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 struct trace_event_functions enter_syscall_print_funcs = {
  35         .trace          = print_syscall_enter,
  36 };
  37
  38 struct trace_event_functions exit_syscall_print_funcs = {
  39         .trace          = print_syscall_exit,
  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         .raw_init       = init_syscall_trace,
  48 };
  49
  50 struct ftrace_event_class event_class_syscall_exit = {
  51         .system         = "syscalls",
  52         .reg            = syscall_exit_register,
  53         .define_fields  = syscall_exit_define_fields,
  54         .fields         = LIST_HEAD_INIT(event_class_syscall_exit.fields),
  55         .raw_init       = init_syscall_trace,
  56 };
  57
  58 extern struct syscall_metadata *__start_syscalls_metadata[];
  59 extern struct syscall_metadata *__stop_syscalls_metadata[];
  60
  61 static struct syscall_metadata **syscalls_metadata;
  62
  63 static __init struct syscall_metadata *
  64 find_syscall_meta(unsigned long syscall)
  65 {
  66         struct syscall_metadata **start;
  67         struct syscall_metadata **stop;
  68         char str[KSYM_SYMBOL_LEN];
  69
  70
  71         start = __start_syscalls_metadata;
  72         stop = __stop_syscalls_metadata;
  73         kallsyms_lookup(syscall, NULL, NULL, NULL, str);
  74
  75         for ( ; start < stop; start++) {
  76                 /*
  77                  * Only compare after the "sys" prefix. Archs that use
  78                  * syscall wrappers may have syscalls symbols aliases prefixed
  79                  * with "SyS" instead of "sys", leading to an unwanted
  80                  * mismatch.
  81                  */
  82                 if ((*start)->name && !strcmp((*start)->name + 3, str + 3))
  83                         return *start;
  84         }
  85         return NULL;
  86 }
  87
  88 static struct syscall_metadata *syscall_nr_to_meta(int nr)
  89 {
  90         if (!syscalls_metadata || nr >= NR_syscalls || nr < 0)
  91                 return NULL;
  92
  93         return syscalls_metadata[nr];
  94 }
  95
  96 enum print_line_t
  97 print_syscall_enter(struct trace_iterator *iter, int flags,
  98                     struct trace_event *event)
  99 {
 100         struct trace_seq *s = &iter->seq;
 101         struct trace_entry *ent = iter->ent;
 102         struct syscall_trace_enter *trace;
 103         struct syscall_metadata *entry;
 104         int i, ret, syscall;
 105
 106         trace = (typeof(trace))ent;
 107         syscall = trace->nr;
 108         entry = syscall_nr_to_meta(syscall);
 109
 110         if (!entry)
 111                 goto end;
 112
 113         if (entry->enter_event->event.type != ent->type) {
 114                 WARN_ON_ONCE(1);
 115                 goto end;
 116         }
 117
 118         ret = trace_seq_printf(s, "%s(", entry->name);
 119         if (!ret)
 120                 return TRACE_TYPE_PARTIAL_LINE;
 121
 122         for (i = 0; i < entry->nb_args; i++) {
 123                 /* parameter types */
 124                 if (trace_flags & TRACE_ITER_VERBOSE) {
 125                         ret = trace_seq_printf(s, "%s ", entry->types[i]);
 126                         if (!ret)
 127                                 return TRACE_TYPE_PARTIAL_LINE;
 128                 }
 129                 /* parameter values */
 130                 ret = trace_seq_printf(s, "%s: %lx%s", entry->args[i],
 131                                        trace->args[i],
 132                                        i == entry->nb_args - 1 ? "" : ", ");
 133                 if (!ret)
 134                         return TRACE_TYPE_PARTIAL_LINE;
 135         }
 136
 137         ret = trace_seq_putc(s, ')');
 138         if (!ret)
 139                 return TRACE_TYPE_PARTIAL_LINE;
 140
 141 end:
 142         ret =  trace_seq_putc(s, '\n');
 143         if (!ret)
 144                 return TRACE_TYPE_PARTIAL_LINE;
 145
 146         return TRACE_TYPE_HANDLED;
 147 }
 148
 149 enum print_line_t
 150 print_syscall_exit(struct trace_iterator *iter, int flags,
 151                    struct trace_event *event)
 152 {
 153         struct trace_seq *s = &iter->seq;
 154         struct trace_entry *ent = iter->ent;
 155         struct syscall_trace_exit *trace;
 156         int syscall;
 157         struct syscall_metadata *entry;
 158         int ret;
 159
 160         trace = (typeof(trace))ent;
 161         syscall = trace->nr;
 162         entry = syscall_nr_to_meta(syscall);
 163
 164         if (!entry) {
 165                 trace_seq_printf(s, "\n");
 166                 return TRACE_TYPE_HANDLED;
 167         }
 168
 169         if (entry->exit_event->event.type != ent->type) {
 170                 WARN_ON_ONCE(1);
 171                 return TRACE_TYPE_UNHANDLED;
 172         }
 173
 174         ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
 175                                 trace->ret);
 176         if (!ret)
 177                 return TRACE_TYPE_PARTIAL_LINE;
 178
 179         return TRACE_TYPE_HANDLED;
 180 }
 181
 182 extern char *__bad_type_size(void);
 183
 184 #define SYSCALL_FIELD(type, name)                                       \
 185         sizeof(type) != sizeof(trace.name) ?                            \
 186                 __bad_type_size() :                                     \
 187                 #type, #name, offsetof(typeof(trace), name),            \
 188                 sizeof(trace.name), is_signed_type(type)
 189
 190 static
 191 int  __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 192 {
 193         int i;
 194         int pos = 0;
 195
 196         /* When len=0, we just calculate the needed length */
 197 #define LEN_OR_ZERO (len ? len - pos : 0)
 198
 199         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 200         for (i = 0; i < entry->nb_args; i++) {
 201                 pos += snprintf(buf + pos, LEN_OR_ZERO, "%s: 0x%%0%zulx%s",
 202                                 entry->args[i], sizeof(unsigned long),
 203                                 i == entry->nb_args - 1 ? "" : ", ");
 204         }
 205         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"");
 206
 207         for (i = 0; i < entry->nb_args; i++) {
 208                 pos += snprintf(buf + pos, LEN_OR_ZERO,
 209                                 ", ((unsigned long)(REC->%s))", entry->args[i]);
 210         }
 211
 212 #undef LEN_OR_ZERO
 213
 214         /* return the length of print_fmt */
 215         return pos;
 216 }
 217
 218 static int set_syscall_print_fmt(struct ftrace_event_call *call)
 219 {
 220         char *print_fmt;
 221         int len;
 222         struct syscall_metadata *entry = call->data;
 223
 224         if (entry->enter_event != call) {
 225                 call->print_fmt = "\"0x%lx\", REC->ret";
 226                 return 0;
 227         }
 228
 229         /* First: called with 0 length to calculate the needed length */
 230         len = __set_enter_print_fmt(entry, NULL, 0);
 231
 232         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 233         if (!print_fmt)
 234                 return -ENOMEM;
 235
 236         /* Second: actually write the @print_fmt */
 237         __set_enter_print_fmt(entry, print_fmt, len + 1);
 238         call->print_fmt = print_fmt;
 239
 240         return 0;
 241 }
 242
 243 static void free_syscall_print_fmt(struct ftrace_event_call *call)
 244 {
 245         struct syscall_metadata *entry = call->data;
 246
 247         if (entry->enter_event == call)
 248                 kfree(call->print_fmt);
 249 }
 250
 251 static int syscall_enter_define_fields(struct ftrace_event_call *call)
 252 {
 253         struct syscall_trace_enter trace;
 254         struct syscall_metadata *meta = call->data;
 255         int ret;
 256         int i;
 257         int offset = offsetof(typeof(trace), args);
 258
 259         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 260         if (ret)
 261                 return ret;
 262
 263         for (i = 0; i < meta->nb_args; i++) {
 264                 ret = trace_define_field(call, meta->types[i],
 265                                          meta->args[i], offset,
 266                                          sizeof(unsigned long), 0,
 267                                          FILTER_OTHER);
 268                 offset += sizeof(unsigned long);
 269         }
 270
 271         return ret;
 272 }
 273
 274 static int syscall_exit_define_fields(struct ftrace_event_call *call)
 275 {
 276         struct syscall_trace_exit trace;
 277         int ret;
 278
 279         ret = trace_define_field(call, SYSCALL_FIELD(int, nr), FILTER_OTHER);
 280         if (ret)
 281                 return ret;
 282
 283         ret = trace_define_field(call, SYSCALL_FIELD(long, ret),
 284                                  FILTER_OTHER);
 285
 286         return ret;
 287 }
 288
 289 void ftrace_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 290 {
 291         struct syscall_trace_enter *entry;
 292         struct syscall_metadata *sys_data;
 293         struct ring_buffer_event *event;
 294         struct ring_buffer *buffer;
 295         int size;
 296         int syscall_nr;
 297
 298         syscall_nr = syscall_get_nr(current, regs);
 299         if (syscall_nr < 0)
 300                 return;
 301         if (!test_bit(syscall_nr, enabled_enter_syscalls))
 302                 return;
 303
 304         sys_data = syscall_nr_to_meta(syscall_nr);
 305         if (!sys_data)
 306                 return;
 307
 308         size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
 309
 310         event = trace_current_buffer_lock_reserve(&buffer,
 311                         sys_data->enter_event->event.type, size, 0, 0);
 312         if (!event)
 313                 return;
 314
 315         entry = ring_buffer_event_data(event);
 316         entry->nr = syscall_nr;
 317         syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
 318
 319         if (!filter_current_check_discard(buffer, sys_data->enter_event,
 320                                           entry, event))
 321                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 322 }
 323
 324 void ftrace_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 325 {
 326         struct syscall_trace_exit *entry;
 327         struct syscall_metadata *sys_data;
 328         struct ring_buffer_event *event;
 329         struct ring_buffer *buffer;
 330         int syscall_nr;
 331
 332         syscall_nr = syscall_get_nr(current, regs);
 333         if (syscall_nr < 0)
 334                 return;
 335         if (!test_bit(syscall_nr, enabled_exit_syscalls))
 336                 return;
 337
 338         sys_data = syscall_nr_to_meta(syscall_nr);
 339         if (!sys_data)
 340                 return;
 341
 342         event = trace_current_buffer_lock_reserve(&buffer,
 343                         sys_data->exit_event->event.type, sizeof(*entry), 0, 0);
 344         if (!event)
 345                 return;
 346
 347         entry = ring_buffer_event_data(event);
 348         entry->nr = syscall_nr;
 349         entry->ret = syscall_get_return_value(current, regs);
 350
 351         if (!filter_current_check_discard(buffer, sys_data->exit_event,
 352                                           entry, event))
 353                 trace_current_buffer_unlock_commit(buffer, event, 0, 0);
 354 }
 355
 356 int reg_event_syscall_enter(struct ftrace_event_call *call)
 357 {
 358         int ret = 0;
 359         int num;
 360
 361         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 362         if (num < 0 || num >= NR_syscalls)
 363                 return -ENOSYS;
 364         mutex_lock(&syscall_trace_lock);
 365         if (!sys_refcount_enter)
 366                 ret = register_trace_sys_enter(ftrace_syscall_enter, NULL);
 367         if (!ret) {
 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
 379         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 380         if (num < 0 || num >= NR_syscalls)
 381                 return;
 382         mutex_lock(&syscall_trace_lock);
 383         sys_refcount_enter--;
 384         clear_bit(num, enabled_enter_syscalls);
 385         if (!sys_refcount_enter)
 386                 unregister_trace_sys_enter(ftrace_syscall_enter, NULL);
 387         mutex_unlock(&syscall_trace_lock);
 388 }
 389
 390 int reg_event_syscall_exit(struct ftrace_event_call *call)
 391 {
 392         int ret = 0;
 393         int num;
 394
 395         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 396         if (num < 0 || num >= NR_syscalls)
 397                 return -ENOSYS;
 398         mutex_lock(&syscall_trace_lock);
 399         if (!sys_refcount_exit)
 400                 ret = register_trace_sys_exit(ftrace_syscall_exit, NULL);
 401         if (!ret) {
 402                 set_bit(num, enabled_exit_syscalls);
 403                 sys_refcount_exit++;
 404         }
 405         mutex_unlock(&syscall_trace_lock);
 406         return ret;
 407 }
 408
 409 void unreg_event_syscall_exit(struct ftrace_event_call *call)
 410 {
 411         int num;
 412
 413         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 414         if (num < 0 || num >= NR_syscalls)
 415                 return;
 416         mutex_lock(&syscall_trace_lock);
 417         sys_refcount_exit--;
 418         clear_bit(num, enabled_exit_syscalls);
 419         if (!sys_refcount_exit)
 420                 unregister_trace_sys_exit(ftrace_syscall_exit, NULL);
 421         mutex_unlock(&syscall_trace_lock);
 422 }
 423
 424 int init_syscall_trace(struct ftrace_event_call *call)
 425 {
 426         int id;
 427
 428         if (set_syscall_print_fmt(call) < 0)
 429                 return -ENOMEM;
 430
 431         id = trace_event_raw_init(call);
 432
 433         if (id < 0) {
 434                 free_syscall_print_fmt(call);
 435                 return id;
 436         }
 437
 438         return id;
 439 }
 440
 441 unsigned long __init arch_syscall_addr(int nr)
 442 {
 443         return (unsigned long)sys_call_table[nr];
 444 }
 445
 446 int __init init_ftrace_syscalls(void)
 447 {
 448         struct syscall_metadata *meta;
 449         unsigned long addr;
 450         int i;
 451
 452         syscalls_metadata = kzalloc(sizeof(*syscalls_metadata) *
 453                                         NR_syscalls, GFP_KERNEL);
 454         if (!syscalls_metadata) {
 455                 WARN_ON(1);
 456                 return -ENOMEM;
 457         }
 458
 459         for (i = 0; i < NR_syscalls; i++) {
 460                 addr = arch_syscall_addr(i);
 461                 meta = find_syscall_meta(addr);
 462                 if (!meta)
 463                         continue;
 464
 465                 meta->syscall_nr = i;
 466                 syscalls_metadata[i] = meta;
 467         }
 468
 469         return 0;
 470 }
 471 core_initcall(init_ftrace_syscalls);
 472
 473 #ifdef CONFIG_PERF_EVENTS
 474
 475 static DECLARE_BITMAP(enabled_perf_enter_syscalls, NR_syscalls);
 476 static DECLARE_BITMAP(enabled_perf_exit_syscalls, NR_syscalls);
 477 static int sys_perf_refcount_enter;
 478 static int sys_perf_refcount_exit;
 479
 480 static void perf_syscall_enter(void *ignore, struct pt_regs *regs, long id)
 481 {
 482         struct syscall_metadata *sys_data;
 483         struct syscall_trace_enter *rec;
 484         struct hlist_head *head;
 485         int syscall_nr;
 486         int rctx;
 487         int size;
 488
 489         syscall_nr = syscall_get_nr(current, regs);
 490         if (!test_bit(syscall_nr, enabled_perf_enter_syscalls))
 491                 return;
 492
 493         sys_data = syscall_nr_to_meta(syscall_nr);
 494         if (!sys_data)
 495                 return;
 496
 497         /* get the size after alignment with the u32 buffer size field */
 498         size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 499         size = ALIGN(size + sizeof(u32), sizeof(u64));
 500         size -= sizeof(u32);
 501
 502         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 503                       "perf buffer not large enough"))
 504                 return;
 505
 506         rec = (struct syscall_trace_enter *)perf_trace_buf_prepare(size,
 507                                 sys_data->enter_event->event.type, regs, &rctx);
 508         if (!rec)
 509                 return;
 510
 511         rec->nr = syscall_nr;
 512         syscall_get_arguments(current, regs, 0, sys_data->nb_args,
 513                                (unsigned long *)&rec->args);
 514
 515         head = this_cpu_ptr(sys_data->enter_event->perf_events);
 516         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head);
 517 }
 518
 519 int perf_sysenter_enable(struct ftrace_event_call *call)
 520 {
 521         int ret = 0;
 522         int num;
 523
 524         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 525
 526         mutex_lock(&syscall_trace_lock);
 527         if (!sys_perf_refcount_enter)
 528                 ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 529         if (ret) {
 530                 pr_info("event trace: Could not activate"
 531                                 "syscall entry trace point");
 532         } else {
 533                 set_bit(num, enabled_perf_enter_syscalls);
 534                 sys_perf_refcount_enter++;
 535         }
 536         mutex_unlock(&syscall_trace_lock);
 537         return ret;
 538 }
 539
 540 void perf_sysenter_disable(struct ftrace_event_call *call)
 541 {
 542         int num;
 543
 544         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 545
 546         mutex_lock(&syscall_trace_lock);
 547         sys_perf_refcount_enter--;
 548         clear_bit(num, enabled_perf_enter_syscalls);
 549         if (!sys_perf_refcount_enter)
 550                 unregister_trace_sys_enter(perf_syscall_enter, NULL);
 551         mutex_unlock(&syscall_trace_lock);
 552 }
 553
 554 static void perf_syscall_exit(void *ignore, struct pt_regs *regs, long ret)
 555 {
 556         struct syscall_metadata *sys_data;
 557         struct syscall_trace_exit *rec;
 558         struct hlist_head *head;
 559         int syscall_nr;
 560         int rctx;
 561         int size;
 562
 563         syscall_nr = syscall_get_nr(current, regs);
 564         if (!test_bit(syscall_nr, enabled_perf_exit_syscalls))
 565                 return;
 566
 567         sys_data = syscall_nr_to_meta(syscall_nr);
 568         if (!sys_data)
 569                 return;
 570
 571         /* We can probably do that at build time */
 572         size = ALIGN(sizeof(*rec) + sizeof(u32), sizeof(u64));
 573         size -= sizeof(u32);
 574
 575         /*
 576          * Impossible, but be paranoid with the future
 577          * How to put this check outside runtime?
 578          */
 579         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE,
 580                 "exit event has grown above perf buffer size"))
 581                 return;
 582
 583         rec = (struct syscall_trace_exit *)perf_trace_buf_prepare(size,
 584                                 sys_data->exit_event->event.type, regs, &rctx);
 585         if (!rec)
 586                 return;
 587
 588         rec->nr = syscall_nr;
 589         rec->ret = syscall_get_return_value(current, regs);
 590
 591         head = this_cpu_ptr(sys_data->exit_event->perf_events);
 592         perf_trace_buf_submit(rec, size, rctx, 0, 1, regs, head);
 593 }
 594
 595 int perf_sysexit_enable(struct ftrace_event_call *call)
 596 {
 597         int ret = 0;
 598         int num;
 599
 600         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 601
 602         mutex_lock(&syscall_trace_lock);
 603         if (!sys_perf_refcount_exit)
 604                 ret = register_trace_sys_exit(perf_syscall_exit, NULL);
 605         if (ret) {
 606                 pr_info("event trace: Could not activate"
 607                                 "syscall exit trace point");
 608         } else {
 609                 set_bit(num, enabled_perf_exit_syscalls);
 610                 sys_perf_refcount_exit++;
 611         }
 612         mutex_unlock(&syscall_trace_lock);
 613         return ret;
 614 }
 615
 616 void perf_sysexit_disable(struct ftrace_event_call *call)
 617 {
 618         int num;
 619
 620         num = ((struct syscall_metadata *)call->data)->syscall_nr;
 621
 622         mutex_lock(&syscall_trace_lock);
 623         sys_perf_refcount_exit--;
 624         clear_bit(num, enabled_perf_exit_syscalls);
 625         if (!sys_perf_refcount_exit)
 626                 unregister_trace_sys_exit(perf_syscall_exit, NULL);
 627         mutex_unlock(&syscall_trace_lock);
 628 }
 629
 630 #endif /* CONFIG_PERF_EVENTS */
 631
 632 static int syscall_enter_register(struct ftrace_event_call *event,
 633                                  enum trace_reg type)
 634 {
 635         switch (type) {
 636         case TRACE_REG_REGISTER:
 637                 return reg_event_syscall_enter(event);
 638         case TRACE_REG_UNREGISTER:
 639                 unreg_event_syscall_enter(event);
 640                 return 0;
 641
 642 #ifdef CONFIG_PERF_EVENTS
 643         case TRACE_REG_PERF_REGISTER:
 644                 return perf_sysenter_enable(event);
 645         case TRACE_REG_PERF_UNREGISTER:
 646                 perf_sysenter_disable(event);
 647                 return 0;
 648 #endif
 649         }
 650         return 0;
 651 }
 652
 653 static int syscall_exit_register(struct ftrace_event_call *event,
 654                                  enum trace_reg type)
 655 {
 656         switch (type) {
 657         case TRACE_REG_REGISTER:
 658                 return reg_event_syscall_exit(event);
 659         case TRACE_REG_UNREGISTER:
 660                 unreg_event_syscall_exit(event);
 661                 return 0;
 662
 663 #ifdef CONFIG_PERF_EVENTS
 664         case TRACE_REG_PERF_REGISTER:
 665                 return perf_sysexit_enable(event);
 666         case TRACE_REG_PERF_UNREGISTER:
 667                 perf_sysexit_disable(event);
 668                 return 0;
 669 #endif
 670         }
 671         return 0;
 672 }