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->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 extern char *__bad_type_size(void);
 205
 206 #define SYSCALL_FIELD(_type, _name) {                                   \
 207         .type = #_type, .name = #_name,                                 \
 208         .size = sizeof(_type), .align = __alignof__(_type),             \
 209         .is_signed = is_signed_type(_type), .filter_type = FILTER_OTHER }
 210
 211 static int __init
 212 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 213 {
 214         int i;
 215         int pos = 0;
 216
 217         /* When len=0, we just calculate the needed length */
 218 #define LEN_OR_ZERO (len ? len - pos : 0)
 219
 220         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 221         for (i = 0; i < entry->nb_args; i++) {
 222                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 223                                 entry->args[i], sizeof(unsigned long),
 224                                 i == entry->nb_args - 1 ? "" : ", ");
 225         }
 226         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 227
 228         for (i = 0; i < entry->nb_args; i++) {
 229                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 230                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 231         }
 232
 233 #undef LEN_OR_ZERO
 234
 235         /* return the length of print_fmt */
 236         return pos;
 237 }
 238
 239 static int __init set_syscall_print_fmt(struct trace_event_call *call)
 240 {
 241         char *print_fmt;
 242         int len;
 243         struct syscall_metadata *entry = call->data;
 244
 245         if (entry->enter_event != call) {
 246                 call->print_fmt = "\"0x%lx\", REC->ret";
 247                 return 0;
 248         }
 249
 250         /* First: called with 0 length to calculate the needed length */
 251         len = __set_enter_print_fmt(entry, NULL, 0);
 252
 253         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 254         if (!print_fmt)
 255                 return -ENOMEM;
 256
 257         /* Second: actually write the @print_fmt */
 258         __set_enter_print_fmt(entry, print_fmt, len + 1);
 259         call->print_fmt = print_fmt;
 260
 261         return 0;
 262 }
 263
 264 static void __init free_syscall_print_fmt(struct trace_event_call *call)
 265 {
 266         struct syscall_metadata *entry = call->data;
 267
 268         if (entry->enter_event == call)
 269                 kfree(call->print_fmt);
 270 }
 271
 272 static int __init syscall_enter_define_fields(struct trace_event_call *call)
 273 {
 274         struct syscall_trace_enter trace;
 275         struct syscall_metadata *meta = call->data;
 276         int offset = offsetof(typeof(trace), args);
 277         int ret, i;
 278
 279         for (i = 0; i < meta->nb_args; i++) {
 280                 ret = trace_define_field(call, meta->types[i],
 281                                          meta->args[i], offset,
 282                                          sizeof(unsigned long), 0,
 283                                          FILTER_OTHER);
 284                 if (ret)
 285                         break;
 286                 offset += sizeof(unsigned long);
 287         }
 288
 289         return ret;
 290 }
 291
 292 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 293 {
 294         struct trace_array *tr = data;
 295         struct trace_event_file *trace_file;
 296         struct syscall_trace_enter *entry;
 297         struct syscall_metadata *sys_data;
 298         struct ring_buffer_event *event;
 299         struct ring_buffer *buffer;
 300         unsigned long irq_flags;
 301         unsigned long args[6];
 302         int pc;
 303         int syscall_nr;
 304         int size;
 305
 306         syscall_nr = trace_get_syscall_nr(current, regs);
 307         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 308                 return;
 309
 310         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE) */
 311         trace_file = rcu_dereference_sched(tr->enter_syscall_files[syscall_nr]);
 312         if (!trace_file)
 313                 return;
 314
 315         if (trace_trigger_soft_disabled(trace_file))
 316                 return;
 317
 318         sys_data = syscall_nr_to_meta(syscall_nr);
 319         if (!sys_data)
 320                 return;
 321
 322         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 323
 324         local_save_flags(irq_flags);
 325         pc = preempt_count();
 326
 327         buffer = tr->trace_buffer.buffer;
 328         event = trace_buffer_lock_reserve(buffer,
 329                         sys_data->enter_event->event.type, size, irq_flags, pc);
 330         if (!event)
 331                 return;
 332
 333         entry = ring_buffer_event_data(event);
 334         entry->nr = syscall_nr;
 335         syscall_get_arguments(current, regs, args);
 336         memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
 337
 338         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 339                                     irq_flags, pc);
 340 }
 341
 342 static void ftrace_syscall_exit(void *data, struct pt_regs *regs, long ret)
 343 {
 344         struct trace_array *tr = data;
 345         struct trace_event_file *trace_file;
 346         struct syscall_trace_exit *entry;
 347         struct syscall_metadata *sys_data;
 348         struct ring_buffer_event *event;
 349         struct ring_buffer *buffer;
 350         unsigned long irq_flags;
 351         int pc;
 352         int syscall_nr;
 353
 354         syscall_nr = trace_get_syscall_nr(current, regs);
 355         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 356                 return;
 357
 358         /* Here we're inside tp handler's rcu_read_lock_sched (__DO_TRACE()) */
 359         trace_file = rcu_dereference_sched(tr->exit_syscall_files[syscall_nr]);
 360         if (!trace_file)
 361                 return;
 362
 363         if (trace_trigger_soft_disabled(trace_file))
 364                 return;
 365
 366         sys_data = syscall_nr_to_meta(syscall_nr);
 367         if (!sys_data)
 368                 return;
 369
 370         local_save_flags(irq_flags);
 371         pc = preempt_count();
 372
 373         buffer = tr->trace_buffer.buffer;
 374         event = trace_buffer_lock_reserve(buffer,
 375                         sys_data->exit_event->event.type, sizeof(*entry),
 376                         irq_flags, pc);
 377         if (!event)
 378                 return;
 379
 380         entry = ring_buffer_event_data(event);
 381         entry->nr = syscall_nr;
 382         entry->ret = syscall_get_return_value(current, regs);
 383
 384         event_trigger_unlock_commit(trace_file, buffer, event, entry,
 385                                     irq_flags, pc);
 386 }
 387
 388 static int reg_event_syscall_enter(struct trace_event_file *file,
 389                                    struct trace_event_call *call)
 390 {
 391         struct trace_array *tr = file->tr;
 392         int ret = 0;
 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 -ENOSYS;
 398         mutex_lock(&syscall_trace_lock);
 399         if (!tr->sys_refcount_enter)
 400                 ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 401         if (!ret) {
 402                 rcu_assign_pointer(tr->enter_syscall_files[num], file);
 403                 tr->sys_refcount_enter++;
 404         }
 405         mutex_unlock(&syscall_trace_lock);
 406         return ret;
 407 }
 408
 409 static void unreg_event_syscall_enter(struct trace_event_file *file,
 410                                       struct trace_event_call *call)
 411 {
 412         struct trace_array *tr = file->tr;
 413         int num;
 414
 415         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 416         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 417                 return;
 418         mutex_lock(&syscall_trace_lock);
 419         tr->sys_refcount_enter--;
 420         RCU_INIT_POINTER(tr->enter_syscall_files[num], NULL);
 421         if (!tr->sys_refcount_enter)
 422                 unregister_trace_sys_enter(ftrace_syscall_enter, tr);
 423         mutex_unlock(&syscall_trace_lock);
 424 }
 425
 426 static int reg_event_syscall_exit(struct trace_event_file *file,
 427                                   struct trace_event_call *call)
 428 {
 429         struct trace_array *tr = file->tr;
 430         int ret = 0;
 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 -ENOSYS;
 436         mutex_lock(&syscall_trace_lock);
 437         if (!tr->sys_refcount_exit)
 438                 ret = register_trace_sys_exit(ftrace_syscall_exit, tr);
 439         if (!ret) {
 440                 rcu_assign_pointer(tr->exit_syscall_files[num], file);
 441                 tr->sys_refcount_exit++;
 442         }
 443         mutex_unlock(&syscall_trace_lock);
 444         return ret;
 445 }
 446
 447 static void unreg_event_syscall_exit(struct trace_event_file *file,
 448                                      struct trace_event_call *call)
 449 {
 450         struct trace_array *tr = file->tr;
 451         int num;
 452
 453         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 454         if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 455                 return;
 456         mutex_lock(&syscall_trace_lock);
 457         tr->sys_refcount_exit--;
 458         RCU_INIT_POINTER(tr->exit_syscall_files[num], NULL);
 459         if (!tr->sys_refcount_exit)
 460                 unregister_trace_sys_exit(ftrace_syscall_exit, tr);
 461         mutex_unlock(&syscall_trace_lock);
 462 }
 463
 464 static int __init init_syscall_trace(struct trace_event_call *call)
 465 {
 466         int id;
 467         int num;
 468
 469         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 470         if (num < 0 || num >= NR_syscalls) {
 471                 pr_debug("syscall %s metadata not mapped, disabling ftrace event\n",
 472                                 ((struct syscall_metadata *)call->data)->name);
 473                 return -ENOSYS;
 474         }
 475
 476         if (set_syscall_print_fmt(call) < 0)
 477                 return -ENOMEM;
 478
 479         id = trace_event_raw_init(call);
 480
 481         if (id < 0) {
 482                 free_syscall_print_fmt(call);
 483                 return id;
 484         }
 485
 486         return id;
 487 }
 488
 489 static struct trace_event_fields __refdata syscall_enter_fields_array[] = {
 490         SYSCALL_FIELD(int, __syscall_nr),
 491         { .type = TRACE_FUNCTION_TYPE,
 492           .define_fields = syscall_enter_define_fields },
 493         {}
 494 };
 495
 496 struct trace_event_functions enter_syscall_print_funcs = {
 497         .trace          = print_syscall_enter,
 498 };
 499
 500 struct trace_event_functions exit_syscall_print_funcs = {
 501         .trace          = print_syscall_exit,
 502 };
 503
 504 struct trace_event_class __refdata event_class_syscall_enter = {
 505         .system         = "syscalls",
 506         .reg            = syscall_enter_register,
 507         .fields_array   = syscall_enter_fields_array,
 508         .get_fields     = syscall_get_enter_fields,
 509         .raw_init       = init_syscall_trace,
 510 };
 511
 512 struct trace_event_class __refdata event_class_syscall_exit = {
 513         .system         = "syscalls",
 514         .reg            = syscall_exit_register,
 515         .fields_array   = (struct trace_event_fields[]){
 516                 SYSCALL_FIELD(int, __syscall_nr),
 517                 SYSCALL_FIELD(long, ret),
 518                 {}
 519         },
 520         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
 521         .raw_init       = init_syscall_trace,
 522 };
 523
 524 unsigned long __init __weak arch_syscall_addr(int nr)
 525 {
 526         return (unsigned long)sys_call_table[nr];
 527 }
 528
 529 void __init init_ftrace_syscalls(void)
 530 {
 531         struct syscall_metadata *meta;
 532         unsigned long addr;
 533         int i;
 534         void *ret;
 535
 536         if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
 537                 syscalls_metadata = kcalloc(NR_syscalls,
 538                                         sizeof(*syscalls_metadata),
 539                                         GFP_KERNEL);
 540                 if (!syscalls_metadata) {
 541                         WARN_ON(1);
 542                         return;
 543                 }
 544         }
 545
 546         for (i = 0; i < NR_syscalls; i++) {
 547                 addr = arch_syscall_addr(i);
 548                 meta = find_syscall_meta(addr);
 549                 if (!meta)
 550                         continue;
 551
 552                 meta->syscall_nr = i;
 553
 554                 if (!IS_ENABLED(CONFIG_HAVE_SPARSE_SYSCALL_NR)) {
 555                         syscalls_metadata[i] = meta;
 556                 } else {
 557                         ret = xa_store(&syscalls_metadata_sparse, i, meta,
 558                                         GFP_KERNEL);
 559                         WARN(xa_is_err(ret),
 560                                 "Syscall memory allocation failed\n");
 561                 }
 562
 563         }
 564 }
 565
 566 #ifdef CONFIG_PERF_EVENTS
 567
 568 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 569 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 570 static int sys_perf_refcount_enter;
 571 static int sys_perf_refcount_exit;
 572
 573 static int perf_call_bpf_enter(struct trace_event_call *call, struct pt_regs *regs,
 574                                struct syscall_metadata *sys_data,
 575                                struct syscall_trace_enter *rec)
 576 {
 577         struct syscall_tp_t {
 578                 unsigned long long regs;
 579                 unsigned long syscall_nr;
 580                 unsigned long args[SYSCALL_DEFINE_MAXARGS];
 581         } param;
 582         int i;
 583
 584         *(struct pt_regs **)&param = regs;
 585         param.syscall_nr = rec->nr;
 586         for (i = 0; i < sys_data->nb_args; i++)
 587                 param.args[i] = rec->args[i];
 588         return trace_call_bpf(call, &param);
 589 }
 590
 591 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 592 {
 593         struct syscall_metadata *sys_data;
 594         struct syscall_trace_enter *rec;
 595         struct hlist_head *head;
 596         unsigned long args[6];
 597         bool valid_prog_array;
 598         int syscall_nr;
 599         int rctx;
 600         int size;
 601
 602         syscall_nr = trace_get_syscall_nr(current, regs);
 603         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 604                 return;
 605         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 606                 return;
 607
 608         sys_data = syscall_nr_to_meta(syscall_nr);
 609         if (!sys_data)
 610                 return;
 611
 612         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 613         valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
 614         if (!valid_prog_array && hlist_empty(head))
 615                 return;
 616
 617         /* get the size after alignment with the u32 buffer size field */
 618         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 619         size = ALIGN(size + sizeof(u32), sizeof(u64));
 620         size -= sizeof(u32);
 621
 622         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 623         if (!rec)
 624                 return;
 625
 626         rec->nr = syscall_nr;
 627         syscall_get_arguments(current, regs, args);
 628         memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
 629
 630         if ((valid_prog_array &&
 631              !perf_call_bpf_enter(sys_data->enter_event, regs, sys_data, rec)) ||
 632             hlist_empty(head)) {
 633                 perf_swevent_put_recursion_context(rctx);
 634                 return;
 635         }
 636
 637         perf_trace_buf_submit(rec, size, rctx,
 638                               sys_data->enter_event->event.type, 1, regs,
 639                               head, NULL);
 640 }
 641
 642 static int perf_sysenter_enable(struct trace_event_call *call)
 643 {
 644         int ret = 0;
 645         int num;
 646
 647         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 648
 649         mutex_lock(&syscall_trace_lock);
 650         if (!sys_perf_refcount_enter)
 651                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 652         if (ret) {
 653                 pr_info("event trace: Could not activate syscall entry trace point");
 654         } else {
 655                 set_bit(num, enabled_perf_enter_syscalls);
 656                 sys_perf_refcount_enter++;
 657         }
 658         mutex_unlock(&syscall_trace_lock);
 659         return ret;
 660 }
 661
 662 static void perf_sysenter_disable(struct trace_event_call *call)
 663 {
 664         int num;
 665
 666         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 667
 668         mutex_lock(&syscall_trace_lock);
 669         sys_perf_refcount_enter--;
 670         clear_bit(num, enabled_perf_enter_syscalls);
 671         if (!sys_perf_refcount_enter)
 672                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 673         mutex_unlock(&syscall_trace_lock);
 674 }
 675
 676 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,
 677                               struct syscall_trace_exit *rec)
 678 {
 679         struct syscall_tp_t {
 680                 unsigned long long regs;
 681                 unsigned long syscall_nr;
 682                 unsigned long ret;
 683         } param;
 684
 685         *(struct pt_regs **)&param = regs;
 686         param.syscall_nr = rec->nr;
 687         param.ret = rec->ret;
 688         return trace_call_bpf(call, &param);
 689 }
 690
 691 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 692 {
 693         struct syscall_metadata *sys_data;
 694         struct syscall_trace_exit *rec;
 695         struct hlist_head *head;
 696         bool valid_prog_array;
 697         int syscall_nr;
 698         int rctx;
 699         int size;
 700
 701         syscall_nr = trace_get_syscall_nr(current, regs);
 702         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
 703                 return;
 704         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 705                 return;
 706
 707         sys_data = syscall_nr_to_meta(syscall_nr);
 708         if (!sys_data)
 709                 return;
 710
 711         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 712         valid_prog_array = bpf_prog_array_valid(sys_data->exit_event);
 713         if (!valid_prog_array && hlist_empty(head))
 714                 return;
 715
 716         /* We can probably do that at build time */
 717         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 718         size -= sizeof(u32);
 719
 720         rec = perf_trace_buf_alloc(size, NULL, &rctx);
 721         if (!rec)
 722                 return;
 723
 724         rec->nr = syscall_nr;
 725         rec->ret = syscall_get_return_value(current, regs);
 726
 727         if ((valid_prog_array &&
 728              !perf_call_bpf_exit(sys_data->exit_event, regs, rec)) ||
 729             hlist_empty(head)) {
 730                 perf_swevent_put_recursion_context(rctx);
 731                 return;
 732         }
 733
 734         perf_trace_buf_submit(rec, size, rctx, sys_data->exit_event->event.type,
 735                               1, regs, head, NULL);
 736 }
 737
 738 static int perf_sysexit_enable(struct trace_event_call *call)
 739 {
 740         int ret = 0;
 741         int num;
 742
 743         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 744
 745         mutex_lock(&syscall_trace_lock);
 746         if (!sys_perf_refcount_exit)
 747                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 748         if (ret) {
 749                 pr_info("event trace: Could not activate syscall exit trace point");
 750         } else {
 751                 set_bit(num, enabled_perf_exit_syscalls);
 752                 sys_perf_refcount_exit++;
 753         }
 754         mutex_unlock(&syscall_trace_lock);
 755         return ret;
 756 }
 757
 758 static void perf_sysexit_disable(struct trace_event_call *call)
 759 {
 760         int num;
 761
 762         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 763
 764         mutex_lock(&syscall_trace_lock);
 765         sys_perf_refcount_exit--;
 766         clear_bit(num, enabled_perf_exit_syscalls);
 767         if (!sys_perf_refcount_exit)
 768                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 769         mutex_unlock(&syscall_trace_lock);
 770 }
 771
 772 #endif /* CONFIG_PERF_EVENTS */
 773
 774 static int syscall_enter_register(struct trace_event_call *event,
 775                                  enum trace_reg type, void *data)
 776 {
 777         struct trace_event_file *file = data;
 778
 779         switch (type) {
 780         case TRACE_REG_REGISTER:
 781                 return reg_event_syscall_enter(file, event);
 782         case TRACE_REG_UNREGISTER:
 783                 unreg_event_syscall_enter(file, event);
 784                 return 0;
 785
 786 #ifdef CONFIG_PERF_EVENTS
 787         case TRACE_REG_PERF_REGISTER:
 788                 return perf_sysenter_enable(event);
 789         case TRACE_REG_PERF_UNREGISTER:
 790                 perf_sysenter_disable(event);
 791                 return 0;
 792         case TRACE_REG_PERF_OPEN:
 793         case TRACE_REG_PERF_CLOSE:
 794         case TRACE_REG_PERF_ADD:
 795         case TRACE_REG_PERF_DEL:
 796                 return 0;
 797 #endif
 798         }
 799         return 0;
 800 }
 801
 802 static int syscall_exit_register(struct trace_event_call *event,
 803                                  enum trace_reg type, void *data)
 804 {
 805         struct trace_event_file *file = data;
 806
 807         switch (type) {
 808         case TRACE_REG_REGISTER:
 809                 return reg_event_syscall_exit(file, event);
 810         case TRACE_REG_UNREGISTER:
 811                 unreg_event_syscall_exit(file, event);
 812                 return 0;
 813
 814 #ifdef CONFIG_PERF_EVENTS
 815         case TRACE_REG_PERF_REGISTER:
 816                 return perf_sysexit_enable(event);
 817         case TRACE_REG_PERF_UNREGISTER:
 818                 perf_sysexit_disable(event);
 819                 return 0;
 820         case TRACE_REG_PERF_OPEN:
 821         case TRACE_REG_PERF_CLOSE:
 822         case TRACE_REG_PERF_ADD:
 823         case TRACE_REG_PERF_DEL:
 824                 return 0;
 825 #endif
 826         }
 827         return 0;
 828 }