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 enum print_line_t
  71 print_syscall_enter(struct trace_iterator *iter, int flags)
  72 {
  73         struct trace_seq *s = &iter->seq;
  74         struct trace_entry *ent = iter->ent;
  75         struct syscall_trace_enter *trace;
  76         struct syscall_metadata *entry;
  77         int i, ret, syscall;
  78
  79         trace = (typeof(trace))ent;
  80         syscall = trace->nr;
  81         entry = syscall_nr_to_meta(syscall);
  82
  83         if (!entry)
  84                 goto end;
  85
  86         if (entry->enter_event->id != ent->type) {
  87                 WARN_ON_ONCE(1);
  88                 goto end;
  89         }
  90
  91         ret = trace_seq_printf(s, "%s(", entry->name);
  92         if (!ret)
  93                 return TRACE_TYPE_PARTIAL_LINE;
  94
  95         for (i = 0; i < entry->nb_args; i++) {
  96                 /* parameter types */
  97                 if (trace_flags & TRACE_ITER_VERBOSE) {
  98                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
  99                         if (!ret)
 100                                 return TRACE_TYPE_PARTIAL_LINE;
 101                 }
 102                 /* parameter values */
 103                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 104                                        trace->args[i],
 105                                        i == entry->nb_args - 1 ? "" : ", ");
 106                 if (!ret)
 107                         return TRACE_TYPE_PARTIAL_LINE;
 108         }
 109
 110         ret = trace_seq_putc(s, ')');
 111         if (!ret)
 112                 return TRACE_TYPE_PARTIAL_LINE;
 113
 114 end:
 115         ret =  trace_seq_putc(s, '\n');
 116         if (!ret)
 117                 return TRACE_TYPE_PARTIAL_LINE;
 118
 119         return TRACE_TYPE_HANDLED;
 120 }
 121
 122 enum print_line_t
 123 print_syscall_exit(struct trace_iterator *iter, int flags)
 124 {
 125         struct trace_seq *s = &iter->seq;
 126         struct trace_entry *ent = iter->ent;
 127         struct syscall_trace_exit *trace;
 128         int syscall;
 129         struct syscall_metadata *entry;
 130         int ret;
 131
 132         trace = (typeof(trace))ent;
 133         syscall = trace->nr;
 134         entry = syscall_nr_to_meta(syscall);
 135
 136         if (!entry) {
 137                 trace_seq_printf(s, "\n");
 138                 return TRACE_TYPE_HANDLED;
 139         }
 140
 141         if (entry->exit_event->id != ent->type) {
 142                 WARN_ON_ONCE(1);
 143                 return TRACE_TYPE_UNHANDLED;
 144         }
 145
 146         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 147                                 trace->ret);
 148         if (!ret)
 149                 return TRACE_TYPE_PARTIAL_LINE;
 150
 151         return TRACE_TYPE_HANDLED;
 152 }
 153
 154 extern char *__bad_type_size(void);
 155
 156 #define SYSCALL_FIELD(type, name)                                       \
 157         sizeof(type) != sizeof(trace.name) ?                            \
 158                 __bad_type_size() :                                     \
 159                 #type, #name, offsetof(typeof(trace), name),            \
 160                 sizeof(trace.name), is_signed_type(type)
 161
 162 int syscall_enter_format(struct ftrace_event_call *call, struct trace_seq *s)
 163 {
 164         int i;
 165         int ret;
 166         struct syscall_metadata *entry = call->data;
 167         struct syscall_trace_enter trace;
 168         int offset = offsetof(struct syscall_trace_enter, args);
 169
 170         ret = trace_seq_printf(s, "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 171                                "\tsigned:%u;\n",
 172                                SYSCALL_FIELD(int, nr));
 173         if (!ret)
 174                 return 0;
 175
 176         for (i = 0; i < entry->nb_args; i++) {
 177                 ret = trace_seq_printf(s, "\tfield:%s %s;", entry->types[i],
 178                                         entry->args[i]);
 179                 if (!ret)
 180                         return 0;
 181                 ret = trace_seq_printf(s, "\toffset:%d;\tsize:%zu;"
 182                                        "\tsigned:%u;\n", offset,
 183                                        sizeof(unsigned long),
 184                                        is_signed_type(unsigned long));
 185                 if (!ret)
 186                         return 0;
 187                 offset += sizeof(unsigned long);
 188         }
 189
 190         trace_seq_puts(s, "\nprint fmt: \"");
 191         for (i = 0; i < entry->nb_args; i++) {
 192                 ret = trace_seq_printf(s, "%s: 0x%%0%zulx%s", entry->args[i],
 193                                         sizeof(unsigned long),
 194                                         i == entry->nb_args - 1 ? "" : ", ");
 195                 if (!ret)
 196                         return 0;
 197         }
 198         trace_seq_putc(s, '"');
 199
 200         for (i = 0; i < entry->nb_args; i++) {
 201                 ret = trace_seq_printf(s, ", ((unsigned long)(REC->%s))",
 202                                        entry->args[i]);
 203                 if (!ret)
 204                         return 0;
 205         }
 206
 207         return trace_seq_putc(s, '\n');
 208 }
 209
 210 int syscall_exit_format(struct ftrace_event_call *call, struct trace_seq *s)
 211 {
 212         int ret;
 213         struct syscall_trace_exit trace;
 214
 215         ret = trace_seq_printf(s,
 216                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 217                                "\tsigned:%u;\n"
 218                                "\tfield:%s %s;\toffset:%zu;\tsize:%zu;"
 219                                "\tsigned:%u;\n",
 220                                SYSCALL_FIELD(int, nr),
 221                                SYSCALL_FIELD(long, ret));
 222         if (!ret)
 223                 return 0;
 224
 225         return trace_seq_printf(s, "\nprint fmt: \"0x%%lx\", REC->ret\n");
 226 }
 227
 228 int syscall_enter_define_fields(struct ftrace_event_call *call)
 229 {
 230         struct syscall_trace_enter trace;
 231         struct syscall_metadata *meta = call->data;
 232         int ret;
 233         int i;
 234         int offset = offsetof(typeof(trace), args);
 235
 236         ret = trace_define_common_fields(call);
 237         if (ret)
 238                 return ret;
 239
 240         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 241         if (ret)
 242                 return ret;
 243
 244         for (i = 0; i < meta->nb_args; i++) {
 245                 ret = trace_define_field(call, meta->types[i],
 246                                          meta->args[i], offset,
 247                                          sizeof(unsigned long), 0,
 248                                          FILTER_OTHER);
 249                 offset += sizeof(unsigned long);
 250         }
 251
 252         return ret;
 253 }
 254
 255 int syscall_exit_define_fields(struct ftrace_event_call *call)
 256 {
 257         struct syscall_trace_exit trace;
 258         int ret;
 259
 260         ret = trace_define_common_fields(call);
 261         if (ret)
 262                 return ret;
 263
 264         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 265         if (ret)
 266                 return ret;
 267
 268         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 269                                  FILTER_OTHER);
 270
 271         return ret;
 272 }
 273
 274 void ftrace_syscall_enter(struct pt_regs *regs, long id)
 275 {
 276         struct syscall_trace_enter *entry;
 277         struct syscall_metadata *sys_data;
 278         struct ring_buffer_event *event;
 279         struct ring_buffer *buffer;
 280         int size;
 281         int syscall_nr;
 282
 283         syscall_nr = syscall_get_nr(current, regs);
 284         if (syscall_nr < 0)
 285                 return;
 286         if (!test_bit(syscall_nr, enabled_enter_syscalls))
 287                 return;
 288
 289         sys_data = syscall_nr_to_meta(syscall_nr);
 290         if (!sys_data)
 291                 return;
 292
 293         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 294
 295         event = trace_current_buffer_lock_reserve(&buffer,
 296                         sys_data->enter_event->id, size, 0, 0);
 297         if (!event)
 298                 return;
 299
 300         entry = ring_buffer_event_data(event);
 301         entry->nr = syscall_nr;
 302         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 303
 304         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 305                                           entry, event))
 306                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 307 }
 308
 309 void ftrace_syscall_exit(struct pt_regs *regs, long ret)
 310 {
 311         struct syscall_trace_exit *entry;
 312         struct syscall_metadata *sys_data;
 313         struct ring_buffer_event *event;
 314         struct ring_buffer *buffer;
 315         int syscall_nr;
 316
 317         syscall_nr = syscall_get_nr(current, regs);
 318         if (syscall_nr < 0)
 319                 return;
 320         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 321                 return;
 322
 323         sys_data = syscall_nr_to_meta(syscall_nr);
 324         if (!sys_data)
 325                 return;
 326
 327         event = trace_current_buffer_lock_reserve(&buffer,
 328                         sys_data->exit_event->id, sizeof(*entry), 0, 0);
 329         if (!event)
 330                 return;
 331
 332         entry = ring_buffer_event_data(event);
 333         entry->nr = syscall_nr;
 334         entry->ret = syscall_get_return_value(current, regs);
 335
 336         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 337                                           entry, event))
 338                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 339 }
 340
 341 int reg_event_syscall_enter(struct ftrace_event_call *call)
 342 {
 343         int ret = 0;
 344         int num;
 345         const char *name;
 346
 347         name = ((struct syscall_metadata *)call->data)->name;
 348         num = syscall_name_to_nr(name);
 349         if (num < 0 || num >= NR_syscalls)
 350                 return -ENOSYS;
 351         mutex_lock(&syscall_trace_lock);
 352         if (!sys_refcount_enter)
 353                 ret = register_trace_sys_enter(ftrace_syscall_enter);
 354         if (ret) {
 355                 pr_info("event trace: Could not activate"
 356                                 "syscall entry trace point");
 357         } else {
 358                 set_bit(num, enabled_enter_syscalls);
 359                 sys_refcount_enter++;
 360         }
 361         mutex_unlock(&syscall_trace_lock);
 362         return ret;
 363 }
 364
 365 void unreg_event_syscall_enter(struct ftrace_event_call *call)
 366 {
 367         int num;
 368         const char *name;
 369
 370         name = ((struct syscall_metadata *)call->data)->name;
 371         num = syscall_name_to_nr(name);
 372         if (num < 0 || num >= NR_syscalls)
 373                 return;
 374         mutex_lock(&syscall_trace_lock);
 375         sys_refcount_enter--;
 376         clear_bit(num, enabled_enter_syscalls);
 377         if (!sys_refcount_enter)
 378                 unregister_trace_sys_enter(ftrace_syscall_enter);
 379         mutex_unlock(&syscall_trace_lock);
 380 }
 381
 382 int reg_event_syscall_exit(struct ftrace_event_call *call)
 383 {
 384         int ret = 0;
 385         int num;
 386         const char *name;
 387
 388         name = ((struct syscall_metadata *)call->data)->name;
 389         num = syscall_name_to_nr(name);
 390         if (num < 0 || num >= NR_syscalls)
 391                 return -ENOSYS;
 392         mutex_lock(&syscall_trace_lock);
 393         if (!sys_refcount_exit)
 394                 ret = register_trace_sys_exit(ftrace_syscall_exit);
 395         if (ret) {
 396                 pr_info("event trace: Could not activate"
 397                                 "syscall exit trace point");
 398         } else {
 399                 set_bit(num, enabled_exit_syscalls);
 400                 sys_refcount_exit++;
 401         }
 402         mutex_unlock(&syscall_trace_lock);
 403         return ret;
 404 }
 405
 406 void unreg_event_syscall_exit(struct ftrace_event_call *call)
 407 {
 408         int num;
 409         const char *name;
 410
 411         name = ((struct syscall_metadata *)call->data)->name;
 412         num = syscall_name_to_nr(name);
 413         if (num < 0 || num >= NR_syscalls)
 414                 return;
 415         mutex_lock(&syscall_trace_lock);
 416         sys_refcount_exit--;
 417         clear_bit(num, enabled_exit_syscalls);
 418         if (!sys_refcount_exit)
 419                 unregister_trace_sys_exit(ftrace_syscall_exit);
 420         mutex_unlock(&syscall_trace_lock);
 421 }
 422
 423 int __init init_ftrace_syscalls(void)
 424 {
 425         struct syscall_metadata *meta;
 426         unsigned long addr;
 427         int i;
 428
 429         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 430                                         NR_syscalls, GFP_KERNEL);
 431         if (!syscalls_metadata) {
 432                 WARN_ON(1);
 433                 return -ENOMEM;
 434         }
 435
 436         for (i = 0; i < NR_syscalls; i++) {
 437                 addr = arch_syscall_addr(i);
 438                 meta = find_syscall_meta(addr);
 439                 syscalls_metadata[i] = meta;
 440         }
 441
 442         return 0;
 443 }
 444 core_initcall(init_ftrace_syscalls);
 445
 446 #ifdef CONFIG_EVENT_PROFILE
 447
 448 static DECLARE_BITMAP(enabled_prof_enter_syscalls, NR_syscalls);
 449 static DECLARE_BITMAP(enabled_prof_exit_syscalls, NR_syscalls);
 450 static int sys_prof_refcount_enter;
 451 static int sys_prof_refcount_exit;
 452
 453 static void prof_syscall_enter(struct pt_regs *regs, long id)
 454 {
 455         struct syscall_metadata *sys_data;
 456         struct syscall_trace_enter *rec;
 457         unsigned long flags;
 458         char *trace_buf;
 459         char *raw_data;
 460         int syscall_nr;
 461         int rctx;
 462         int size;
 463         int cpu;
 464
 465         syscall_nr = syscall_get_nr(current, regs);
 466         if (!test_bit(syscall_nr, enabled_prof_enter_syscalls))
 467                 return;
 468
 469         sys_data = syscall_nr_to_meta(syscall_nr);
 470         if (!sys_data)
 471                 return;
 472
 473         /* get the size after alignment with the u32 buffer size field */
 474         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 475         size = ALIGN(size + sizeof(u32), sizeof(u64));
 476         size -= sizeof(u32);
 477
 478         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 479                       "profile buffer not large enough"))
 480                 return;
 481
 482         /* Protect the per cpu buffer, begin the rcu read side */
 483         local_irq_save(flags);
 484
 485         rctx = perf_swevent_get_recursion_context();
 486         if (rctx < 0)
 487                 goto end_recursion;
 488
 489         cpu = smp_processor_id();
 490
 491         trace_buf = rcu_dereference(perf_trace_buf);
 492
 493         if (!trace_buf)
 494                 goto end;
 495
 496         raw_data = per_cpu_ptr(trace_buf, cpu);
 497
 498         /* zero the dead bytes from align to not leak stack to user */
 499         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 500
 501         rec = (struct syscall_trace_enter *) raw_data;
 502         tracing_generic_entry_update(&rec->ent, 0, 0);
 503         rec->ent.type = sys_data->enter_event->id;
 504         rec->nr = syscall_nr;
 505         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 506                                (unsigned long *)&rec->args);
 507         perf_tp_event(sys_data->enter_event->id, 0, 1, rec, size);
 508
 509 end:
 510         perf_swevent_put_recursion_context(rctx);
 511 end_recursion:
 512         local_irq_restore(flags);
 513 }
 514
 515 int reg_prof_syscall_enter(char *name)
 516 {
 517         int ret = 0;
 518         int num;
 519
 520         num = syscall_name_to_nr(name);
 521         if (num < 0 || num >= NR_syscalls)
 522                 return -ENOSYS;
 523
 524         mutex_lock(&syscall_trace_lock);
 525         if (!sys_prof_refcount_enter)
 526                 ret = register_trace_sys_enter(prof_syscall_enter);
 527         if (ret) {
 528                 pr_info("event trace: Could not activate"
 529                                 "syscall entry trace point");
 530         } else {
 531                 set_bit(num, enabled_prof_enter_syscalls);
 532                 sys_prof_refcount_enter++;
 533         }
 534         mutex_unlock(&syscall_trace_lock);
 535         return ret;
 536 }
 537
 538 void unreg_prof_syscall_enter(char *name)
 539 {
 540         int num;
 541
 542         num = syscall_name_to_nr(name);
 543         if (num < 0 || num >= NR_syscalls)
 544                 return;
 545
 546         mutex_lock(&syscall_trace_lock);
 547         sys_prof_refcount_enter--;
 548         clear_bit(num, enabled_prof_enter_syscalls);
 549         if (!sys_prof_refcount_enter)
 550                 unregister_trace_sys_enter(prof_syscall_enter);
 551         mutex_unlock(&syscall_trace_lock);
 552 }
 553
 554 static void prof_syscall_exit(struct pt_regs *regs, long ret)
 555 {
 556         struct syscall_metadata *sys_data;
 557         struct syscall_trace_exit *rec;
 558         unsigned long flags;
 559         int syscall_nr;
 560         char *trace_buf;
 561         char *raw_data;
 562         int rctx;
 563         int size;
 564         int cpu;
 565
 566         syscall_nr = syscall_get_nr(current, regs);
 567         if (!test_bit(syscall_nr, enabled_prof_exit_syscalls))
 568                 return;
 569
 570         sys_data = syscall_nr_to_meta(syscall_nr);
 571         if (!sys_data)
 572                 return;
 573
 574         /* We can probably do that at build time */
 575         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 576         size -= sizeof(u32);
 577
 578         /*
 579          * Impossible, but be paranoid with the future
 580          * How to put this check outside runtime?
 581          */
 582         if (WARN_ONCE(size > FTRACE_MAX_PROFILE_SIZE,
 583                 "exit event has grown above profile buffer size"))
 584                 return;
 585
 586         /* Protect the per cpu buffer, begin the rcu read side */
 587         local_irq_save(flags);
 588
 589         rctx = perf_swevent_get_recursion_context();
 590         if (rctx < 0)
 591                 goto end_recursion;
 592
 593         cpu = smp_processor_id();
 594
 595         trace_buf = rcu_dereference(perf_trace_buf);
 596
 597         if (!trace_buf)
 598                 goto end;
 599
 600         raw_data = per_cpu_ptr(trace_buf, cpu);
 601
 602         /* zero the dead bytes from align to not leak stack to user */
 603         *(u64 *)(&raw_data[size - sizeof(u64)]) = 0ULL;
 604
 605         rec = (struct syscall_trace_exit *)raw_data;
 606
 607         tracing_generic_entry_update(&rec->ent, 0, 0);
 608         rec->ent.type = sys_data->exit_event->id;
 609         rec->nr = syscall_nr;
 610         rec->ret = syscall_get_return_value(current, regs);
 611
 612         perf_tp_event(sys_data->exit_event->id, 0, 1, rec, size);
 613
 614 end:
 615         perf_swevent_put_recursion_context(rctx);
 616 end_recursion:
 617         local_irq_restore(flags);
 618 }
 619
 620 int reg_prof_syscall_exit(char *name)
 621 {
 622         int ret = 0;
 623         int num;
 624
 625         num = syscall_name_to_nr(name);
 626         if (num < 0 || num >= NR_syscalls)
 627                 return -ENOSYS;
 628
 629         mutex_lock(&syscall_trace_lock);
 630         if (!sys_prof_refcount_exit)
 631                 ret = register_trace_sys_exit(prof_syscall_exit);
 632         if (ret) {
 633                 pr_info("event trace: Could not activate"
 634                                 "syscall entry trace point");
 635         } else {
 636                 set_bit(num, enabled_prof_exit_syscalls);
 637                 sys_prof_refcount_exit++;
 638         }
 639         mutex_unlock(&syscall_trace_lock);
 640         return ret;
 641 }
 642
 643 void unreg_prof_syscall_exit(char *name)
 644 {
 645         int num;
 646
 647         num = syscall_name_to_nr(name);
 648         if (num < 0 || num >= NR_syscalls)
 649                 return;
 650
 651         mutex_lock(&syscall_trace_lock);
 652         sys_prof_refcount_exit--;
 653         clear_bit(num, enabled_prof_exit_syscalls);
 654         if (!sys_prof_refcount_exit)
 655                 unregister_trace_sys_exit(prof_syscall_exit);
 656         mutex_unlock(&syscall_trace_lock);
 657 }
 658
 659 #endif
 660
 661