kernel/trace/trace_syscalls.c

   1 #include <trace/syscall.h>
   2 #include <trace/events/syscalls.h>
   3 #include <linux/kernel.h>
   4 #include <linux/ftrace.h>
   5 #include <linux/perf_event.h>
   6 #include <asm/syscall.h>
   7
   8 #include "trace_output.h"
   9 #include "trace.h"
  10
  11 static DEFINE_MUTEX(syscall_trace_lock);
  12 static int sys_refcount_enter;
  13 static int sys_refcount_exit;
  14 static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
  15 static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
  16
  17 extern unsigned long __start_syscalls_metadata[];
  18 extern unsigned long __stop_syscalls_metadata[];
  19
  20 static struct syscall_metadata **syscalls_metadata;
  21
  22 static struct syscall_metadata *find_syscall_meta(unsigned long syscall)
  23 {
  24         struct syscall_metadata *start;
  25         struct syscall_metadata *stop;
  26         char str[KSYM_SYMBOL_LEN];
  27
  28
  29         start = (struct syscall_metadata *)__start_syscalls_metadata;
  30         stop = (struct syscall_metadata *)__stop_syscalls_metadata;
  31         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
  32
  33         for ( ; start < stop; start++) {
  34                 /*
  35                  * Only compare after the "sys" prefix. Archs that use
  36                  * syscall wrappers may have syscalls symbols aliases prefixed
  37                  * with "SyS" instead of "sys", leading to an unwanted
  38                  * mismatch.
  39                  */
  40                 if (start->name && !strcmp(start->name + 3, str + 3))
  41                         return start;
  42         }
  43         return NULL;
  44 }
  45
  46 static struct syscall_metadata *syscall_nr_to_meta(int nr)
  47 {
  48         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
  49                 return NULL;
  50
  51         return syscalls_metadata[nr];
  52 }
  53
  54 enum print_line_t
  55 print_syscall_enter(struct trace_iterator *iter, int flags)
  56 {
  57         struct trace_seq *s = &iter->seq;
  58         struct trace_entry *ent = iter->ent;
  59         struct syscall_trace_enter *trace;
  60         struct syscall_metadata *entry;
  61         int i, ret, syscall;
  62
  63         trace = (typeof(trace))ent;
  64         syscall = trace->nr;
  65         entry = syscall_nr_to_meta(syscall);
  66
  67         if (!entry)
  68                 goto end;
  69
  70         if (entry->enter_event->id != ent->type) {
  71                 WARN_ON_ONCE(1);
  72                 goto end;
  73         }
  74
  75         ret = trace_seq_printf(s, "%s(", entry->name);
  76         if (!ret)
  77                 return TRACE_TYPE_PARTIAL_LINE;
  78
  79         for (i = 0; i < entry->nb_args; i++) {
  80                 /* parameter types */
  81                 if (trace_flags & TRACE_ITER_VERBOSE) {
  82                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
  83                         if (!ret)
  84                                 return TRACE_TYPE_PARTIAL_LINE;
  85                 }
  86                 /* parameter values */
  87                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
  88                                        trace->args[i],
  89                                        i == entry->nb_args - 1 ? "" : ", ");
  90                 if (!ret)
  91                         return TRACE_TYPE_PARTIAL_LINE;
  92         }
  93
  94         ret = trace_seq_putc(s, ')');
  95         if (!ret)
  96                 return TRACE_TYPE_PARTIAL_LINE;
  97
  98 end:
  99         ret =  trace_seq_putc(s, '\n');
 100         if (!ret)
 101                 return TRACE_TYPE_PARTIAL_LINE;
 102
 103         return TRACE_TYPE_HANDLED;
 104 }
 105
 106 enum print_line_t
 107 print_syscall_exit(struct trace_iterator *iter, int flags)
 108 {
 109         struct trace_seq *s = &iter->seq;
 110         struct trace_entry *ent = iter->ent;
 111         struct syscall_trace_exit *trace;
 112         int syscall;
 113         struct syscall_metadata *entry;
 114         int ret;
 115
 116         trace = (typeof(trace))ent;
 117         syscall = trace->nr;
 118         entry = syscall_nr_to_meta(syscall);
 119
 120         if (!entry) {
 121                 trace_seq_printf(s, "\n");
 122                 return TRACE_TYPE_HANDLED;
 123         }
 124
 125         if (entry->exit_event->id != ent->type) {
 126                 WARN_ON_ONCE(1);
 127                 return TRACE_TYPE_UNHANDLED;
 128         }
 129
 130         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 131                                 trace->ret);
 132         if (!ret)
 133                 return TRACE_TYPE_PARTIAL_LINE;
 134
 135         return TRACE_TYPE_HANDLED;
 136 }
 137
 138 extern char *__bad_type_size(void);
 139
 140 #define SYSCALL_FIELD(type, name)                                       \
 141         sizeof(type) != sizeof(trace.name) ?                            \
 142                 __bad_type_size() :                                     \
 143                 #type, #name, offsetof(typeof(trace), name),            \
 144                 sizeof(trace.name), is_signed_type(type)
 145
 146 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
 147 {
 148         int i;
 149         int ret;
 150         struct syscall_metadata *entry = call->data;
 151         struct syscall_trace_enter trace;
 152         int offset = offsetof(struct syscall_trace_enter, args);
 153
 154         ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 155                                "\tsigned:%u;\n",
 156                                SYSCALL_FIELD(int, nr));
 157         if (!ret)
 158                 return 0;
 159
 160         for (i = 0; i < entry->nb_args; i++) {
 161                 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
 162                                         entry->args[i]);
 163                 if (!ret)
 164                         return 0;
 165                 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;"
 166                                        "\tsigned:%u;\n", offset,
 167                                        sizeof(unsigned long),
 168                                        is_signed_type(unsigned long));
 169                 if (!ret)
 170                         return 0;
 171                 offset += sizeof(unsigned long);
 172         }
 173
 174         trace_seq_puts(s, "\nprint fmt: \"");
 175         for (i = 0; i < entry->nb_args; i++) {
 176                 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
 177                                         sizeof(unsigned long),
 178                                         i == entry->nb_args - 1 ? "" : ", ");
 179                 if (!ret)
 180                         return 0;
 181         }
 182         trace_seq_putc(s, '"');
 183
 184         for (i = 0; i < entry->nb_args; i++) {
 185                 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
 186                                        entry->args[i]);
 187                 if (!ret)
 188                         return 0;
 189         }
 190
 191         return trace_seq_putc(s, '\n');
 192 }
 193
 194 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
 195 {
 196         int ret;
 197         struct syscall_trace_exit trace;
 198
 199         ret = trace_seq_printf(s,
 200                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 201                                "\tsigned:%u;\n"
 202                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 203                                "\tsigned:%u;\n",
 204                                SYSCALL_FIELD(int, nr),
 205                                SYSCALL_FIELD(long, ret));
 206         if (!ret)
 207                 return 0;
 208
 209         return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
 210 }
 211
 212 int syscall_enter_define_fields(struct ftrace_event_call *call)
 213 {
 214         struct syscall_trace_enter trace;
 215         struct syscall_metadata *meta = call->data;
 216         int ret;
 217         int i;
 218         int offset = offsetof(typeof(trace), args);
 219
 220         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 221         if (ret)
 222                 return ret;
 223
 224         for (i = 0; i < meta->nb_args; i++) {
 225                 ret = trace_define_field(call, meta->types[i],
 226                                          meta->args[i], offset,
 227                                          sizeof(unsigned long), 0,
 228                                          FILTER_OTHER);
 229                 offset += sizeof(unsigned long);
 230         }
 231
 232         return ret;
 233 }
 234
 235 int syscall_exit_define_fields(struct ftrace_event_call *call)
 236 {
 237         struct syscall_trace_exit trace;
 238         int ret;
 239
 240         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 241         if (ret)
 242                 return ret;
 243
 244         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 245                                  FILTER_OTHER);
 246
 247         return ret;
 248 }
 249
 250 void ftrace_syscall_enter(struct pt_regs *regs, long id)
 251 {
 252         struct syscall_trace_enter *entry;
 253         struct syscall_metadata *sys_data;
 254         struct ring_buffer_event *event;
 255         struct ring_buffer *buffer;
 256         int size;
 257         int syscall_nr;
 258
 259         syscall_nr = syscall_get_nr(current, regs);
 260         if (syscall_nr < 0)
 261                 return;
 262         if (!test_bit(syscall_nr, enabled_enter_syscalls))
 263                 return;
 264
 265         sys_data = syscall_nr_to_meta(syscall_nr);
 266         if (!sys_data)
 267                 return;
 268
 269         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 270
 271         event = trace_current_buffer_lock_reserve(&buffer,
 272                         sys_data->enter_event->id, size, 0, 0);
 273         if (!event)
 274                 return;
 275
 276         entry = ring_buffer_event_data(event);
 277         entry->nr = syscall_nr;
 278         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 279
 280         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 281                                           entry, event))
 282                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 283 }
 284
 285 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
 286 {
 287         struct syscall_trace_exit *entry;
 288         struct syscall_metadata *sys_data;
 289         struct ring_buffer_event *event;
 290         struct ring_buffer *buffer;
 291         int syscall_nr;
 292
 293         syscall_nr = syscall_get_nr(current, regs);
 294         if (syscall_nr < 0)
 295                 return;
 296         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 297                 return;
 298
 299         sys_data = syscall_nr_to_meta(syscall_nr);
 300         if (!sys_data)
 301                 return;
 302
 303         event = trace_current_buffer_lock_reserve(&buffer,
 304                         sys_data->exit_event->id, sizeof(*entry), 0, 0);
 305         if (!event)
 306                 return;
 307
 308         entry = ring_buffer_event_data(event);
 309         entry->nr = syscall_nr;
 310         entry->ret = syscall_get_return_value(current, regs);
 311
 312         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 313                                           entry, event))
 314                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 315 }
 316
 317 int reg_event_syscall_enter(struct ftrace_event_call *call)
 318 {
 319         int ret = 0;
 320         int num;
 321
 322         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 323         if (num < 0 || num >= NR_syscalls)
 324                 return -ENOSYS;
 325         mutex_lock(&syscall_trace_lock);
 326         if (!sys_refcount_enter)
 327                 ret = register_trace_sys_enter(ftrace_syscall_enter);
 328         if (ret) {
 329                 pr_info("event trace: Could not activate"
 330                                 "syscall entry trace point");
 331         } else {
 332                 set_bit(num, enabled_enter_syscalls);
 333                 sys_refcount_enter++;
 334         }
 335         mutex_unlock(&syscall_trace_lock);
 336         return ret;
 337 }
 338
 339 void unreg_event_syscall_enter(struct ftrace_event_call *call)
 340 {
 341         int num;
 342
 343         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 344         if (num < 0 || num >= NR_syscalls)
 345                 return;
 346         mutex_lock(&syscall_trace_lock);
 347         sys_refcount_enter--;
 348         clear_bit(num, enabled_enter_syscalls);
 349         if (!sys_refcount_enter)
 350                 unregister_trace_sys_enter(ftrace_syscall_enter);
 351         mutex_unlock(&syscall_trace_lock);
 352 }
 353
 354 int reg_event_syscall_exit(struct ftrace_event_call *call)
 355 {
 356         int ret = 0;
 357         int num;
 358
 359         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 360         if (num < 0 || num >= NR_syscalls)
 361                 return -ENOSYS;
 362         mutex_lock(&syscall_trace_lock);
 363         if (!sys_refcount_exit)
 364                 ret = register_trace_sys_exit(ftrace_syscall_exit);
 365         if (ret) {
 366                 pr_info("event trace: Could not activate"
 367                                 "syscall exit trace point");
 368         } else {
 369                 set_bit(num, enabled_exit_syscalls);
 370                 sys_refcount_exit++;
 371         }
 372         mutex_unlock(&syscall_trace_lock);
 373         return ret;
 374 }
 375
 376 void unreg_event_syscall_exit(struct ftrace_event_call *call)
 377 {
 378         int num;
 379
 380         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 381         if (num < 0 || num >= NR_syscalls)
 382                 return;
 383         mutex_lock(&syscall_trace_lock);
 384         sys_refcount_exit--;
 385         clear_bit(num, enabled_exit_syscalls);
 386         if (!sys_refcount_exit)
 387                 unregister_trace_sys_exit(ftrace_syscall_exit);
 388         mutex_unlock(&syscall_trace_lock);
 389 }
 390
 391 int init_syscall_trace(struct ftrace_event_call *call)
 392 {
 393         int id;
 394
 395         id = register_ftrace_event(call->event);
 396         if (!id)
 397                 return -ENODEV;
 398         call->id = id;
 399         INIT_LIST_HEAD(&call->fields);
 400         return 0;
 401 }
 402
 403 int __init init_ftrace_syscalls(void)
 404 {
 405         struct syscall_metadata *meta;
 406         unsigned long addr;
 407         int i;
 408
 409         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 410                                         NR_syscalls, GFP_KERNEL);
 411         if (!syscalls_metadata) {
 412                 WARN_ON(1);
 413                 return -ENOMEM;
 414         }
 415
 416         for (i = 0; i < NR_syscalls; i++) {
 417                 addr = arch_syscall_addr(i);
 418                 meta = find_syscall_meta(addr);
 419                 if (!meta)
 420                         continue;
 421
 422                 meta->syscall_nr = i;
 423                 syscalls_metadata[i] = meta;
 424         }
 425
 426         return 0;
 427 }
 428 core_initcall(init_ftrace_syscalls);
 429
 430 #ifdef CONFIG_EVENT_PROFILE
 431
 432 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 433 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 434 static int sys_prof_refcount_enter;
 435 static int sys_prof_refcount_exit;
 436
 437 static void prof_syscall_enter(struct pt_regs *regs, long id)
 438 {
 439         struct syscall_metadata *sys_data;
 440         struct syscall_trace_enter *rec;
 441         unsigned long flags;
 442         char *trace_buf;
 443         char *raw_data;
 444         int syscall_nr;
 445         int rctx;
 446         int size;
 447         int cpu;
 448
 449         syscall_nr = syscall_get_nr(current, regs);
 450         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 451                 return;
 452
 453         sys_data = syscall_nr_to_meta(syscall_nr);
 454         if (!sys_data)
 455                 return;
 456
 457         /* get the size after alignment with the u32 buffer size field */
 458         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 459         size = ALIGN(size + sizeof(u32), sizeof(u64));
 460         size -= sizeof(u32);
 461
 462         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 463                       "profile buffer not large enough"))
 464                 return;
 465
 466         /* Protect the per cpu buffer, begin the rcu read side */
 467         local_irq_save(flags);
 468
 469         rctx = perf_swevent_get_recursion_context();
 470         if (rctx < 0)
 471                 goto end_recursion;
 472
 473         cpu = smp_processor_id();
 474
 475         trace_buf = rcu_dereference(perf_trace_buf);
 476
 477         if (!trace_buf)
 478                 goto end;
 479
 480         raw_data = per_cpu_ptr(trace_buf, cpu);
 481
 482         /* zero the dead bytes from align to not leak stack to user */
 483         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 484
 485         rec = (struct syscall_trace_enter *) raw_data;
 486         tracing_generic_entry_update(&rec->ent, 0, 0);
 487         rec->ent.type = sys_data->enter_event->id;
 488         rec->nr = syscall_nr;
 489         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 490                                (unsigned long *)&rec->args);
 491         perf_tp_event(sys_data->enter_event->id, 0, 1, rec, size);
 492
 493 end:
 494         perf_swevent_put_recursion_context(rctx);
 495 end_recursion:
 496         local_irq_restore(flags);
 497 }
 498
 499 int prof_sysenter_enable(struct ftrace_event_call *call)
 500 {
 501         int ret = 0;
 502         int num;
 503
 504         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 505
 506         mutex_lock(&syscall_trace_lock);
 507         if (!sys_prof_refcount_enter)
 508                 ret = register_trace_sys_enter(prof_syscall_enter);
 509         if (ret) {
 510                 pr_info("event trace: Could not activate"
 511                                 "syscall entry trace point");
 512         } else {
 513                 set_bit(num, enabled_prof_enter_syscalls);
 514                 sys_prof_refcount_enter++;
 515         }
 516         mutex_unlock(&syscall_trace_lock);
 517         return ret;
 518 }
 519
 520 void prof_sysenter_disable(struct ftrace_event_call *call)
 521 {
 522         int num;
 523
 524         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 525
 526         mutex_lock(&syscall_trace_lock);
 527         sys_prof_refcount_enter--;
 528         clear_bit(num, enabled_prof_enter_syscalls);
 529         if (!sys_prof_refcount_enter)
 530                 unregister_trace_sys_enter(prof_syscall_enter);
 531         mutex_unlock(&syscall_trace_lock);
 532 }
 533
 534 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 535 {
 536         struct syscall_metadata *sys_data;
 537         struct syscall_trace_exit *rec;
 538         unsigned long flags;
 539         int syscall_nr;
 540         char *trace_buf;
 541         char *raw_data;
 542         int rctx;
 543         int size;
 544         int cpu;
 545
 546         syscall_nr = syscall_get_nr(current, regs);
 547         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 548                 return;
 549
 550         sys_data = syscall_nr_to_meta(syscall_nr);
 551         if (!sys_data)
 552                 return;
 553
 554         /* We can probably do that at build time */
 555         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 556         size -= sizeof(u32);
 557
 558         /*
 559          * Impossible, but be paranoid with the future
 560          * How to put this check outside runtime?
 561          */
 562         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 563                 "exit event has grown above profile buffer size"))
 564                 return;
 565
 566         /* Protect the per cpu buffer, begin the rcu read side */
 567         local_irq_save(flags);
 568
 569         rctx = perf_swevent_get_recursion_context();
 570         if (rctx < 0)
 571                 goto end_recursion;
 572
 573         cpu = smp_processor_id();
 574
 575         trace_buf = rcu_dereference(perf_trace_buf);
 576
 577         if (!trace_buf)
 578                 goto end;
 579
 580         raw_data = per_cpu_ptr(trace_buf, cpu);
 581
 582         /* zero the dead bytes from align to not leak stack to user */
 583         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 584
 585         rec = (struct syscall_trace_exit *)raw_data;
 586
 587         tracing_generic_entry_update(&rec->ent, 0, 0);
 588         rec->ent.type = sys_data->exit_event->id;
 589         rec->nr = syscall_nr;
 590         rec->ret = syscall_get_return_value(current, regs);
 591
 592         perf_tp_event(sys_data->exit_event->id, 0, 1, rec, size);
 593
 594 end:
 595         perf_swevent_put_recursion_context(rctx);
 596 end_recursion:
 597         local_irq_restore(flags);
 598 }
 599
 600 int prof_sysexit_enable(struct ftrace_event_call *call)
 601 {
 602         int ret = 0;
 603         int num;
 604
 605         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 606
 607         mutex_lock(&syscall_trace_lock);
 608         if (!sys_prof_refcount_exit)
 609                 ret = register_trace_sys_exit(prof_syscall_exit);
 610         if (ret) {
 611                 pr_info("event trace: Could not activate"
 612                                 "syscall entry trace point");
 613         } else {
 614                 set_bit(num, enabled_prof_exit_syscalls);
 615                 sys_prof_refcount_exit++;
 616         }
 617         mutex_unlock(&syscall_trace_lock);
 618         return ret;
 619 }
 620
 621 void prof_sysexit_disable(struct ftrace_event_call *call)
 622 {
 623         int num;
 624
 625         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 626
 627         mutex_lock(&syscall_trace_lock);
 628         sys_prof_refcount_exit--;
 629         clear_bit(num, enabled_prof_exit_syscalls);
 630         if (!sys_prof_refcount_exit)
 631                 unregister_trace_sys_exit(prof_syscall_exit);
 632         mutex_unlock(&syscall_trace_lock);
 633 }
 634
 635 #endif
 636
 637