
 526         for (i = 0; i < NR_syscalls; i++) {
 527                 addr = arch_syscall_addr(i);
 528                 meta = find_syscall_meta(addr);
 529                 if (!meta)
 530                         continue;
 531
 532                 meta->syscall_nr = i;
 533
 534                 if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
 535                         syscalls_metadata[i] = meta;
 536                 } else {
 537                         ret = xa_store(&syscalls_metadata_sparse, i, meta,
 538                                         GFP_KERNEL);
 539                         WARN(xa_is_err(ret),
 540                                 "Syscall memory allocation failed\n");
 541                 }
 542
 543         }
 544 }
 545
 546 #ifdef CONFIG_PERF_EVENTS
 547
 548 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 549 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 550 static int sys_perf_refcount_enter;
 551 static int sys_perf_refcount_exit;
 552
 553 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
 554                                struct syscall_metadata *sys_data,
 555                                struct syscall_trace_enter *rec)
 556 {
 557         struct syscall_tp_t {
 558                 struct trace_entry ent;
 559                 int syscall_nr;
 560                 unsigned long args[SYSCALL_DEFINE_MAXARGS];
 561         } __aligned(8) param;
 562         int i;
 563
 564         BUILD_BUG_ON(sizeof(param.ent) < sizeof(void *));
 565
 566         /* bpf prog requires 'regs' to be the first member in the ctx (a.k.a. &param) */
 567         perf_fetch_caller_regs(regs);
 568         *(struct pt_regs **)&param = regs;
 569         param.syscall_nr = rec->nr;
 570         for (i = 0; i < sys_data->nb_args; i++)
 571                 param.args[i] = rec->args[i];
 572         return trace_call_bpf(call, &param);
 573 }
 574
 575 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 576 {
 577         struct syscall_metadata *sys_data;
 578         struct syscall_trace_enter *rec;
 579         struct pt_regs *fake_regs;
 580         struct hlist_head *head;
 581         unsigned long args[6];
 582         bool valid_prog_array;
 583         int syscall_nr;
 584         int rctx;
 585         int size;
 586
 587         syscall_nr = trace_get_syscall_nr(current, regs);
 588         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 589                 return;
 590         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 591                 return;
 592
 593         sys_data = syscall_nr_to_meta(syscall_nr);
 594         if (!sys_data)
 595                 return;
 596
 597         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 598         valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
 599         if (!valid_prog_array && hlist_empty(head))
 600                 return;
 601
 602         /* get the size after alignment with the u32 buffer size field */
 603         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 604         size = ALIGN(size + sizeof(u32), sizeof(u64));
 605         size -= sizeof(u32);
 606
 607         rec = perf_trace_buf_alloc(size, &fake_regs, &rctx);
 608         if (!rec)
 609                 return;
 610
 611         rec->nr = syscall_nr;
 612         syscall_get_arguments(current, regs, args);
 613         memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
 614
 615         if ((valid_prog_array &&
 616              !perf_call_bpf_enter(sys_data->enter_event, fake_regs, sys_data, rec)) ||
 617             hlist_empty(head)) {
 618                 perf_swevent_put_recursion_context(rctx);
 619                 return;
 620         }
 621
 622         perf_trace_buf_submit(rec, size, rctx,
 623                               sys_data->enter_event->event.type, 1, regs,
 624                               head, NULL);
 625 }
 626
 627 static int perf_sysenter_enable(struct trace_event_call *call)
 628 {
 629         int ret = 0;
 630         int num;
 631
 632         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 633
 634         mutex_lock(&syscall_trace_lock);
 635         if (!sys_perf_refcount_enter)
 636                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 637         if (ret) {
 638                 pr_info("event trace: Could not activate syscall entry trace point");
 639         } else {
 640                 set_bit(num, enabled_perf_enter_syscalls);
 641                 sys_perf_refcount_enter++;
 642         }
 643         mutex_unlock(&syscall_trace_lock);
 644         return ret;
 645 }
 646
 647 static void perf_sysenter_disable(struct trace_event_call *call)
 648 {
 649         int num;
 650
 651         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 652
 653         mutex_lock(&syscall_trace_lock);
 654         sys_perf_refcount_enter--;
 655         clear_bit(num, enabled_perf_enter_syscalls);
 656         if (!sys_perf_refcount_enter)
 657                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 658         mutex_unlock(&syscall_trace_lock);
 659 }
 660
 661 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
 662                               struct syscall_trace_exit *rec)
 663 {
 664         struct syscall_tp_t {
 665                 struct trace_entry ent;
 666                 int syscall_nr;
 667                 unsigned long ret;
 668         } __aligned(8) param;
 669
 670         /* bpf prog requires 'regs' to be the first member in the ctx (a.k.a. &param) */
 671         perf_fetch_caller_regs(regs);
 672         *(struct pt_regs **)&param = regs;
 673         param.syscall_nr = rec->nr;
 674         param.ret = rec->ret;
 675         return trace_call_bpf(call, &param);
 676 }
 677
 678 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 679 {
 680         struct syscall_metadata *sys_data;
 681         struct syscall_trace_exit *rec;
 682         struct pt_regs *fake_regs;
 683         struct hlist_head *head;
 684         bool valid_prog_array;
 685         int syscall_nr;
 686         int rctx;
 687         int size;
 688
 689         syscall_nr = trace_get_syscall_nr(current, regs);
 690         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 691                 return;
 692         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 693                 return;
 694
 695         sys_data = syscall_nr_to_meta(syscall_nr);
 696         if (!sys_data)
 697                 return;
 698
 699         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 700         valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
 701         if (!valid_prog_array && hlist_empty(head))
 702                 return;
 703
 704         /* We can probably do that at build time */
 705         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 706         size -= sizeof(u32);
 707
 708         rec = perf_trace_buf_alloc(size, &fake_regs, &rctx);
 709         if (!rec)
 710                 return;
 711
 712         rec->nr = syscall_nr;
 713         rec->ret = syscall_get_return_value(current, regs);
 714
 715         if ((valid_prog_array &&
 716              !perf_call_bpf_exit(sys_data->exit_event, fake_regs, rec)) ||
 717             hlist_empty(head)) {
 718                 perf_swevent_put_recursion_context(rctx);
 719                 return;
 720         }
 721
 722         perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
 723                               1, regs, head, NULL);
 724 }
 725
 726 static int perf_sysexit_enable(struct trace_event_call *call)
 727 {
 728         int ret = 0;
 729         int num;
 730
 731         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 732
 733         mutex_lock(&syscall_trace_lock);
 734         if (!sys_perf_refcount_exit)
 735                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 736         if (ret) {
 737                 pr_info("event trace: Could not activate syscall exit trace point");
 738         } else {
 739                 set_bit(num, enabled_perf_exit_syscalls);
 740                 sys_perf_refcount_exit++;
 741         }
 742         mutex_unlock(&syscall_trace_lock);
 743         return ret;
 744 }
 745
 746 static void perf_sysexit_disable(struct trace_event_call *call)
 747 {
 748         int num;
 749
 750         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 751
 752         mutex_lock(&syscall_trace_lock);
 753         sys_perf_refcount_exit--;
 754         clear_bit(num, enabled_perf_exit_syscalls);
 755         if (!sys_perf_refcount_exit)
 756                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 757         mutex_unlock(&syscall_trace_lock);
 758 }
 759
 760 #endif /* CONFIG_PERF_EVENTS */
 761
 762 static int syscall_enter_register(struct trace_event_call *event,
 763                                  enum trace_reg type, void *data)
 764 {
 765         struct trace_event_file *file = data;
 766
 767         switch (type) {
 768         case TRACE_REG_REGISTER:
 769                 return reg_event_syscall_enter(file, event);
 770         case TRACE_REG_UNREGISTER:
 771                 unreg_event_syscall_enter(file, event);
 772                 return 0;
 773
 774 #ifdef CONFIG_PERF_EVENTS
 775         case TRACE_REG_PERF_REGISTER:
 776                 return perf_sysenter_enable(event);
 777         case TRACE_REG_PERF_UNREGISTER:
 778                 perf_sysenter_disable(event);
 779                 return 0;
 780         case TRACE_REG_PERF_OPEN:
 781         case TRACE_REG_PERF_CLOSE:
 782         case TRACE_REG_PERF_ADD:
 783         case TRACE_REG_PERF_DEL:
 784                 return 0;
 785 #endif
 786         }
 787         return 0;
 788 }
 789
 790 static int syscall_exit_register(struct trace_event_call *event,
 791                                  enum trace_reg type, void *data)
 792 {
 793         struct trace_event_file *file = data;
 794
 795         switch (type) {
 796         case TRACE_REG_REGISTER:
 797                 return reg_event_syscall_exit(file, event);
 798         case TRACE_REG_UNREGISTER:
 799                 unreg_event_syscall_exit(file, event);
 800                 return 0;
 801
 802 #ifdef CONFIG_PERF_EVENTS
 803         case TRACE_REG_PERF_REGISTER:
 804                 return perf_sysexit_enable(event);
 805         case TRACE_REG_PERF_UNREGISTER:
 806                 perf_sysexit_disable(event);
 807                 return 0;
 808         case TRACE_REG_PERF_OPEN:
 809         case TRACE_REG_PERF_CLOSE:
 810         case TRACE_REG_PERF_ADD:
 811         case TRACE_REG_PERF_DEL:
 812                 return 0;
 813 #endif
 814         }
 815         return 0;
 816 }