kernel/trace/trace_syscalls.c

   1 #include <trace/syscall.h>
   2 #include <trace/events/syscalls.h>
   3 #include <linux/slab.h>
   4 #include <linux/kernel.h>
   5 #include <linux/ftrace.h>
   6 #include <linux/perf_event.h>
   7 #include <asm/syscall.h>
   8
   9 #include "trace_output.h"
  10 #include "trace.h"
  11
  12 static DEFINE_MUTEX(syscall_trace_lock);
  13 static int sys_refcount_enter;
  14 static int sys_refcount_exit;
  15 static DECLARE_BITMAP(enabled_enter_syscalls, NR_syscalls);
  16 static DECLARE_BITMAP(enabled_exit_syscalls, NR_syscalls);
  17
  18 static int syscall_enter_register(struct ftrace_event_call *event,
  19                                  enum trace_reg type);
  20 static int syscall_exit_register(struct ftrace_event_call *event,
  21                                  enum trace_reg type);
  22
  23 static int syscall_enter_define_fields(struct ftrace_event_call *call);
  24 static int syscall_exit_define_fields(struct ftrace_event_call *call);
  25
  26 static struct list_head *
  27 syscall_get_enter_fields(struct ftrace_event_call *call)
  28 {
  29         struct syscall_metadata *entry = call->data;
  30
  31         return &entry->enter_fields;
  32 }
  33
  34 static struct list_head *
  35 syscall_get_exit_fields(struct ftrace_event_call *call)
  36 {
  37         struct syscall_metadata *entry = call->data;
  38
  39         return &entry->exit_fields;
  40 }
  41
  42 struct ftrace_event_class event_class_syscall_enter = {
  43         .system                 = "syscalls",
  44         .reg                    = syscall_enter_register,
  45         .define_fields          = syscall_enter_define_fields,
  46         .get_fields             = syscall_get_enter_fields,
  47 };
  48
  49 struct ftrace_event_class event_class_syscall_exit = {
  50         .system                 = "syscalls",
  51         .reg                    = syscall_exit_register,
  52         .define_fields          = syscall_exit_define_fields,
  53         .get_fields             = syscall_get_exit_fields,
  54 };
  55
  56 extern unsigned long __start_syscalls_metadata[];
  57 extern unsigned long __stop_syscalls_metadata[];
  58
  59 static struct syscall_metadata **syscalls_metadata;
  60
  61 static struct syscall_metadata *find_syscall_meta(unsigned long syscall)
  62 {
  63         struct syscall_metadata *start;
  64         struct syscall_metadata *stop;
  65         char str[KSYM_SYMBOL_LEN];
  66
  67
  68         start = (struct syscall_metadata *)__start_syscalls_metadata;
  69         stop = (struct syscall_metadata *)__stop_syscalls_metadata;
  70         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
  71
  72         for ( ; start < stop; start++) {
  73                 /*
  74                  * Only compare after the "sys" prefix. Archs that use
  75                  * syscall wrappers may have syscalls symbols aliases prefixed
  76                  * with "SyS" instead of "sys", leading to an unwanted
  77                  * mismatch.
  78                  */
  79                 if (start->name && !strcmp(start->name + 3, str + 3))
  80                         return start;
  81         }
  82         return NULL;
  83 }
  84
  85 static struct syscall_metadata *syscall_nr_to_meta(int nr)
  86 {
  87         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
  88                 return NULL;
  89
  90         return syscalls_metadata[nr];
  91 }
  92
  93 enum print_line_t
  94 print_syscall_enter(struct trace_iterator *iter, int flags)
  95 {
  96         struct trace_seq *s = &iter->seq;
  97         struct trace_entry *ent = iter->ent;
  98         struct syscall_trace_enter *trace;
  99         struct syscall_metadata *entry;
 100         int i, ret, syscall;
 101
 102         trace = (typeof(trace))ent;
 103         syscall = trace->nr;
 104         entry = syscall_nr_to_meta(syscall);
 105
 106         if (!entry)
 107                 goto end;
 108
 109         if (entry->enter_event->id != ent->type) {
 110                 WARN_ON_ONCE(1);
 111                 goto end;
 112         }
 113
 114         ret = trace_seq_printf(s, "%s(", entry->name);
 115         if (!ret)
 116                 return TRACE_TYPE_PARTIAL_LINE;
 117
 118         for (i = 0; i < entry->nb_args; i++) {
 119                 /* parameter types */
 120                 if (trace_flags & TRACE_ITER_VERBOSE) {
 121                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
 122                         if (!ret)
 123                                 return TRACE_TYPE_PARTIAL_LINE;
 124                 }
 125                 /* parameter values */
 126                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 127                                        trace->args[i],
 128                                        i == entry->nb_args - 1 ? "" : ", ");
 129                 if (!ret)
 130                         return TRACE_TYPE_PARTIAL_LINE;
 131         }
 132
 133         ret = trace_seq_putc(s, ')');
 134         if (!ret)
 135                 return TRACE_TYPE_PARTIAL_LINE;
 136
 137 end:
 138         ret =  trace_seq_putc(s, '\n');
 139         if (!ret)
 140                 return TRACE_TYPE_PARTIAL_LINE;
 141
 142         return TRACE_TYPE_HANDLED;
 143 }
 144
 145 enum print_line_t
 146 print_syscall_exit(struct trace_iterator *iter, int flags)
 147 {
 148         struct trace_seq *s = &iter->seq;
 149         struct trace_entry *ent = iter->ent;
 150         struct syscall_trace_exit *trace;
 151         int syscall;
 152         struct syscall_metadata *entry;
 153         int ret;
 154
 155         trace = (typeof(trace))ent;
 156         syscall = trace->nr;
 157         entry = syscall_nr_to_meta(syscall);
 158
 159         if (!entry) {
 160                 trace_seq_printf(s, "\n");
 161                 return TRACE_TYPE_HANDLED;
 162         }
 163
 164         if (entry->exit_event->id != ent->type) {
 165                 WARN_ON_ONCE(1);
 166                 return TRACE_TYPE_UNHANDLED;
 167         }
 168
 169         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 170                                 trace->ret);
 171         if (!ret)
 172                 return TRACE_TYPE_PARTIAL_LINE;
 173
 174         return TRACE_TYPE_HANDLED;
 175 }
 176
 177 extern char *__bad_type_size(void);
 178
 179 #define SYSCALL_FIELD(type, name)                                       \
 180         sizeof(type) != sizeof(trace.name) ?                            \
 181                 __bad_type_size() :                                     \
 182                 #type, #name, offsetof(typeof(trace), name),            \
 183                 sizeof(trace.name), is_signed_type(type)
 184
 185 static
 186 int  __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 187 {
 188         int i;
 189         int pos = 0;
 190
 191         /* When len=0, we just calculate the needed length */
 192 #define LEN_OR_ZERO (len ? len - pos : 0)
 193
 194         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 195         for (i = 0; i < entry->nb_args; i++) {
 196                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 197                                 entry->args[i], sizeof(unsigned long),
 198                                 i == entry->nb_args - 1 ? "" : ", ");
 199         }
 200         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 201
 202         for (i = 0; i < entry->nb_args; i++) {
 203                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 204                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 205         }
 206
 207 #undef LEN_OR_ZERO
 208
 209         /* return the length of print_fmt */
 210         return pos;
 211 }
 212
 213 static int set_syscall_print_fmt(struct ftrace_event_call *call)
 214 {
 215         char *print_fmt;
 216         int len;
 217         struct syscall_metadata *entry = call->data;
 218
 219         if (entry->enter_event != call) {
 220                 call->print_fmt = "\"0x%lx\", REC->ret";
 221                 return 0;
 222         }
 223
 224         /* First: called with 0 length to calculate the needed length */
 225         len = __set_enter_print_fmt(entry, NULL, 0);
 226
 227         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 228         if (!print_fmt)
 229                 return -ENOMEM;
 230
 231         /* Second: actually write the @print_fmt */
 232         __set_enter_print_fmt(entry, print_fmt, len + 1);
 233         call->print_fmt = print_fmt;
 234
 235         return 0;
 236 }
 237
 238 static void free_syscall_print_fmt(struct ftrace_event_call *call)
 239 {
 240         struct syscall_metadata *entry = call->data;
 241
 242         if (entry->enter_event == call)
 243                 kfree(call->print_fmt);
 244 }
 245
 246 static int syscall_enter_define_fields(struct ftrace_event_call *call)
 247 {
 248         struct syscall_trace_enter trace;
 249         struct syscall_metadata *meta = call->data;
 250         int ret;
 251         int i;
 252         int offset = offsetof(typeof(trace), args);
 253
 254         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 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 static 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_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(void *ignore, 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,
 306                         sys_data->enter_event->id, 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(void *ignore, 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,
 338                         sys_data->exit_event->id, 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
 356         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 357         if (num < 0 || num >= NR_syscalls)
 358                 return -ENOSYS;
 359         mutex_lock(&syscall_trace_lock);
 360         if (!sys_refcount_enter)
 361                 ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
 362         if (!ret) {
 363                 set_bit(num, enabled_enter_syscalls);
 364                 sys_refcount_enter++;
 365         }
 366         mutex_unlock(&syscall_trace_lock);
 367         return ret;
 368 }
 369
 370 void unreg_event_syscall_enter(struct ftrace_event_call *call)
 371 {
 372         int num;
 373
 374         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 375         if (num < 0 || num >= NR_syscalls)
 376                 return;
 377         mutex_lock(&syscall_trace_lock);
 378         sys_refcount_enter--;
 379         clear_bit(num, enabled_enter_syscalls);
 380         if (!sys_refcount_enter)
 381                 unregister_trace_sys_enter(ftrace_syscall_enter, NULL);
 382         mutex_unlock(&syscall_trace_lock);
 383 }
 384
 385 int reg_event_syscall_exit(struct ftrace_event_call *call)
 386 {
 387         int ret = 0;
 388         int num;
 389
 390         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 391         if (num < 0 || num >= NR_syscalls)
 392                 return -ENOSYS;
 393         mutex_lock(&syscall_trace_lock);
 394         if (!sys_refcount_exit)
 395                 ret = register_trace_sys_exit(ftrace_syscall_exit, NULL);
 396         if (!ret) {
 397                 set_bit(num, enabled_exit_syscalls);
 398                 sys_refcount_exit++;
 399         }
 400         mutex_unlock(&syscall_trace_lock);
 401         return ret;
 402 }
 403
 404 void unreg_event_syscall_exit(struct ftrace_event_call *call)
 405 {
 406         int num;
 407
 408         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 409         if (num < 0 || num >= NR_syscalls)
 410                 return;
 411         mutex_lock(&syscall_trace_lock);
 412         sys_refcount_exit--;
 413         clear_bit(num, enabled_exit_syscalls);
 414         if (!sys_refcount_exit)
 415                 unregister_trace_sys_exit(ftrace_syscall_exit, NULL);
 416         mutex_unlock(&syscall_trace_lock);
 417 }
 418
 419 int init_syscall_trace(struct ftrace_event_call *call)
 420 {
 421         int id;
 422
 423         if (set_syscall_print_fmt(call) < 0)
 424                 return -ENOMEM;
 425
 426         id = trace_event_raw_init(call);
 427
 428         if (id < 0) {
 429                 free_syscall_print_fmt(call);
 430                 return id;
 431         }
 432
 433         return id;
 434 }
 435
 436 unsigned long __init arch_syscall_addr(int nr)
 437 {
 438         return (unsigned long)sys_call_table[nr];
 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                 if (!meta)
 458                         continue;
 459
 460                 meta->syscall_nr = i;
 461                 syscalls_metadata[i] = meta;
 462         }
 463
 464         return 0;
 465 }
 466 core_initcall(init_ftrace_syscalls);
 467
 468 #ifdef CONFIG_PERF_EVENTS
 469
 470 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 471 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 472 static int sys_perf_refcount_enter;
 473 static int sys_perf_refcount_exit;
 474
 475 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 476 {
 477         struct syscall_metadata *sys_data;
 478         struct syscall_trace_enter *rec;
 479         unsigned long flags;
 480         int syscall_nr;
 481         int rctx;
 482         int size;
 483
 484         syscall_nr = syscall_get_nr(current, regs);
 485         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 486                 return;
 487
 488         sys_data = syscall_nr_to_meta(syscall_nr);
 489         if (!sys_data)
 490                 return;
 491
 492         /* get the size after alignment with the u32 buffer size field */
 493         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 494         size = ALIGN(size + sizeof(u32), sizeof(u64));
 495         size -= sizeof(u32);
 496
 497         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 498                       "perf buffer not large enough"))
 499                 return;
 500
 501         rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
 502                                 sys_data->enter_event->id, &rctx, &flags);
 503         if (!rec)
 504                 return;
 505
 506         rec->nr = syscall_nr;
 507         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 508                                (unsigned long *)&rec->args);
 509         perf_trace_buf_submit(rec, size, rctx, 0, 1, flags, regs);
 510 }
 511
 512 int perf_sysenter_enable(struct ftrace_event_call *call)
 513 {
 514         int ret = 0;
 515         int num;
 516
 517         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 518
 519         mutex_lock(&syscall_trace_lock);
 520         if (!sys_perf_refcount_enter)
 521                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 522         if (ret) {
 523                 pr_info("event trace: Could not activate"
 524                                 "syscall entry trace point");
 525         } else {
 526                 set_bit(num, enabled_perf_enter_syscalls);
 527                 sys_perf_refcount_enter++;
 528         }
 529         mutex_unlock(&syscall_trace_lock);
 530         return ret;
 531 }
 532
 533 void perf_sysenter_disable(struct ftrace_event_call *call)
 534 {
 535         int num;
 536
 537         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 538
 539         mutex_lock(&syscall_trace_lock);
 540         sys_perf_refcount_enter--;
 541         clear_bit(num, enabled_perf_enter_syscalls);
 542         if (!sys_perf_refcount_enter)
 543                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 544         mutex_unlock(&syscall_trace_lock);
 545 }
 546
 547 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 548 {
 549         struct syscall_metadata *sys_data;
 550         struct syscall_trace_exit *rec;
 551         unsigned long flags;
 552         int syscall_nr;
 553         int rctx;
 554         int size;
 555
 556         syscall_nr = syscall_get_nr(current, regs);
 557         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 558                 return;
 559
 560         sys_data = syscall_nr_to_meta(syscall_nr);
 561         if (!sys_data)
 562                 return;
 563
 564         /* We can probably do that at build time */
 565         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 566         size -= sizeof(u32);
 567
 568         /*
 569          * Impossible, but be paranoid with the future
 570          * How to put this check outside runtime?
 571          */
 572         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 573                 "exit event has grown above perf buffer size"))
 574                 return;
 575
 576         rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
 577                                 sys_data->exit_event->id, &rctx, &flags);
 578         if (!rec)
 579                 return;
 580
 581         rec->nr = syscall_nr;
 582         rec->ret = syscall_get_return_value(current, regs);
 583
 584         perf_trace_buf_submit(rec, size, rctx, 0, 1, flags, regs);
 585 }
 586
 587 int perf_sysexit_enable(struct ftrace_event_call *call)
 588 {
 589         int ret = 0;
 590         int num;
 591
 592         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 593
 594         mutex_lock(&syscall_trace_lock);
 595         if (!sys_perf_refcount_exit)
 596                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 597         if (ret) {
 598                 pr_info("event trace: Could not activate"
 599                                 "syscall exit trace point");
 600         } else {
 601                 set_bit(num, enabled_perf_exit_syscalls);
 602                 sys_perf_refcount_exit++;
 603         }
 604         mutex_unlock(&syscall_trace_lock);
 605         return ret;
 606 }
 607
 608 void perf_sysexit_disable(struct ftrace_event_call *call)
 609 {
 610         int num;
 611
 612         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 613
 614         mutex_lock(&syscall_trace_lock);
 615         sys_perf_refcount_exit--;
 616         clear_bit(num, enabled_perf_exit_syscalls);
 617         if (!sys_perf_refcount_exit)
 618                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 619         mutex_unlock(&syscall_trace_lock);
 620 }
 621
 622 #endif /* CONFIG_PERF_EVENTS */
 623
 624 static int syscall_enter_register(struct ftrace_event_call *event,
 625                                  enum trace_reg type)
 626 {
 627         switch (type) {
 628         case TRACE_REG_REGISTER:
 629                 return reg_event_syscall_enter(event);
 630         case TRACE_REG_UNREGISTER:
 631                 unreg_event_syscall_enter(event);
 632                 return 0;
 633
 634 #ifdef CONFIG_PERF_EVENTS
 635         case TRACE_REG_PERF_REGISTER:
 636                 return perf_sysenter_enable(event);
 637         case TRACE_REG_PERF_UNREGISTER:
 638                 perf_sysenter_disable(event);
 639                 return 0;
 640 #endif
 641         }
 642         return 0;
 643 }
 644
 645 static int syscall_exit_register(struct ftrace_event_call *event,
 646                                  enum trace_reg type)
 647 {
 648         switch (type) {
 649         case TRACE_REG_REGISTER:
 650                 return reg_event_syscall_exit(event);
 651         case TRACE_REG_UNREGISTER:
 652                 unreg_event_syscall_exit(event);
 653                 return 0;
 654
 655 #ifdef CONFIG_PERF_EVENTS
 656         case TRACE_REG_PERF_REGISTER:
 657                 return perf_sysexit_enable(event);
 658         case TRACE_REG_PERF_UNREGISTER:
 659                 perf_sysexit_disable(event);
 660                 return 0;
 661 #endif
 662         }
 663         return 0;
 664 }