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),            \
 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 nr;
 176         int ret;
 177         struct syscall_metadata *entry;
 178         struct syscall_trace_enter trace;
 179         int offset = offsetof(struct syscall_trace_enter, args);
 180
 181         nr = syscall_name_to_nr(call->data);
 182         entry = syscall_nr_to_meta(nr);
 183
 184         if (!entry)
 185                 return 0;
 186
 187         ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 188                                "\tsigned:%u;\n",
 189                                SYSCALL_FIELD(int, nr));
 190         if (!ret)
 191                 return 0;
 192
 193         for (i = 0; i < entry->nb_args; i++) {
 194                 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
 195                                         entry->args[i]);
 196                 if (!ret)
 197                         return 0;
 198                 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;"
 199                                        "\tsigned:%u;\n", offset,
 200                                        sizeof(unsigned long),
 201                                        is_signed_type(unsigned long));
 202                 if (!ret)
 203                         return 0;
 204                 offset += sizeof(unsigned long);
 205         }
 206
 207         trace_seq_puts(s, "\nprint fmt: \"");
 208         for (i = 0; i < entry->nb_args; i++) {
 209                 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
 210                                         sizeof(unsigned long),
 211                                         i == entry->nb_args - 1 ? "" : ", ");
 212                 if (!ret)
 213                         return 0;
 214         }
 215         trace_seq_putc(s, '"');
 216
 217         for (i = 0; i < entry->nb_args; i++) {
 218                 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
 219                                        entry->args[i]);
 220                 if (!ret)
 221                         return 0;
 222         }
 223
 224         return trace_seq_putc(s, '\n');
 225 }
 226
 227 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
 228 {
 229         int ret;
 230         struct syscall_trace_exit trace;
 231
 232         ret = trace_seq_printf(s,
 233                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 234                                "\tsigned:%u;\n"
 235                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 236                                "\tsigned:%u;\n",
 237                                SYSCALL_FIELD(int, nr),
 238                                SYSCALL_FIELD(long, ret));
 239         if (!ret)
 240                 return 0;
 241
 242         return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
 243 }
 244
 245 int syscall_enter_define_fields(struct ftrace_event_call *call)
 246 {
 247         struct syscall_trace_enter trace;
 248         struct syscall_metadata *meta;
 249         int ret;
 250         int nr;
 251         int i;
 252         int offset = offsetof(typeof(trace), args);
 253
 254         nr = syscall_name_to_nr(call->data);
 255         meta = syscall_nr_to_meta(nr);
 256
 257         if (!meta)
 258                 return 0;
 259
 260         ret = trace_define_common_fields(call);
 261         if (ret)
 262                 return ret;
 263
 264         for (i = 0; i < meta->nb_args; i++) {
 265                 ret = trace_define_field(call, meta->types[i],
 266                                          meta->args[i], offset,
 267                                          sizeof(unsigned long), 0,
 268                                          FILTER_OTHER);
 269                 offset += sizeof(unsigned long);
 270         }
 271
 272         return ret;
 273 }
 274
 275 int syscall_exit_define_fields(struct ftrace_event_call *call)
 276 {
 277         struct syscall_trace_exit trace;
 278         int ret;
 279
 280         ret = trace_define_common_fields(call);
 281         if (ret)
 282                 return ret;
 283
 284         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 285                                  FILTER_OTHER);
 286
 287         return ret;
 288 }
 289
 290 void ftrace_syscall_enter(struct pt_regs *regs, long id)
 291 {
 292         struct syscall_trace_enter *entry;
 293         struct syscall_metadata *sys_data;
 294         struct ring_buffer_event *event;
 295         struct ring_buffer *buffer;
 296         int size;
 297         int syscall_nr;
 298
 299         syscall_nr = syscall_get_nr(current, regs);
 300         if (syscall_nr < 0)
 301                 return;
 302         if (!test_bit(syscall_nr, enabled_enter_syscalls))
 303                 return;
 304
 305         sys_data = syscall_nr_to_meta(syscall_nr);
 306         if (!sys_data)
 307                 return;
 308
 309         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 310
 311         event = trace_current_buffer_lock_reserve(&buffer, sys_data->enter_id,
 312                                                   size, 0, 0);
 313         if (!event)
 314                 return;
 315
 316         entry = ring_buffer_event_data(event);
 317         entry->nr = syscall_nr;
 318         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 319
 320         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 321                                           entry, event))
 322                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 323 }
 324
 325 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
 326 {
 327         struct syscall_trace_exit *entry;
 328         struct syscall_metadata *sys_data;
 329         struct ring_buffer_event *event;
 330         struct ring_buffer *buffer;
 331         int syscall_nr;
 332
 333         syscall_nr = syscall_get_nr(current, regs);
 334         if (syscall_nr < 0)
 335                 return;
 336         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 337                 return;
 338
 339         sys_data = syscall_nr_to_meta(syscall_nr);
 340         if (!sys_data)
 341                 return;
 342
 343         event = trace_current_buffer_lock_reserve(&buffer, sys_data->exit_id,
 344                                 sizeof(*entry), 0, 0);
 345         if (!event)
 346                 return;
 347
 348         entry = ring_buffer_event_data(event);
 349         entry->nr = syscall_nr;
 350         entry->ret = syscall_get_return_value(current, regs);
 351
 352         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 353                                           entry, event))
 354                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 355 }
 356
 357 int reg_event_syscall_enter(struct ftrace_event_call *call)
 358 {
 359         int ret = 0;
 360         int num;
 361         char *name;
 362
 363         name = (char *)call->data;
 364         num = syscall_name_to_nr(name);
 365         if (num < 0 || num >= NR_syscalls)
 366                 return -ENOSYS;
 367         mutex_lock(&syscall_trace_lock);
 368         if (!sys_refcount_enter)
 369                 ret = register_trace_sys_enter(ftrace_syscall_enter);
 370         if (ret) {
 371                 pr_info("event trace: Could not activate"
 372                                 "syscall entry trace point");
 373         } else {
 374                 set_bit(num, enabled_enter_syscalls);
 375                 sys_refcount_enter++;
 376         }
 377         mutex_unlock(&syscall_trace_lock);
 378         return ret;
 379 }
 380
 381 void unreg_event_syscall_enter(struct ftrace_event_call *call)
 382 {
 383         int num;
 384         char *name;
 385
 386         name = (char *)call->data;
 387         num = syscall_name_to_nr(name);
 388         if (num < 0 || num >= NR_syscalls)
 389                 return;
 390         mutex_lock(&syscall_trace_lock);
 391         sys_refcount_enter--;
 392         clear_bit(num, enabled_enter_syscalls);
 393         if (!sys_refcount_enter)
 394                 unregister_trace_sys_enter(ftrace_syscall_enter);
 395         mutex_unlock(&syscall_trace_lock);
 396 }
 397
 398 int reg_event_syscall_exit(struct ftrace_event_call *call)
 399 {
 400         int ret = 0;
 401         int num;
 402         char *name;
 403
 404         name = call->data;
 405         num = syscall_name_to_nr(name);
 406         if (num < 0 || num >= NR_syscalls)
 407                 return -ENOSYS;
 408         mutex_lock(&syscall_trace_lock);
 409         if (!sys_refcount_exit)
 410                 ret = register_trace_sys_exit(ftrace_syscall_exit);
 411         if (ret) {
 412                 pr_info("event trace: Could not activate"
 413                                 "syscall exit trace point");
 414         } else {
 415                 set_bit(num, enabled_exit_syscalls);
 416                 sys_refcount_exit++;
 417         }
 418         mutex_unlock(&syscall_trace_lock);
 419         return ret;
 420 }
 421
 422 void unreg_event_syscall_exit(struct ftrace_event_call *call)
 423 {
 424         int num;
 425         char *name;
 426
 427         name = call->data;
 428         num = syscall_name_to_nr(name);
 429         if (num < 0 || num >= NR_syscalls)
 430                 return;
 431         mutex_lock(&syscall_trace_lock);
 432         sys_refcount_exit--;
 433         clear_bit(num, enabled_exit_syscalls);
 434         if (!sys_refcount_exit)
 435                 unregister_trace_sys_exit(ftrace_syscall_exit);
 436         mutex_unlock(&syscall_trace_lock);
 437 }
 438
 439 struct trace_event event_syscall_enter = {
 440         .trace                  = print_syscall_enter,
 441 };
 442
 443 struct trace_event event_syscall_exit = {
 444         .trace                  = print_syscall_exit,
 445 };
 446
 447 int __init init_ftrace_syscalls(void)
 448 {
 449         struct syscall_metadata *meta;
 450         unsigned long addr;
 451         int i;
 452
 453         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 454                                         NR_syscalls, GFP_KERNEL);
 455         if (!syscalls_metadata) {
 456                 WARN_ON(1);
 457                 return -ENOMEM;
 458         }
 459
 460         for (i = 0; i < NR_syscalls; i++) {
 461                 addr = arch_syscall_addr(i);
 462                 meta = find_syscall_meta(addr);
 463                 syscalls_metadata[i] = meta;
 464         }
 465
 466         return 0;
 467 }
 468 core_initcall(init_ftrace_syscalls);
 469
 470 #ifdef CONFIG_EVENT_PROFILE
 471
 472 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 473 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 474 static int sys_prof_refcount_enter;
 475 static int sys_prof_refcount_exit;
 476
 477 static void prof_syscall_enter(struct pt_regs *regs, long id)
 478 {
 479         struct syscall_metadata *sys_data;
 480         struct syscall_trace_enter *rec;
 481         unsigned long flags;
 482         char *raw_data;
 483         int syscall_nr;
 484         int size;
 485         int cpu;
 486
 487         syscall_nr = syscall_get_nr(current, regs);
 488         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 489                 return;
 490
 491         sys_data = syscall_nr_to_meta(syscall_nr);
 492         if (!sys_data)
 493                 return;
 494
 495         /* get the size after alignment with the u32 buffer size field */
 496         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 497         size = ALIGN(size + sizeof(u32), sizeof(u64));
 498         size -= sizeof(u32);
 499
 500         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 501                       "profile buffer not large enough"))
 502                 return;
 503
 504         /* Protect the per cpu buffer, begin the rcu read side */
 505         local_irq_save(flags);
 506
 507         cpu = smp_processor_id();
 508
 509         if (in_nmi())
 510                 raw_data = rcu_dereference(trace_profile_buf_nmi);
 511         else
 512                 raw_data = rcu_dereference(trace_profile_buf);
 513
 514         if (!raw_data)
 515                 goto end;
 516
 517         raw_data = per_cpu_ptr(raw_data, cpu);
 518
 519         /* zero the dead bytes from align to not leak stack to user */
 520         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 521
 522         rec = (struct syscall_trace_enter *) raw_data;
 523         tracing_generic_entry_update(&rec->ent, 0, 0);
 524         rec->ent.type = sys_data->enter_id;
 525         rec->nr = syscall_nr;
 526         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 527                                (unsigned long *)&rec->args);
 528         perf_tp_event(sys_data->enter_id, 0, 1, rec, size);
 529
 530 end:
 531         local_irq_restore(flags);
 532 }
 533
 534 int reg_prof_syscall_enter(char *name)
 535 {
 536         int ret = 0;
 537         int num;
 538
 539         num = syscall_name_to_nr(name);
 540         if (num < 0 || num >= NR_syscalls)
 541                 return -ENOSYS;
 542
 543         mutex_lock(&syscall_trace_lock);
 544         if (!sys_prof_refcount_enter)
 545                 ret = register_trace_sys_enter(prof_syscall_enter);
 546         if (ret) {
 547                 pr_info("event trace: Could not activate"
 548                                 "syscall entry trace point");
 549         } else {
 550                 set_bit(num, enabled_prof_enter_syscalls);
 551                 sys_prof_refcount_enter++;
 552         }
 553         mutex_unlock(&syscall_trace_lock);
 554         return ret;
 555 }
 556
 557 void unreg_prof_syscall_enter(char *name)
 558 {
 559         int num;
 560
 561         num = syscall_name_to_nr(name);
 562         if (num < 0 || num >= NR_syscalls)
 563                 return;
 564
 565         mutex_lock(&syscall_trace_lock);
 566         sys_prof_refcount_enter--;
 567         clear_bit(num, enabled_prof_enter_syscalls);
 568         if (!sys_prof_refcount_enter)
 569                 unregister_trace_sys_enter(prof_syscall_enter);
 570         mutex_unlock(&syscall_trace_lock);
 571 }
 572
 573 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 574 {
 575         struct syscall_metadata *sys_data;
 576         struct syscall_trace_exit *rec;
 577         unsigned long flags;
 578         int syscall_nr;
 579         char *raw_data;
 580         int size;
 581         int cpu;
 582
 583         syscall_nr = syscall_get_nr(current, regs);
 584         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 585                 return;
 586
 587         sys_data = syscall_nr_to_meta(syscall_nr);
 588         if (!sys_data)
 589                 return;
 590
 591         /* We can probably do that at build time */
 592         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 593         size -= sizeof(u32);
 594
 595         /*
 596          * Impossible, but be paranoid with the future
 597          * How to put this check outside runtime?
 598          */
 599         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 600                 "exit event has grown above profile buffer size"))
 601                 return;
 602
 603         /* Protect the per cpu buffer, begin the rcu read side */
 604         local_irq_save(flags);
 605         cpu = smp_processor_id();
 606
 607         if (in_nmi())
 608                 raw_data = rcu_dereference(trace_profile_buf_nmi);
 609         else
 610                 raw_data = rcu_dereference(trace_profile_buf);
 611
 612         if (!raw_data)
 613                 goto end;
 614
 615         raw_data = per_cpu_ptr(raw_data, cpu);
 616
 617         /* zero the dead bytes from align to not leak stack to user */
 618         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 619
 620         rec = (struct syscall_trace_exit *)raw_data;
 621
 622         tracing_generic_entry_update(&rec->ent, 0, 0);
 623         rec->ent.type = sys_data->exit_id;
 624         rec->nr = syscall_nr;
 625         rec->ret = syscall_get_return_value(current, regs);
 626
 627         perf_tp_event(sys_data->exit_id, 0, 1, rec, size);
 628
 629 end:
 630         local_irq_restore(flags);
 631 }
 632
 633 int reg_prof_syscall_exit(char *name)
 634 {
 635         int ret = 0;
 636         int num;
 637
 638         num = syscall_name_to_nr(name);
 639         if (num < 0 || num >= NR_syscalls)
 640                 return -ENOSYS;
 641
 642         mutex_lock(&syscall_trace_lock);
 643         if (!sys_prof_refcount_exit)
 644                 ret = register_trace_sys_exit(prof_syscall_exit);
 645         if (ret) {
 646                 pr_info("event trace: Could not activate"
 647                                 "syscall entry trace point");
 648         } else {
 649                 set_bit(num, enabled_prof_exit_syscalls);
 650                 sys_prof_refcount_exit++;
 651         }
 652         mutex_unlock(&syscall_trace_lock);
 653         return ret;
 654 }
 655
 656 void unreg_prof_syscall_exit(char *name)
 657 {
 658         int num;
 659
 660         num = syscall_name_to_nr(name);
 661         if (num < 0 || num >= NR_syscalls)
 662                 return;
 663
 664         mutex_lock(&syscall_trace_lock);
 665         sys_prof_refcount_exit--;
 666         clear_bit(num, enabled_prof_exit_syscalls);
 667         if (!sys_prof_refcount_exit)
 668                 unregister_trace_sys_exit(prof_syscall_exit);
 669         mutex_unlock(&syscall_trace_lock);
 670 }
 671
 672 #endif
 673
 674