SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 int syscall_name_to_nr(char *name)
  55 {
  56         int i;
  57
  58         if (!syscalls_metadata)
  59                 return -1;
  60
  61         for (i = 0; i < NR_syscalls; i++) {
  62                 if (syscalls_metadata[i]) {
  63                         if (!strcmp(syscalls_metadata[i]->name, name))
  64                                 return i;
  65                 }
  66         }
  67         return -1;
  68 }
  69
  70 void set_syscall_enter_id(int num, int id)
  71 {
  72         syscalls_metadata[num]->enter_id = id;
  73 }
  74
  75 void set_syscall_exit_id(int num, int id)
  76 {
  77         syscalls_metadata[num]->exit_id = id;
  78 }
  79
  80 enum print_line_t
  81 print_syscall_enter(struct trace_iterator *iter, int flags)
  82 {
  83         struct trace_seq *s = &iter->seq;
  84         struct trace_entry *ent = iter->ent;
  85         struct syscall_trace_enter *trace;
  86         struct syscall_metadata *entry;
  87         int i, ret, syscall;
  88
  89         trace = (typeof(trace))ent;
  90         syscall = trace->nr;
  91         entry = syscall_nr_to_meta(syscall);
  92
  93         if (!entry)
  94                 goto end;
  95
  96         if (entry->enter_id != ent->type) {
  97                 WARN_ON_ONCE(1);
  98                 goto end;
  99         }
 100
 101         ret = trace_seq_printf(s, "%s(", entry->name);
 102         if (!ret)
 103                 return TRACE_TYPE_PARTIAL_LINE;
 104
 105         for (i = 0; i < entry->nb_args; i++) {
 106                 /* parameter types */
 107                 if (trace_flags & TRACE_ITER_VERBOSE) {
 108                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
 109                         if (!ret)
 110                                 return TRACE_TYPE_PARTIAL_LINE;
 111                 }
 112                 /* parameter values */
 113                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 114                                        trace->args[i],
 115                                        i == entry->nb_args - 1 ? "" : ", ");
 116                 if (!ret)
 117                         return TRACE_TYPE_PARTIAL_LINE;
 118         }
 119
 120         ret = trace_seq_putc(s, ')');
 121         if (!ret)
 122                 return TRACE_TYPE_PARTIAL_LINE;
 123
 124 end:
 125         ret =  trace_seq_putc(s, '\n');
 126         if (!ret)
 127                 return TRACE_TYPE_PARTIAL_LINE;
 128
 129         return TRACE_TYPE_HANDLED;
 130 }
 131
 132 enum print_line_t
 133 print_syscall_exit(struct trace_iterator *iter, int flags)
 134 {
 135         struct trace_seq *s = &iter->seq;
 136         struct trace_entry *ent = iter->ent;
 137         struct syscall_trace_exit *trace;
 138         int syscall;
 139         struct syscall_metadata *entry;
 140         int ret;
 141
 142         trace = (typeof(trace))ent;
 143         syscall = trace->nr;
 144         entry = syscall_nr_to_meta(syscall);
 145
 146         if (!entry) {
 147                 trace_seq_printf(s, "\n");
 148                 return TRACE_TYPE_HANDLED;
 149         }
 150
 151         if (entry->exit_id != ent->type) {
 152                 WARN_ON_ONCE(1);
 153                 return TRACE_TYPE_UNHANDLED;
 154         }
 155
 156         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 157                                 trace->ret);
 158         if (!ret)
 159                 return TRACE_TYPE_PARTIAL_LINE;
 160
 161         return TRACE_TYPE_HANDLED;
 162 }
 163
 164 extern char *__bad_type_size(void);
 165
 166 #define SYSCALL_FIELD(type, name)                                       \
 167         sizeof(type) != sizeof(trace.name) ?                            \
 168                 __bad_type_size() :                                     \
 169                 #type, #name, offsetof(typeof(trace), name), sizeof(trace.name)
 170
 171 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
 172 {
 173         int i;
 174         int nr;
 175         int ret;
 176         struct syscall_metadata *entry;
 177         struct syscall_trace_enter trace;
 178         int offset = offsetof(struct syscall_trace_enter, args);
 179
 180         nr = syscall_name_to_nr(call->data);
 181         entry = syscall_nr_to_meta(nr);
 182
 183         if (!entry)
 184                 return 0;
 185
 186         ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n",
 187                                SYSCALL_FIELD(int, nr));
 188         if (!ret)
 189                 return 0;
 190
 191         for (i = 0; i < entry->nb_args; i++) {
 192                 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
 193                                         entry->args[i]);
 194                 if (!ret)
 195                         return 0;
 196                 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;\n", offset,
 197                                        sizeof(unsigned long));
 198                 if (!ret)
 199                         return 0;
 200                 offset += sizeof(unsigned long);
 201         }
 202
 203         trace_seq_puts(s, "\nprint fmt: \"");
 204         for (i = 0; i < entry->nb_args; i++) {
 205                 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
 206                                         sizeof(unsigned long),
 207                                         i == entry->nb_args - 1 ? "" : ", ");
 208                 if (!ret)
 209                         return 0;
 210         }
 211         trace_seq_putc(s, '"');
 212
 213         for (i = 0; i < entry->nb_args; i++) {
 214                 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
 215                                        entry->args[i]);
 216                 if (!ret)
 217                         return 0;
 218         }
 219
 220         return trace_seq_putc(s, '\n');
 221 }
 222
 223 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
 224 {
 225         int ret;
 226         struct syscall_trace_exit trace;
 227
 228         ret = trace_seq_printf(s,
 229                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
 230                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n",
 231                                SYSCALL_FIELD(int, nr),
 232                                SYSCALL_FIELD(unsigned long, ret));
 233         if (!ret)
 234                 return 0;
 235
 236         return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
 237 }
 238
 239 int syscall_enter_define_fields(struct ftrace_event_call *call)
 240 {
 241         struct syscall_trace_enter trace;
 242         struct syscall_metadata *meta;
 243         int ret;
 244         int nr;
 245         int i;
 246         int offset = offsetof(typeof(trace), args);
 247
 248         nr = syscall_name_to_nr(call->data);
 249         meta = syscall_nr_to_meta(nr);
 250
 251         if (!meta)
 252                 return 0;
 253
 254         ret = trace_define_common_fields(call);
 255         if (ret)
 256                 return ret;
 257
 258         for (i = 0; i < meta->nb_args; i++) {
 259                 ret = trace_define_field(call, meta->types[i],
 260                                          meta->args[i], offset,
 261                                          sizeof(unsigned long), 0,
 262                                          FILTER_OTHER);
 263                 offset += sizeof(unsigned long);
 264         }
 265
 266         return ret;
 267 }
 268
 269 int syscall_exit_define_fields(struct ftrace_event_call *call)
 270 {
 271         struct syscall_trace_exit trace;
 272         int ret;
 273
 274         ret = trace_define_common_fields(call);
 275         if (ret)
 276                 return ret;
 277
 278         ret = trace_define_field(call, SYSCALL_FIELD(unsigned long, ret), 0,
 279                                  FILTER_OTHER);
 280
 281         return ret;
 282 }
 283
 284 void ftrace_syscall_enter(struct pt_regs *regs, long id)
 285 {
 286         struct syscall_trace_enter *entry;
 287         struct syscall_metadata *sys_data;
 288         struct ring_buffer_event *event;
 289         struct ring_buffer *buffer;
 290         int size;
 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_enter_syscalls))
 297                 return;
 298
 299         sys_data = syscall_nr_to_meta(syscall_nr);
 300         if (!sys_data)
 301                 return;
 302
 303         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 304
 305         event = trace_current_buffer_lock_reserve(&buffer, sys_data->enter_id,
 306                                                   size, 0, 0);
 307         if (!event)
 308                 return;
 309
 310         entry = ring_buffer_event_data(event);
 311         entry->nr = syscall_nr;
 312         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 313
 314         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 315                                           entry, event))
 316                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 317 }
 318
 319 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
 320 {
 321         struct syscall_trace_exit *entry;
 322         struct syscall_metadata *sys_data;
 323         struct ring_buffer_event *event;
 324         struct ring_buffer *buffer;
 325         int syscall_nr;
 326
 327         syscall_nr = syscall_get_nr(current, regs);
 328         if (syscall_nr < 0)
 329                 return;
 330         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 331                 return;
 332
 333         sys_data = syscall_nr_to_meta(syscall_nr);
 334         if (!sys_data)
 335                 return;
 336
 337         event = trace_current_buffer_lock_reserve(&buffer, sys_data->exit_id,
 338                                 sizeof(*entry), 0, 0);
 339         if (!event)
 340                 return;
 341
 342         entry = ring_buffer_event_data(event);
 343         entry->nr = syscall_nr;
 344         entry->ret = syscall_get_return_value(current, regs);
 345
 346         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 347                                           entry, event))
 348                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 349 }
 350
 351 int reg_event_syscall_enter(void *ptr)
 352 {
 353         int ret = 0;
 354         int num;
 355         char *name;
 356
 357         name = (char *)ptr;
 358         num = syscall_name_to_nr(name);
 359         if (num < 0 || num >= NR_syscalls)
 360                 return -ENOSYS;
 361         mutex_lock(&syscall_trace_lock);
 362         if (!sys_refcount_enter)
 363                 ret = register_trace_sys_enter(ftrace_syscall_enter);
 364         if (ret) {
 365                 pr_info("event trace: Could not activate"
 366                                 "syscall entry trace point");
 367         } else {
 368                 set_bit(num, enabled_enter_syscalls);
 369                 sys_refcount_enter++;
 370         }
 371         mutex_unlock(&syscall_trace_lock);
 372         return ret;
 373 }
 374
 375 void unreg_event_syscall_enter(void *ptr)
 376 {
 377         int num;
 378         char *name;
 379
 380         name = (char *)ptr;
 381         num = syscall_name_to_nr(name);
 382         if (num < 0 || num >= NR_syscalls)
 383                 return;
 384         mutex_lock(&syscall_trace_lock);
 385         sys_refcount_enter--;
 386         clear_bit(num, enabled_enter_syscalls);
 387         if (!sys_refcount_enter)
 388                 unregister_trace_sys_enter(ftrace_syscall_enter);
 389         mutex_unlock(&syscall_trace_lock);
 390 }
 391
 392 int reg_event_syscall_exit(void *ptr)
 393 {
 394         int ret = 0;
 395         int num;
 396         char *name;
 397
 398         name = (char *)ptr;
 399         num = syscall_name_to_nr(name);
 400         if (num < 0 || num >= NR_syscalls)
 401                 return -ENOSYS;
 402         mutex_lock(&syscall_trace_lock);
 403         if (!sys_refcount_exit)
 404                 ret = register_trace_sys_exit(ftrace_syscall_exit);
 405         if (ret) {
 406                 pr_info("event trace: Could not activate"
 407                                 "syscall exit trace point");
 408         } else {
 409                 set_bit(num, enabled_exit_syscalls);
 410                 sys_refcount_exit++;
 411         }
 412         mutex_unlock(&syscall_trace_lock);
 413         return ret;
 414 }
 415
 416 void unreg_event_syscall_exit(void *ptr)
 417 {
 418         int num;
 419         char *name;
 420
 421         name = (char *)ptr;
 422         num = syscall_name_to_nr(name);
 423         if (num < 0 || num >= NR_syscalls)
 424                 return;
 425         mutex_lock(&syscall_trace_lock);
 426         sys_refcount_exit--;
 427         clear_bit(num, enabled_exit_syscalls);
 428         if (!sys_refcount_exit)
 429                 unregister_trace_sys_exit(ftrace_syscall_exit);
 430         mutex_unlock(&syscall_trace_lock);
 431 }
 432
 433 struct trace_event event_syscall_enter = {
 434         .trace                  = print_syscall_enter,
 435 };
 436
 437 struct trace_event event_syscall_exit = {
 438         .trace                  = print_syscall_exit,
 439 };
 440
 441 int __init init_ftrace_syscalls(void)
 442 {
 443         struct syscall_metadata *meta;
 444         unsigned long addr;
 445         int i;
 446
 447         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 448                                         NR_syscalls, GFP_KERNEL);
 449         if (!syscalls_metadata) {
 450                 WARN_ON(1);
 451                 return -ENOMEM;
 452         }
 453
 454         for (i = 0; i < NR_syscalls; i++) {
 455                 addr = arch_syscall_addr(i);
 456                 meta = find_syscall_meta(addr);
 457                 syscalls_metadata[i] = meta;
 458         }
 459
 460         return 0;
 461 }
 462 core_initcall(init_ftrace_syscalls);
 463
 464 #ifdef CONFIG_EVENT_PROFILE
 465
 466 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 467 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 468 static int sys_prof_refcount_enter;
 469 static int sys_prof_refcount_exit;
 470
 471 static void prof_syscall_enter(struct pt_regs *regs, long id)
 472 {
 473         struct syscall_metadata *sys_data;
 474         struct syscall_trace_enter *rec;
 475         unsigned long flags;
 476         char *raw_data;
 477         int syscall_nr;
 478         int size;
 479         int cpu;
 480
 481         syscall_nr = syscall_get_nr(current, regs);
 482         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 483                 return;
 484
 485         sys_data = syscall_nr_to_meta(syscall_nr);
 486         if (!sys_data)
 487                 return;
 488
 489         /* get the size after alignment with the u32 buffer size field */
 490         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 491         size = ALIGN(size + sizeof(u32), sizeof(u64));
 492         size -= sizeof(u32);
 493
 494         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 495                       "profile buffer not large enough"))
 496                 return;
 497
 498         /* Protect the per cpu buffer, begin the rcu read side */
 499         local_irq_save(flags);
 500
 501         cpu = smp_processor_id();
 502
 503         if (in_nmi())
 504                 raw_data = rcu_dereference(trace_profile_buf_nmi);
 505         else
 506                 raw_data = rcu_dereference(trace_profile_buf);
 507
 508         if (!raw_data)
 509                 goto end;
 510
 511         raw_data = per_cpu_ptr(raw_data, cpu);
 512
 513         /* zero the dead bytes from align to not leak stack to user */
 514         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 515
 516         rec = (struct syscall_trace_enter *) raw_data;
 517         tracing_generic_entry_update(&rec->ent, 0, 0);
 518         rec->ent.type = sys_data->enter_id;
 519         rec->nr = syscall_nr;
 520         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 521                                (unsigned long *)&rec->args);
 522         perf_tp_event(sys_data->enter_id, 0, 1, rec, size);
 523
 524 end:
 525         local_irq_restore(flags);
 526 }
 527
 528 int reg_prof_syscall_enter(char *name)
 529 {
 530         int ret = 0;
 531         int num;
 532
 533         num = syscall_name_to_nr(name);
 534         if (num < 0 || num >= NR_syscalls)
 535                 return -ENOSYS;
 536
 537         mutex_lock(&syscall_trace_lock);
 538         if (!sys_prof_refcount_enter)
 539                 ret = register_trace_sys_enter(prof_syscall_enter);
 540         if (ret) {
 541                 pr_info("event trace: Could not activate"
 542                                 "syscall entry trace point");
 543         } else {
 544                 set_bit(num, enabled_prof_enter_syscalls);
 545                 sys_prof_refcount_enter++;
 546         }
 547         mutex_unlock(&syscall_trace_lock);
 548         return ret;
 549 }
 550
 551 void unreg_prof_syscall_enter(char *name)
 552 {
 553         int num;
 554
 555         num = syscall_name_to_nr(name);
 556         if (num < 0 || num >= NR_syscalls)
 557                 return;
 558
 559         mutex_lock(&syscall_trace_lock);
 560         sys_prof_refcount_enter--;
 561         clear_bit(num, enabled_prof_enter_syscalls);
 562         if (!sys_prof_refcount_enter)
 563                 unregister_trace_sys_enter(prof_syscall_enter);
 564         mutex_unlock(&syscall_trace_lock);
 565 }
 566
 567 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 568 {
 569         struct syscall_metadata *sys_data;
 570         struct syscall_trace_exit *rec;
 571         unsigned long flags;
 572         int syscall_nr;
 573         char *raw_data;
 574         int size;
 575         int cpu;
 576
 577         syscall_nr = syscall_get_nr(current, regs);
 578         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 579                 return;
 580
 581         sys_data = syscall_nr_to_meta(syscall_nr);
 582         if (!sys_data)
 583                 return;
 584
 585         /* We can probably do that at build time */
 586         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 587         size -= sizeof(u32);
 588
 589         /*
 590          * Impossible, but be paranoid with the future
 591          * How to put this check outside runtime?
 592          */
 593         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 594                 "exit event has grown above profile buffer size"))
 595                 return;
 596
 597         /* Protect the per cpu buffer, begin the rcu read side */
 598         local_irq_save(flags);
 599         cpu = smp_processor_id();
 600
 601         if (in_nmi())
 602                 raw_data = rcu_dereference(trace_profile_buf_nmi);
 603         else
 604                 raw_data = rcu_dereference(trace_profile_buf);
 605
 606         if (!raw_data)
 607                 goto end;
 608
 609         raw_data = per_cpu_ptr(raw_data, cpu);
 610
 611         /* zero the dead bytes from align to not leak stack to user */
 612         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 613
 614         rec = (struct syscall_trace_exit *)raw_data;
 615
 616         tracing_generic_entry_update(&rec->ent, 0, 0);
 617         rec->ent.type = sys_data->exit_id;
 618         rec->nr = syscall_nr;
 619         rec->ret = syscall_get_return_value(current, regs);
 620
 621         perf_tp_event(sys_data->exit_id, 0, 1, rec, size);
 622
 623 end:
 624         local_irq_restore(flags);
 625 }
 626
 627 int reg_prof_syscall_exit(char *name)
 628 {
 629         int ret = 0;
 630         int num;
 631
 632         num = syscall_name_to_nr(name);
 633         if (num < 0 || num >= NR_syscalls)
 634                 return -ENOSYS;
 635
 636         mutex_lock(&syscall_trace_lock);
 637         if (!sys_prof_refcount_exit)
 638                 ret = register_trace_sys_exit(prof_syscall_exit);
 639         if (ret) {
 640                 pr_info("event trace: Could not activate"
 641                                 "syscall entry trace point");
 642         } else {
 643                 set_bit(num, enabled_prof_exit_syscalls);
 644                 sys_prof_refcount_exit++;
 645         }
 646         mutex_unlock(&syscall_trace_lock);
 647         return ret;
 648 }
 649
 650 void unreg_prof_syscall_exit(char *name)
 651 {
 652         int num;
 653
 654         num = syscall_name_to_nr(name);
 655         if (num < 0 || num >= NR_syscalls)
 656                 return;
 657
 658         mutex_lock(&syscall_trace_lock);
 659         sys_prof_refcount_exit--;
 660         clear_bit(num, enabled_prof_exit_syscalls);
 661         if (!sys_prof_refcount_exit)
 662                 unregister_trace_sys_exit(prof_syscall_exit);
 663         mutex_unlock(&syscall_trace_lock);
 664 }
 665
 666 #endif
 667
 668