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_common_fields(call);
 221         if (ret)
 222                 return ret;
 223
 224         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 225         if (ret)
 226                 return ret;
 227
 228         for (i = 0; i < meta->nb_args; i++) {
 229                 ret = trace_define_field(call, meta->types[i],
 230                                          meta->args[i], offset,
 231                                          sizeof(unsigned long), 0,
 232                                          FILTER_OTHER);
 233                 offset += sizeof(unsigned long);
 234         }
 235
 236         return ret;
 237 }
 238
 239 int syscall_exit_define_fields(struct ftrace_event_call *call)
 240 {
 241         struct syscall_trace_exit trace;
 242         int ret;
 243
 244         ret = trace_define_common_fields(call);
 245         if (ret)
 246                 return ret;
 247
 248         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 249         if (ret)
 250                 return ret;
 251
 252         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 253                                  FILTER_OTHER);
 254
 255         return ret;
 256 }
 257
 258 void ftrace_syscall_enter(struct pt_regs *regs, long id)
 259 {
 260         struct syscall_trace_enter *entry;
 261         struct syscall_metadata *sys_data;
 262         struct ring_buffer_event *event;
 263         struct ring_buffer *buffer;
 264         int size;
 265         int syscall_nr;
 266
 267         syscall_nr = syscall_get_nr(current, regs);
 268         if (syscall_nr < 0)
 269                 return;
 270         if (!test_bit(syscall_nr, enabled_enter_syscalls))
 271                 return;
 272
 273         sys_data = syscall_nr_to_meta(syscall_nr);
 274         if (!sys_data)
 275                 return;
 276
 277         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 278
 279         event = trace_current_buffer_lock_reserve(&buffer,
 280                         sys_data->enter_event->id, size, 0, 0);
 281         if (!event)
 282                 return;
 283
 284         entry = ring_buffer_event_data(event);
 285         entry->nr = syscall_nr;
 286         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 287
 288         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 289                                           entry, event))
 290                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 291 }
 292
 293 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
 294 {
 295         struct syscall_trace_exit *entry;
 296         struct syscall_metadata *sys_data;
 297         struct ring_buffer_event *event;
 298         struct ring_buffer *buffer;
 299         int syscall_nr;
 300
 301         syscall_nr = syscall_get_nr(current, regs);
 302         if (syscall_nr < 0)
 303                 return;
 304         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 305                 return;
 306
 307         sys_data = syscall_nr_to_meta(syscall_nr);
 308         if (!sys_data)
 309                 return;
 310
 311         event = trace_current_buffer_lock_reserve(&buffer,
 312                         sys_data->exit_event->id, sizeof(*entry), 0, 0);
 313         if (!event)
 314                 return;
 315
 316         entry = ring_buffer_event_data(event);
 317         entry->nr = syscall_nr;
 318         entry->ret = syscall_get_return_value(current, regs);
 319
 320         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 321                                           entry, event))
 322                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 323 }
 324
 325 int reg_event_syscall_enter(struct ftrace_event_call *call)
 326 {
 327         int ret = 0;
 328         int num;
 329
 330         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 331         if (num < 0 || num >= NR_syscalls)
 332                 return -ENOSYS;
 333         mutex_lock(&syscall_trace_lock);
 334         if (!sys_refcount_enter)
 335                 ret = register_trace_sys_enter(ftrace_syscall_enter);
 336         if (ret) {
 337                 pr_info("event trace: Could not activate"
 338                                 "syscall entry trace point");
 339         } else {
 340                 set_bit(num, enabled_enter_syscalls);
 341                 sys_refcount_enter++;
 342         }
 343         mutex_unlock(&syscall_trace_lock);
 344         return ret;
 345 }
 346
 347 void unreg_event_syscall_enter(struct ftrace_event_call *call)
 348 {
 349         int num;
 350
 351         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 352         if (num < 0 || num >= NR_syscalls)
 353                 return;
 354         mutex_lock(&syscall_trace_lock);
 355         sys_refcount_enter--;
 356         clear_bit(num, enabled_enter_syscalls);
 357         if (!sys_refcount_enter)
 358                 unregister_trace_sys_enter(ftrace_syscall_enter);
 359         mutex_unlock(&syscall_trace_lock);
 360 }
 361
 362 int reg_event_syscall_exit(struct ftrace_event_call *call)
 363 {
 364         int ret = 0;
 365         int num;
 366
 367         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 368         if (num < 0 || num >= NR_syscalls)
 369                 return -ENOSYS;
 370         mutex_lock(&syscall_trace_lock);
 371         if (!sys_refcount_exit)
 372                 ret = register_trace_sys_exit(ftrace_syscall_exit);
 373         if (ret) {
 374                 pr_info("event trace: Could not activate"
 375                                 "syscall exit trace point");
 376         } else {
 377                 set_bit(num, enabled_exit_syscalls);
 378                 sys_refcount_exit++;
 379         }
 380         mutex_unlock(&syscall_trace_lock);
 381         return ret;
 382 }
 383
 384 void unreg_event_syscall_exit(struct ftrace_event_call *call)
 385 {
 386         int num;
 387
 388         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 389         if (num < 0 || num >= NR_syscalls)
 390                 return;
 391         mutex_lock(&syscall_trace_lock);
 392         sys_refcount_exit--;
 393         clear_bit(num, enabled_exit_syscalls);
 394         if (!sys_refcount_exit)
 395                 unregister_trace_sys_exit(ftrace_syscall_exit);
 396         mutex_unlock(&syscall_trace_lock);
 397 }
 398
 399 int init_syscall_trace(struct ftrace_event_call *call)
 400 {
 401         int id;
 402
 403         id = register_ftrace_event(call->event);
 404         if (!id)
 405                 return -ENODEV;
 406         call->id = id;
 407         INIT_LIST_HEAD(&call->fields);
 408         return 0;
 409 }
 410
 411 int __init init_ftrace_syscalls(void)
 412 {
 413         struct syscall_metadata *meta;
 414         unsigned long addr;
 415         int i;
 416
 417         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 418                                         NR_syscalls, GFP_KERNEL);
 419         if (!syscalls_metadata) {
 420                 WARN_ON(1);
 421                 return -ENOMEM;
 422         }
 423
 424         for (i = 0; i < NR_syscalls; i++) {
 425                 addr = arch_syscall_addr(i);
 426                 meta = find_syscall_meta(addr);
 427                 if (!meta)
 428                         continue;
 429
 430                 meta->syscall_nr = i;
 431                 syscalls_metadata[i] = meta;
 432         }
 433
 434         return 0;
 435 }
 436 core_initcall(init_ftrace_syscalls);
 437
 438 #ifdef CONFIG_EVENT_PROFILE
 439
 440 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 441 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 442 static int sys_prof_refcount_enter;
 443 static int sys_prof_refcount_exit;
 444
 445 static void prof_syscall_enter(struct pt_regs *regs, long id)
 446 {
 447         struct syscall_metadata *sys_data;
 448         struct syscall_trace_enter *rec;
 449         unsigned long flags;
 450         char *trace_buf;
 451         char *raw_data;
 452         int syscall_nr;
 453         int rctx;
 454         int size;
 455         int cpu;
 456
 457         syscall_nr = syscall_get_nr(current, regs);
 458         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 459                 return;
 460
 461         sys_data = syscall_nr_to_meta(syscall_nr);
 462         if (!sys_data)
 463                 return;
 464
 465         /* get the size after alignment with the u32 buffer size field */
 466         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 467         size = ALIGN(size + sizeof(u32), sizeof(u64));
 468         size -= sizeof(u32);
 469
 470         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 471                       "profile buffer not large enough"))
 472                 return;
 473
 474         /* Protect the per cpu buffer, begin the rcu read side */
 475         local_irq_save(flags);
 476
 477         rctx = perf_swevent_get_recursion_context();
 478         if (rctx < 0)
 479                 goto end_recursion;
 480
 481         cpu = smp_processor_id();
 482
 483         trace_buf = rcu_dereference(perf_trace_buf);
 484
 485         if (!trace_buf)
 486                 goto end;
 487
 488         raw_data = per_cpu_ptr(trace_buf, cpu);
 489
 490         /* zero the dead bytes from align to not leak stack to user */
 491         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 492
 493         rec = (struct syscall_trace_enter *) raw_data;
 494         tracing_generic_entry_update(&rec->ent, 0, 0);
 495         rec->ent.type = sys_data->enter_event->id;
 496         rec->nr = syscall_nr;
 497         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 498                                (unsigned long *)&rec->args);
 499         perf_tp_event(sys_data->enter_event->id, 0, 1, rec, size);
 500
 501 end:
 502         perf_swevent_put_recursion_context(rctx);
 503 end_recursion:
 504         local_irq_restore(flags);
 505 }
 506
 507 int prof_sysenter_enable(struct ftrace_event_call *call)
 508 {
 509         int ret = 0;
 510         int num;
 511
 512         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 513
 514         mutex_lock(&syscall_trace_lock);
 515         if (!sys_prof_refcount_enter)
 516                 ret = register_trace_sys_enter(prof_syscall_enter);
 517         if (ret) {
 518                 pr_info("event trace: Could not activate"
 519                                 "syscall entry trace point");
 520         } else {
 521                 set_bit(num, enabled_prof_enter_syscalls);
 522                 sys_prof_refcount_enter++;
 523         }
 524         mutex_unlock(&syscall_trace_lock);
 525         return ret;
 526 }
 527
 528 void prof_sysenter_disable(struct ftrace_event_call *call)
 529 {
 530         int num;
 531
 532         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 533
 534         mutex_lock(&syscall_trace_lock);
 535         sys_prof_refcount_enter--;
 536         clear_bit(num, enabled_prof_enter_syscalls);
 537         if (!sys_prof_refcount_enter)
 538                 unregister_trace_sys_enter(prof_syscall_enter);
 539         mutex_unlock(&syscall_trace_lock);
 540 }
 541
 542 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 543 {
 544         struct syscall_metadata *sys_data;
 545         struct syscall_trace_exit *rec;
 546         unsigned long flags;
 547         int syscall_nr;
 548         char *trace_buf;
 549         char *raw_data;
 550         int rctx;
 551         int size;
 552         int cpu;
 553
 554         syscall_nr = syscall_get_nr(current, regs);
 555         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 556                 return;
 557
 558         sys_data = syscall_nr_to_meta(syscall_nr);
 559         if (!sys_data)
 560                 return;
 561
 562         /* We can probably do that at build time */
 563         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 564         size -= sizeof(u32);
 565
 566         /*
 567          * Impossible, but be paranoid with the future
 568          * How to put this check outside runtime?
 569          */
 570         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 571                 "exit event has grown above profile buffer size"))
 572                 return;
 573
 574         /* Protect the per cpu buffer, begin the rcu read side */
 575         local_irq_save(flags);
 576
 577         rctx = perf_swevent_get_recursion_context();
 578         if (rctx < 0)
 579                 goto end_recursion;
 580
 581         cpu = smp_processor_id();
 582
 583         trace_buf = rcu_dereference(perf_trace_buf);
 584
 585         if (!trace_buf)
 586                 goto end;
 587
 588         raw_data = per_cpu_ptr(trace_buf, cpu);
 589
 590         /* zero the dead bytes from align to not leak stack to user */
 591         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 592
 593         rec = (struct syscall_trace_exit *)raw_data;
 594
 595         tracing_generic_entry_update(&rec->ent, 0, 0);
 596         rec->ent.type = sys_data->exit_event->id;
 597         rec->nr = syscall_nr;
 598         rec->ret = syscall_get_return_value(current, regs);
 599
 600         perf_tp_event(sys_data->exit_event->id, 0, 1, rec, size);
 601
 602 end:
 603         perf_swevent_put_recursion_context(rctx);
 604 end_recursion:
 605         local_irq_restore(flags);
 606 }
 607
 608 int prof_sysexit_enable(struct ftrace_event_call *call)
 609 {
 610         int ret = 0;
 611         int num;
 612
 613         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 614
 615         mutex_lock(&syscall_trace_lock);
 616         if (!sys_prof_refcount_exit)
 617                 ret = register_trace_sys_exit(prof_syscall_exit);
 618         if (ret) {
 619                 pr_info("event trace: Could not activate"
 620                                 "syscall entry trace point");
 621         } else {
 622                 set_bit(num, enabled_prof_exit_syscalls);
 623                 sys_prof_refcount_exit++;
 624         }
 625         mutex_unlock(&syscall_trace_lock);
 626         return ret;
 627 }
 628
 629 void prof_sysexit_disable(struct ftrace_event_call *call)
 630 {
 631         int num;
 632
 633         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 634
 635         mutex_lock(&syscall_trace_lock);
 636         sys_prof_refcount_exit--;
 637         clear_bit(num, enabled_prof_exit_syscalls);
 638         if (!sys_prof_refcount_exit)
 639                 unregister_trace_sys_exit(prof_syscall_exit);
 640         mutex_unlock(&syscall_trace_lock);
 641 }
 642
 643 #endif
 644
 645