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(const 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),            \
 170                 sizeof(trace.name), is_signed_type(type)
 171
 172 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
 173 {
 174         int i;
 175         int ret;
 176         struct syscall_metadata *entry = call->data;
 177         struct syscall_trace_enter trace;
 178         int offset = offsetof(struct syscall_trace_enter, args);
 179
 180         ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 181                                "\tsigned:%u;\n",
 182                                SYSCALL_FIELD(int, nr));
 183         if (!ret)
 184                 return 0;
 185
 186         for (i = 0; i < entry->nb_args; i++) {
 187                 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
 188                                         entry->args[i]);
 189                 if (!ret)
 190                         return 0;
 191                 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;"
 192                                        "\tsigned:%u;\n", offset,
 193                                        sizeof(unsigned long),
 194                                        is_signed_type(unsigned long));
 195                 if (!ret)
 196                         return 0;
 197                 offset += sizeof(unsigned long);
 198         }
 199
 200         trace_seq_puts(s, "\nprint fmt: \"");
 201         for (i = 0; i < entry->nb_args; i++) {
 202                 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
 203                                         sizeof(unsigned long),
 204                                         i == entry->nb_args - 1 ? "" : ", ");
 205                 if (!ret)
 206                         return 0;
 207         }
 208         trace_seq_putc(s, '"');
 209
 210         for (i = 0; i < entry->nb_args; i++) {
 211                 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
 212                                        entry->args[i]);
 213                 if (!ret)
 214                         return 0;
 215         }
 216
 217         return trace_seq_putc(s, '\n');
 218 }
 219
 220 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
 221 {
 222         int ret;
 223         struct syscall_trace_exit trace;
 224
 225         ret = trace_seq_printf(s,
 226                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 227                                "\tsigned:%u;\n"
 228                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 229                                "\tsigned:%u;\n",
 230                                SYSCALL_FIELD(int, nr),
 231                                SYSCALL_FIELD(long, ret));
 232         if (!ret)
 233                 return 0;
 234
 235         return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
 236 }
 237
 238 int syscall_enter_define_fields(struct ftrace_event_call *call)
 239 {
 240         struct syscall_trace_enter trace;
 241         struct syscall_metadata *meta = call->data;
 242         int ret;
 243         int i;
 244         int offset = offsetof(typeof(trace), args);
 245
 246         ret = trace_define_common_fields(call);
 247         if (ret)
 248                 return ret;
 249
 250         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 251         if (ret)
 252                 return ret;
 253
 254         for (i = 0; i < meta->nb_args; i++) {
 255                 ret = trace_define_field(call, meta->types[i],
 256                                          meta->args[i], offset,
 257                                          sizeof(unsigned long), 0,
 258                                          FILTER_OTHER);
 259                 offset += sizeof(unsigned long);
 260         }
 261
 262         return ret;
 263 }
 264
 265 int syscall_exit_define_fields(struct ftrace_event_call *call)
 266 {
 267         struct syscall_trace_exit trace;
 268         int ret;
 269
 270         ret = trace_define_common_fields(call);
 271         if (ret)
 272                 return ret;
 273
 274         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 275         if (ret)
 276                 return ret;
 277
 278         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 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(struct ftrace_event_call *call)
 352 {
 353         int ret = 0;
 354         int num;
 355         const char *name;
 356
 357         name = ((struct syscall_metadata *)call->data)->name;
 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(struct ftrace_event_call *call)
 376 {
 377         int num;
 378         const char *name;
 379
 380         name = ((struct syscall_metadata *)call->data)->name;
 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(struct ftrace_event_call *call)
 393 {
 394         int ret = 0;
 395         int num;
 396         const char *name;
 397
 398         name = ((struct syscall_metadata *)call->data)->name;
 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(struct ftrace_event_call *call)
 417 {
 418         int num;
 419         const char *name;
 420
 421         name = ((struct syscall_metadata *)call->data)->name;
 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 int __init init_ftrace_syscalls(void)
 434 {
 435         struct syscall_metadata *meta;
 436         unsigned long addr;
 437         int i;
 438
 439         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 440                                         NR_syscalls, GFP_KERNEL);
 441         if (!syscalls_metadata) {
 442                 WARN_ON(1);
 443                 return -ENOMEM;
 444         }
 445
 446         for (i = 0; i < NR_syscalls; i++) {
 447                 addr = arch_syscall_addr(i);
 448                 meta = find_syscall_meta(addr);
 449                 syscalls_metadata[i] = meta;
 450         }
 451
 452         return 0;
 453 }
 454 core_initcall(init_ftrace_syscalls);
 455
 456 #ifdef CONFIG_EVENT_PROFILE
 457
 458 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 459 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 460 static int sys_prof_refcount_enter;
 461 static int sys_prof_refcount_exit;
 462
 463 static void prof_syscall_enter(struct pt_regs *regs, long id)
 464 {
 465         struct syscall_metadata *sys_data;
 466         struct syscall_trace_enter *rec;
 467         unsigned long flags;
 468         char *trace_buf;
 469         char *raw_data;
 470         int syscall_nr;
 471         int rctx;
 472         int size;
 473         int cpu;
 474
 475         syscall_nr = syscall_get_nr(current, regs);
 476         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 477                 return;
 478
 479         sys_data = syscall_nr_to_meta(syscall_nr);
 480         if (!sys_data)
 481                 return;
 482
 483         /* get the size after alignment with the u32 buffer size field */
 484         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 485         size = ALIGN(size + sizeof(u32), sizeof(u64));
 486         size -= sizeof(u32);
 487
 488         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 489                       "profile buffer not large enough"))
 490                 return;
 491
 492         /* Protect the per cpu buffer, begin the rcu read side */
 493         local_irq_save(flags);
 494
 495         rctx = perf_swevent_get_recursion_context();
 496         if (rctx < 0)
 497                 goto end_recursion;
 498
 499         cpu = smp_processor_id();
 500
 501         trace_buf = rcu_dereference(perf_trace_buf);
 502
 503         if (!trace_buf)
 504                 goto end;
 505
 506         raw_data = per_cpu_ptr(trace_buf, cpu);
 507
 508         /* zero the dead bytes from align to not leak stack to user */
 509         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 510
 511         rec = (struct syscall_trace_enter *) raw_data;
 512         tracing_generic_entry_update(&rec->ent, 0, 0);
 513         rec->ent.type = sys_data->enter_id;
 514         rec->nr = syscall_nr;
 515         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 516                                (unsigned long *)&rec->args);
 517         perf_tp_event(sys_data->enter_id, 0, 1, rec, size);
 518
 519 end:
 520         perf_swevent_put_recursion_context(rctx);
 521 end_recursion:
 522         local_irq_restore(flags);
 523 }
 524
 525 int reg_prof_syscall_enter(char *name)
 526 {
 527         int ret = 0;
 528         int num;
 529
 530         num = syscall_name_to_nr(name);
 531         if (num < 0 || num >= NR_syscalls)
 532                 return -ENOSYS;
 533
 534         mutex_lock(&syscall_trace_lock);
 535         if (!sys_prof_refcount_enter)
 536                 ret = register_trace_sys_enter(prof_syscall_enter);
 537         if (ret) {
 538                 pr_info("event trace: Could not activate"
 539                                 "syscall entry trace point");
 540         } else {
 541                 set_bit(num, enabled_prof_enter_syscalls);
 542                 sys_prof_refcount_enter++;
 543         }
 544         mutex_unlock(&syscall_trace_lock);
 545         return ret;
 546 }
 547
 548 void unreg_prof_syscall_enter(char *name)
 549 {
 550         int num;
 551
 552         num = syscall_name_to_nr(name);
 553         if (num < 0 || num >= NR_syscalls)
 554                 return;
 555
 556         mutex_lock(&syscall_trace_lock);
 557         sys_prof_refcount_enter--;
 558         clear_bit(num, enabled_prof_enter_syscalls);
 559         if (!sys_prof_refcount_enter)
 560                 unregister_trace_sys_enter(prof_syscall_enter);
 561         mutex_unlock(&syscall_trace_lock);
 562 }
 563
 564 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 565 {
 566         struct syscall_metadata *sys_data;
 567         struct syscall_trace_exit *rec;
 568         unsigned long flags;
 569         int syscall_nr;
 570         char *trace_buf;
 571         char *raw_data;
 572         int rctx;
 573         int size;
 574         int cpu;
 575
 576         syscall_nr = syscall_get_nr(current, regs);
 577         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 578                 return;
 579
 580         sys_data = syscall_nr_to_meta(syscall_nr);
 581         if (!sys_data)
 582                 return;
 583
 584         /* We can probably do that at build time */
 585         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 586         size -= sizeof(u32);
 587
 588         /*
 589          * Impossible, but be paranoid with the future
 590          * How to put this check outside runtime?
 591          */
 592         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 593                 "exit event has grown above profile buffer size"))
 594                 return;
 595
 596         /* Protect the per cpu buffer, begin the rcu read side */
 597         local_irq_save(flags);
 598
 599         rctx = perf_swevent_get_recursion_context();
 600         if (rctx < 0)
 601                 goto end_recursion;
 602
 603         cpu = smp_processor_id();
 604
 605         trace_buf = rcu_dereference(perf_trace_buf);
 606
 607         if (!trace_buf)
 608                 goto end;
 609
 610         raw_data = per_cpu_ptr(trace_buf, cpu);
 611
 612         /* zero the dead bytes from align to not leak stack to user */
 613         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 614
 615         rec = (struct syscall_trace_exit *)raw_data;
 616
 617         tracing_generic_entry_update(&rec->ent, 0, 0);
 618         rec->ent.type = sys_data->exit_id;
 619         rec->nr = syscall_nr;
 620         rec->ret = syscall_get_return_value(current, regs);
 621
 622         perf_tp_event(sys_data->exit_id, 0, 1, rec, size);
 623
 624 end:
 625         perf_swevent_put_recursion_context(rctx);
 626 end_recursion:
 627         local_irq_restore(flags);
 628 }
 629
 630 int reg_prof_syscall_exit(char *name)
 631 {
 632         int ret = 0;
 633         int num;
 634
 635         num = syscall_name_to_nr(name);
 636         if (num < 0 || num >= NR_syscalls)
 637                 return -ENOSYS;
 638
 639         mutex_lock(&syscall_trace_lock);
 640         if (!sys_prof_refcount_exit)
 641                 ret = register_trace_sys_exit(prof_syscall_exit);
 642         if (ret) {
 643                 pr_info("event trace: Could not activate"
 644                                 "syscall entry trace point");
 645         } else {
 646                 set_bit(num, enabled_prof_exit_syscalls);
 647                 sys_prof_refcount_exit++;
 648         }
 649         mutex_unlock(&syscall_trace_lock);
 650         return ret;
 651 }
 652
 653 void unreg_prof_syscall_exit(char *name)
 654 {
 655         int num;
 656
 657         num = syscall_name_to_nr(name);
 658         if (num < 0 || num >= NR_syscalls)
 659                 return;
 660
 661         mutex_lock(&syscall_trace_lock);
 662         sys_prof_refcount_exit--;
 663         clear_bit(num, enabled_prof_exit_syscalls);
 664         if (!sys_prof_refcount_exit)
 665                 unregister_trace_sys_exit(prof_syscall_exit);
 666         mutex_unlock(&syscall_trace_lock);
 667 }
 668
 669 #endif
 670
 671