kernel/trace/trace_eprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * event probes
   4  *
   5  * Part of this code was copied from kernel/trace/trace_kprobe.c written by
   6  * Masami Hiramatsu <mhiramat@kernel.org>
   7  *
   8  * Copyright (C) 2021, VMware Inc, Steven Rostedt <rostedt@goodmis.org>
   9  * Copyright (C) 2021, VMware Inc, Tzvetomir Stoyanov tz.stoyanov@gmail.com>
  10  *
  11  */
  12 #include <linux/module.h>
  13 #include <linux/mutex.h>
  14 #include <linux/ftrace.h>
  15
  16 #include "trace_dynevent.h"
  17 #include "trace_probe.h"
  18 #include "trace_probe_tmpl.h"
  19
  20 #define EPROBE_EVENT_SYSTEM "eprobes"
  21
  22 struct trace_eprobe {
  23         /* tracepoint system */
  24         const char *event_system;
  25
  26         /* tracepoint event */
  27         const char *event_name;
  28
  29         struct trace_event_call *event;
  30
  31         struct dyn_event        devent;
  32         struct trace_probe      tp;
  33 };
  34
  35 struct eprobe_data {
  36         struct trace_event_file *file;
  37         struct trace_eprobe     *ep;
  38 };
  39
  40 static int __trace_eprobe_create(int argc, const char *argv[]);
  41
  42 static void trace_event_probe_cleanup(struct trace_eprobe *ep)
  43 {
  44         if (!ep)
  45                 return;
  46         trace_probe_cleanup(&ep->tp);
  47         kfree(ep->event_name);
  48         kfree(ep->event_system);
  49         if (ep->event)
  50                 trace_event_put_ref(ep->event);
  51         kfree(ep);
  52 }
  53
  54 static struct trace_eprobe *to_trace_eprobe(struct dyn_event *ev)
  55 {
  56         return container_of(ev, struct trace_eprobe, devent);
  57 }
  58
  59 static int eprobe_dyn_event_create(const char *raw_command)
  60 {
  61         return trace_probe_create(raw_command, __trace_eprobe_create);
  62 }
  63
  64 static int eprobe_dyn_event_show(struct seq_file *m, struct dyn_event *ev)
  65 {
  66         struct trace_eprobe *ep = to_trace_eprobe(ev);
  67         int i;
  68
  69         seq_printf(m, "e:%s/%s", trace_probe_group_name(&ep->tp),
  70                                 trace_probe_name(&ep->tp));
  71         seq_printf(m, " %s.%s", ep->event_system, ep->event_name);
  72
  73         for (i = 0; i < ep->tp.nr_args; i++)
  74                 seq_printf(m, " %s=%s", ep->tp.args[i].name, ep->tp.args[i].comm);
  75         seq_putc(m, '\n');
  76
  77         return 0;
  78 }
  79
  80 static int unregister_trace_eprobe(struct trace_eprobe *ep)
  81 {
  82         /* If other probes are on the event, just unregister eprobe */
  83         if (trace_probe_has_sibling(&ep->tp))
  84                 goto unreg;
  85
  86         /* Enabled event can not be unregistered */
  87         if (trace_probe_is_enabled(&ep->tp))
  88                 return -EBUSY;
  89
  90         /* Will fail if probe is being used by ftrace or perf */
  91         if (trace_probe_unregister_event_call(&ep->tp))
  92                 return -EBUSY;
  93
  94 unreg:
  95         dyn_event_remove(&ep->devent);
  96         trace_probe_unlink(&ep->tp);
  97
  98         return 0;
  99 }
 100
 101 static int eprobe_dyn_event_release(struct dyn_event *ev)
 102 {
 103         struct trace_eprobe *ep = to_trace_eprobe(ev);
 104         int ret = unregister_trace_eprobe(ep);
 105
 106         if (!ret)
 107                 trace_event_probe_cleanup(ep);
 108         return ret;
 109 }
 110
 111 static bool eprobe_dyn_event_is_busy(struct dyn_event *ev)
 112 {
 113         struct trace_eprobe *ep = to_trace_eprobe(ev);
 114
 115         return trace_probe_is_enabled(&ep->tp);
 116 }
 117
 118 static bool eprobe_dyn_event_match(const char *system, const char *event,
 119                         int argc, const char **argv, struct dyn_event *ev)
 120 {
 121         struct trace_eprobe *ep = to_trace_eprobe(ev);
 122         const char *slash;
 123
 124         /*
 125          * We match the following:
 126          *  event only                  - match all eprobes with event name
 127          *  system and event only       - match all system/event probes
 128          *
 129          * The below has the above satisfied with more arguments:
 130          *
 131          *  attached system/event       - If the arg has the system and event
 132          *                                the probe is attached to, match
 133          *                                probes with the attachment.
 134          *
 135          *  If any more args are given, then it requires a full match.
 136          */
 137
 138         /*
 139          * If system exists, but this probe is not part of that system
 140          * do not match.
 141          */
 142         if (system && strcmp(trace_probe_group_name(&ep->tp), system) != 0)
 143                 return false;
 144
 145         /* Must match the event name */
 146         if (strcmp(trace_probe_name(&ep->tp), event) != 0)
 147                 return false;
 148
 149         /* No arguments match all */
 150         if (argc < 1)
 151                 return true;
 152
 153         /* First argument is the system/event the probe is attached to */
 154
 155         slash = strchr(argv[0], '/');
 156         if (!slash)
 157                 slash = strchr(argv[0], '.');
 158         if (!slash)
 159                 return false;
 160
 161         if (strncmp(ep->event_system, argv[0], slash - argv[0]))
 162                 return false;
 163         if (strcmp(ep->event_name, slash + 1))
 164                 return false;
 165
 166         argc--;
 167         argv++;
 168
 169         /* If there are no other args, then match */
 170         if (argc < 1)
 171                 return true;
 172
 173         return trace_probe_match_command_args(&ep->tp, argc, argv);
 174 }
 175
 176 static struct dyn_event_operations eprobe_dyn_event_ops = {
 177         .create = eprobe_dyn_event_create,
 178         .show = eprobe_dyn_event_show,
 179         .is_busy = eprobe_dyn_event_is_busy,
 180         .free = eprobe_dyn_event_release,
 181         .match = eprobe_dyn_event_match,
 182 };
 183
 184 static struct trace_eprobe *alloc_event_probe(const char *group,
 185                                               const char *this_event,
 186                                               struct trace_event_call *event,
 187                                               int nargs)
 188 {
 189         struct trace_eprobe *ep;
 190         const char *event_name;
 191         const char *sys_name;
 192         int ret = -ENOMEM;
 193
 194         if (!event)
 195                 return ERR_PTR(-ENODEV);
 196
 197         sys_name = event->class->system;
 198         event_name = trace_event_name(event);
 199
 200         ep = kzalloc(struct_size(ep, tp.args, nargs), GFP_KERNEL);
 201         if (!ep) {
 202                 trace_event_put_ref(event);
 203                 goto error;
 204         }
 205         ep->event = event;
 206         ep->event_name = kstrdup(event_name, GFP_KERNEL);
 207         if (!ep->event_name)
 208                 goto error;
 209         ep->event_system = kstrdup(sys_name, GFP_KERNEL);
 210         if (!ep->event_system)
 211                 goto error;
 212
 213         ret = trace_probe_init(&ep->tp, this_event, group, false);
 214         if (ret < 0)
 215                 goto error;
 216
 217         dyn_event_init(&ep->devent, &eprobe_dyn_event_ops);
 218         return ep;
 219 error:
 220         trace_event_probe_cleanup(ep);
 221         return ERR_PTR(ret);
 222 }
 223
 224 static int trace_eprobe_tp_arg_update(struct trace_eprobe *ep, int i)
 225 {
 226         struct probe_arg *parg = &ep->tp.args[i];
 227         struct ftrace_event_field *field;
 228         struct list_head *head;
 229
 230         head = trace_get_fields(ep->event);
 231         list_for_each_entry(field, head, link) {
 232                 if (!strcmp(parg->code->data, field->name)) {
 233                         kfree(parg->code->data);
 234                         parg->code->data = field;
 235                         return 0;
 236                 }
 237         }
 238         kfree(parg->code->data);
 239         parg->code->data = NULL;
 240         return -ENOENT;
 241 }
 242
 243 static int eprobe_event_define_fields(struct trace_event_call *event_call)
 244 {
 245         struct eprobe_trace_entry_head field;
 246         struct trace_probe *tp;
 247
 248         tp = trace_probe_primary_from_call(event_call);
 249         if (WARN_ON_ONCE(!tp))
 250                 return -ENOENT;
 251
 252         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
 253 }
 254
 255 static struct trace_event_fields eprobe_fields_array[] = {
 256         { .type = TRACE_FUNCTION_TYPE,
 257           .define_fields = eprobe_event_define_fields },
 258         {}
 259 };
 260
 261 /* Event entry printers */
 262 static enum print_line_t
 263 print_eprobe_event(struct trace_iterator *iter, int flags,
 264                    struct trace_event *event)
 265 {
 266         struct eprobe_trace_entry_head *field;
 267         struct trace_event_call *pevent;
 268         struct trace_event *probed_event;
 269         struct trace_seq *s = &iter->seq;
 270         struct trace_eprobe *ep;
 271         struct trace_probe *tp;
 272         unsigned int type;
 273
 274         field = (struct eprobe_trace_entry_head *)iter->ent;
 275         tp = trace_probe_primary_from_call(
 276                 container_of(event, struct trace_event_call, event));
 277         if (WARN_ON_ONCE(!tp))
 278                 goto out;
 279
 280         ep = container_of(tp, struct trace_eprobe, tp);
 281         type = ep->event->event.type;
 282
 283         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
 284
 285         probed_event = ftrace_find_event(type);
 286         if (probed_event) {
 287                 pevent = container_of(probed_event, struct trace_event_call, event);
 288                 trace_seq_printf(s, "%s.%s", pevent->class->system,
 289                                  trace_event_name(pevent));
 290         } else {
 291                 trace_seq_printf(s, "%u", type);
 292         }
 293
 294         trace_seq_putc(s, ')');
 295
 296         if (print_probe_args(s, tp->args, tp->nr_args,
 297                              (u8 *)&field[1], field) < 0)
 298                 goto out;
 299
 300         trace_seq_putc(s, '\n');
 301  out:
 302         return trace_handle_return(s);
 303 }
 304
 305 static unsigned long get_event_field(struct fetch_insn *code, void *rec)
 306 {
 307         struct ftrace_event_field *field = code->data;
 308         unsigned long val;
 309         void *addr;
 310
 311         addr = rec + field->offset;
 312
 313         switch (field->size) {
 314         case 1:
 315                 if (field->is_signed)
 316                         val = *(char *)addr;
 317                 else
 318                         val = *(unsigned char *)addr;
 319                 break;
 320         case 2:
 321                 if (field->is_signed)
 322                         val = *(short *)addr;
 323                 else
 324                         val = *(unsigned short *)addr;
 325                 break;
 326         case 4:
 327                 if (field->is_signed)
 328                         val = *(int *)addr;
 329                 else
 330                         val = *(unsigned int *)addr;
 331                 break;
 332         default:
 333                 if (field->is_signed)
 334                         val = *(long *)addr;
 335                 else
 336                         val = *(unsigned long *)addr;
 337                 break;
 338         }
 339         return val;
 340 }
 341
 342 static int get_eprobe_size(struct trace_probe *tp, void *rec)
 343 {
 344         struct probe_arg *arg;
 345         int i, len, ret = 0;
 346
 347         for (i = 0; i < tp->nr_args; i++) {
 348                 arg = tp->args + i;
 349                 if (unlikely(arg->dynamic)) {
 350                         unsigned long val;
 351
 352                         val = get_event_field(arg->code, rec);
 353                         len = process_fetch_insn_bottom(arg->code + 1, val, NULL, NULL);
 354                         if (len > 0)
 355                                 ret += len;
 356                 }
 357         }
 358
 359         return ret;
 360 }
 361
 362 /* Kprobe specific fetch functions */
 363
 364 /* Note that we don't verify it, since the code does not come from user space */
 365 static int
 366 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 367                    void *base)
 368 {
 369         unsigned long val;
 370
 371         val = get_event_field(code, rec);
 372         return process_fetch_insn_bottom(code + 1, val, dest, base);
 373 }
 374 NOKPROBE_SYMBOL(process_fetch_insn)
 375
 376 /* Return the length of string -- including null terminal byte */
 377 static nokprobe_inline int
 378 fetch_store_strlen_user(unsigned long addr)
 379 {
 380         const void __user *uaddr =  (__force const void __user *)addr;
 381
 382         return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
 383 }
 384
 385 /* Return the length of string -- including null terminal byte */
 386 static nokprobe_inline int
 387 fetch_store_strlen(unsigned long addr)
 388 {
 389         int ret, len = 0;
 390         u8 c;
 391
 392 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 393         if (addr < TASK_SIZE)
 394                 return fetch_store_strlen_user(addr);
 395 #endif
 396
 397         do {
 398                 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
 399                 len++;
 400         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 401
 402         return (ret < 0) ? ret : len;
 403 }
 404
 405 /*
 406  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
 407  * with max length and relative data location.
 408  */
 409 static nokprobe_inline int
 410 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 411 {
 412         const void __user *uaddr =  (__force const void __user *)addr;
 413         int maxlen = get_loc_len(*(u32 *)dest);
 414         void *__dest;
 415         long ret;
 416
 417         if (unlikely(!maxlen))
 418                 return -ENOMEM;
 419
 420         __dest = get_loc_data(dest, base);
 421
 422         ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
 423         if (ret >= 0)
 424                 *(u32 *)dest = make_data_loc(ret, __dest - base);
 425
 426         return ret;
 427 }
 428
 429 /*
 430  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
 431  * length and relative data location.
 432  */
 433 static nokprobe_inline int
 434 fetch_store_string(unsigned long addr, void *dest, void *base)
 435 {
 436         int maxlen = get_loc_len(*(u32 *)dest);
 437         void *__dest;
 438         long ret;
 439
 440 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 441         if ((unsigned long)addr < TASK_SIZE)
 442                 return fetch_store_string_user(addr, dest, base);
 443 #endif
 444
 445         if (unlikely(!maxlen))
 446                 return -ENOMEM;
 447
 448         __dest = get_loc_data(dest, base);
 449
 450         /*
 451          * Try to get string again, since the string can be changed while
 452          * probing.
 453          */
 454         ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
 455         if (ret >= 0)
 456                 *(u32 *)dest = make_data_loc(ret, __dest - base);
 457
 458         return ret;
 459 }
 460
 461 static nokprobe_inline int
 462 probe_mem_read_user(void *dest, void *src, size_t size)
 463 {
 464         const void __user *uaddr =  (__force const void __user *)src;
 465
 466         return copy_from_user_nofault(dest, uaddr, size);
 467 }
 468
 469 static nokprobe_inline int
 470 probe_mem_read(void *dest, void *src, size_t size)
 471 {
 472 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 473         if ((unsigned long)src < TASK_SIZE)
 474                 return probe_mem_read_user(dest, src, size);
 475 #endif
 476         return copy_from_kernel_nofault(dest, src, size);
 477 }
 478
 479 /* eprobe handler */
 480 static inline void
 481 __eprobe_trace_func(struct eprobe_data *edata, void *rec)
 482 {
 483         struct eprobe_trace_entry_head *entry;
 484         struct trace_event_call *call = trace_probe_event_call(&edata->ep->tp);
 485         struct trace_event_buffer fbuffer;
 486         int dsize;
 487
 488         if (WARN_ON_ONCE(call != edata->file->event_call))
 489                 return;
 490
 491         if (trace_trigger_soft_disabled(edata->file))
 492                 return;
 493
 494         dsize = get_eprobe_size(&edata->ep->tp, rec);
 495
 496         entry = trace_event_buffer_reserve(&fbuffer, edata->file,
 497                                            sizeof(*entry) + edata->ep->tp.size + dsize);
 498
 499         if (!entry)
 500                 return;
 501
 502         entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
 503         store_trace_args(&entry[1], &edata->ep->tp, rec, sizeof(*entry), dsize);
 504
 505         trace_event_buffer_commit(&fbuffer);
 506 }
 507
 508 /*
 509  * The event probe implementation uses event triggers to get access to
 510  * the event it is attached to, but is not an actual trigger. The below
 511  * functions are just stubs to fulfill what is needed to use the trigger
 512  * infrastructure.
 513  */
 514 static int eprobe_trigger_init(struct event_trigger_ops *ops,
 515                                struct event_trigger_data *data)
 516 {
 517         return 0;
 518 }
 519
 520 static void eprobe_trigger_free(struct event_trigger_ops *ops,
 521                                 struct event_trigger_data *data)
 522 {
 523
 524 }
 525
 526 static int eprobe_trigger_print(struct seq_file *m,
 527                                 struct event_trigger_ops *ops,
 528                                 struct event_trigger_data *data)
 529 {
 530         /* Do not print eprobe event triggers */
 531         return 0;
 532 }
 533
 534 static void eprobe_trigger_func(struct event_trigger_data *data,
 535                                 struct trace_buffer *buffer, void *rec,
 536                                 struct ring_buffer_event *rbe)
 537 {
 538         struct eprobe_data *edata = data->private_data;
 539
 540         __eprobe_trace_func(edata, rec);
 541 }
 542
 543 static struct event_trigger_ops eprobe_trigger_ops = {
 544         .trigger                = eprobe_trigger_func,
 545         .print                  = eprobe_trigger_print,
 546         .init                   = eprobe_trigger_init,
 547         .free                   = eprobe_trigger_free,
 548 };
 549
 550 static int eprobe_trigger_cmd_parse(struct event_command *cmd_ops,
 551                                     struct trace_event_file *file,
 552                                     char *glob, char *cmd, char *param)
 553 {
 554         return -1;
 555 }
 556
 557 static int eprobe_trigger_reg_func(char *glob,
 558                                    struct event_trigger_data *data,
 559                                    struct trace_event_file *file)
 560 {
 561         return -1;
 562 }
 563
 564 static void eprobe_trigger_unreg_func(char *glob,
 565                                       struct event_trigger_data *data,
 566                                       struct trace_event_file *file)
 567 {
 568
 569 }
 570
 571 static struct event_trigger_ops *eprobe_trigger_get_ops(char *cmd,
 572                                                         char *param)
 573 {
 574         return &eprobe_trigger_ops;
 575 }
 576
 577 static struct event_command event_trigger_cmd = {
 578         .name                   = "eprobe",
 579         .trigger_type           = ETT_EVENT_EPROBE,
 580         .flags                  = EVENT_CMD_FL_NEEDS_REC,
 581         .parse                  = eprobe_trigger_cmd_parse,
 582         .reg                    = eprobe_trigger_reg_func,
 583         .unreg                  = eprobe_trigger_unreg_func,
 584         .unreg_all              = NULL,
 585         .get_trigger_ops        = eprobe_trigger_get_ops,
 586         .set_filter             = NULL,
 587 };
 588
 589 static struct event_trigger_data *
 590 new_eprobe_trigger(struct trace_eprobe *ep, struct trace_event_file *file)
 591 {
 592         struct event_trigger_data *trigger;
 593         struct eprobe_data *edata;
 594
 595         edata = kzalloc(sizeof(*edata), GFP_KERNEL);
 596         trigger = kzalloc(sizeof(*trigger), GFP_KERNEL);
 597         if (!trigger || !edata) {
 598                 kfree(edata);
 599                 kfree(trigger);
 600                 return ERR_PTR(-ENOMEM);
 601         }
 602
 603         trigger->flags = EVENT_TRIGGER_FL_PROBE;
 604         trigger->count = -1;
 605         trigger->ops = &eprobe_trigger_ops;
 606
 607         /*
 608          * EVENT PROBE triggers are not registered as commands with
 609          * register_event_command(), as they are not controlled by the user
 610          * from the trigger file
 611          */
 612         trigger->cmd_ops = &event_trigger_cmd;
 613
 614         INIT_LIST_HEAD(&trigger->list);
 615         RCU_INIT_POINTER(trigger->filter, NULL);
 616
 617         edata->file = file;
 618         edata->ep = ep;
 619         trigger->private_data = edata;
 620
 621         return trigger;
 622 }
 623
 624 static int enable_eprobe(struct trace_eprobe *ep,
 625                          struct trace_event_file *eprobe_file)
 626 {
 627         struct event_trigger_data *trigger;
 628         struct trace_event_file *file;
 629         struct trace_array *tr = eprobe_file->tr;
 630
 631         file = find_event_file(tr, ep->event_system, ep->event_name);
 632         if (!file)
 633                 return -ENOENT;
 634         trigger = new_eprobe_trigger(ep, eprobe_file);
 635         if (IS_ERR(trigger))
 636                 return PTR_ERR(trigger);
 637
 638         list_add_tail_rcu(&trigger->list, &file->triggers);
 639
 640         trace_event_trigger_enable_disable(file, 1);
 641         update_cond_flag(file);
 642
 643         return 0;
 644 }
 645
 646 static struct trace_event_functions eprobe_funcs = {
 647         .trace          = print_eprobe_event
 648 };
 649
 650 static int disable_eprobe(struct trace_eprobe *ep,
 651                           struct trace_array *tr)
 652 {
 653         struct event_trigger_data *trigger;
 654         struct trace_event_file *file;
 655         struct eprobe_data *edata;
 656
 657         file = find_event_file(tr, ep->event_system, ep->event_name);
 658         if (!file)
 659                 return -ENOENT;
 660
 661         list_for_each_entry(trigger, &file->triggers, list) {
 662                 if (!(trigger->flags & EVENT_TRIGGER_FL_PROBE))
 663                         continue;
 664                 edata = trigger->private_data;
 665                 if (edata->ep == ep)
 666                         break;
 667         }
 668         if (list_entry_is_head(trigger, &file->triggers, list))
 669                 return -ENODEV;
 670
 671         list_del_rcu(&trigger->list);
 672
 673         trace_event_trigger_enable_disable(file, 0);
 674         update_cond_flag(file);
 675
 676         /* Make sure nothing is using the edata or trigger */
 677         tracepoint_synchronize_unregister();
 678
 679         kfree(edata);
 680         kfree(trigger);
 681
 682         return 0;
 683 }
 684
 685 static int enable_trace_eprobe(struct trace_event_call *call,
 686                                struct trace_event_file *file)
 687 {
 688         struct trace_probe *pos, *tp;
 689         struct trace_eprobe *ep;
 690         bool enabled;
 691         int ret = 0;
 692
 693         tp = trace_probe_primary_from_call(call);
 694         if (WARN_ON_ONCE(!tp))
 695                 return -ENODEV;
 696         enabled = trace_probe_is_enabled(tp);
 697
 698         /* This also changes "enabled" state */
 699         if (file) {
 700                 ret = trace_probe_add_file(tp, file);
 701                 if (ret)
 702                         return ret;
 703         } else
 704                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
 705
 706         if (enabled)
 707                 return 0;
 708
 709         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 710                 ep = container_of(pos, struct trace_eprobe, tp);
 711                 ret = enable_eprobe(ep, file);
 712                 if (ret)
 713                         break;
 714                 enabled = true;
 715         }
 716
 717         if (ret) {
 718                 /* Failed to enable one of them. Roll back all */
 719                 if (enabled)
 720                         disable_eprobe(ep, file->tr);
 721                 if (file)
 722                         trace_probe_remove_file(tp, file);
 723                 else
 724                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 725         }
 726
 727         return ret;
 728 }
 729
 730 static int disable_trace_eprobe(struct trace_event_call *call,
 731                                 struct trace_event_file *file)
 732 {
 733         struct trace_probe *pos, *tp;
 734         struct trace_eprobe *ep;
 735
 736         tp = trace_probe_primary_from_call(call);
 737         if (WARN_ON_ONCE(!tp))
 738                 return -ENODEV;
 739
 740         if (file) {
 741                 if (!trace_probe_get_file_link(tp, file))
 742                         return -ENOENT;
 743                 if (!trace_probe_has_single_file(tp))
 744                         goto out;
 745                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 746         } else
 747                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 748
 749         if (!trace_probe_is_enabled(tp)) {
 750                 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 751                         ep = container_of(pos, struct trace_eprobe, tp);
 752                         disable_eprobe(ep, file->tr);
 753                 }
 754         }
 755
 756  out:
 757         if (file)
 758                 /*
 759                  * Synchronization is done in below function. For perf event,
 760                  * file == NULL and perf_trace_event_unreg() calls
 761                  * tracepoint_synchronize_unregister() to ensure synchronize
 762                  * event. We don't need to care about it.
 763                  */
 764                 trace_probe_remove_file(tp, file);
 765
 766         return 0;
 767 }
 768
 769 static int eprobe_register(struct trace_event_call *event,
 770                            enum trace_reg type, void *data)
 771 {
 772         struct trace_event_file *file = data;
 773
 774         switch (type) {
 775         case TRACE_REG_REGISTER:
 776                 return enable_trace_eprobe(event, file);
 777         case TRACE_REG_UNREGISTER:
 778                 return disable_trace_eprobe(event, file);
 779 #ifdef CONFIG_PERF_EVENTS
 780         case TRACE_REG_PERF_REGISTER:
 781         case TRACE_REG_PERF_UNREGISTER:
 782         case TRACE_REG_PERF_OPEN:
 783         case TRACE_REG_PERF_CLOSE:
 784         case TRACE_REG_PERF_ADD:
 785         case TRACE_REG_PERF_DEL:
 786                 return 0;
 787 #endif
 788         }
 789         return 0;
 790 }
 791
 792 static inline void init_trace_eprobe_call(struct trace_eprobe *ep)
 793 {
 794         struct trace_event_call *call = trace_probe_event_call(&ep->tp);
 795
 796         call->flags = TRACE_EVENT_FL_EPROBE;
 797         call->event.funcs = &eprobe_funcs;
 798         call->class->fields_array = eprobe_fields_array;
 799         call->class->reg = eprobe_register;
 800 }
 801
 802 static struct trace_event_call *
 803 find_and_get_event(const char *system, const char *event_name)
 804 {
 805         struct trace_event_call *tp_event;
 806         const char *name;
 807
 808         list_for_each_entry(tp_event, &ftrace_events, list) {
 809                 /* Skip other probes and ftrace events */
 810                 if (tp_event->flags &
 811                     (TRACE_EVENT_FL_IGNORE_ENABLE |
 812                      TRACE_EVENT_FL_KPROBE |
 813                      TRACE_EVENT_FL_UPROBE |
 814                      TRACE_EVENT_FL_EPROBE))
 815                         continue;
 816                 if (!tp_event->class->system ||
 817                     strcmp(system, tp_event->class->system))
 818                         continue;
 819                 name = trace_event_name(tp_event);
 820                 if (!name || strcmp(event_name, name))
 821                         continue;
 822                 if (!trace_event_try_get_ref(tp_event)) {
 823                         return NULL;
 824                         break;
 825                 }
 826                 return tp_event;
 827                 break;
 828         }
 829         return NULL;
 830 }
 831
 832 static int trace_eprobe_tp_update_arg(struct trace_eprobe *ep, const char *argv[], int i)
 833 {
 834         unsigned int flags = TPARG_FL_KERNEL | TPARG_FL_TPOINT;
 835         int ret;
 836
 837         ret = traceprobe_parse_probe_arg(&ep->tp, i, argv[i], flags);
 838         if (ret)
 839                 return ret;
 840
 841         if (ep->tp.args[i].code->op == FETCH_OP_TP_ARG)
 842                 ret = trace_eprobe_tp_arg_update(ep, i);
 843
 844         return ret;
 845 }
 846
 847 static int __trace_eprobe_create(int argc, const char *argv[])
 848 {
 849         /*
 850          * Argument syntax:
 851          *      e[:[GRP/]ENAME] SYSTEM.EVENT [FETCHARGS]
 852          * Fetch args:
 853          *  <name>=$<field>[:TYPE]
 854          */
 855         const char *event = NULL, *group = EPROBE_EVENT_SYSTEM;
 856         const char *sys_event = NULL, *sys_name = NULL;
 857         struct trace_event_call *event_call;
 858         struct trace_eprobe *ep = NULL;
 859         char buf1[MAX_EVENT_NAME_LEN];
 860         char buf2[MAX_EVENT_NAME_LEN];
 861         int ret = 0;
 862         int i;
 863
 864         if (argc < 2 || argv[0][0] != 'e')
 865                 return -ECANCELED;
 866
 867         trace_probe_log_init("event_probe", argc, argv);
 868
 869         event = strchr(&argv[0][1], ':');
 870         if (event) {
 871                 event++;
 872                 ret = traceprobe_parse_event_name(&event, &group, buf1,
 873                                                   event - argv[0]);
 874                 if (ret)
 875                         goto parse_error;
 876         } else {
 877                 strscpy(buf1, argv[1], MAX_EVENT_NAME_LEN);
 878                 sanitize_event_name(buf1);
 879                 event = buf1;
 880         }
 881         if (!is_good_name(event) || !is_good_name(group))
 882                 goto parse_error;
 883
 884         sys_event = argv[1];
 885         ret = traceprobe_parse_event_name(&sys_event, &sys_name, buf2,
 886                                           sys_event - argv[1]);
 887         if (ret || !sys_name)
 888                 goto parse_error;
 889         if (!is_good_name(sys_event) || !is_good_name(sys_name))
 890                 goto parse_error;
 891
 892         mutex_lock(&event_mutex);
 893         event_call = find_and_get_event(sys_name, sys_event);
 894         ep = alloc_event_probe(group, event, event_call, argc - 2);
 895         mutex_unlock(&event_mutex);
 896
 897         if (IS_ERR(ep)) {
 898                 ret = PTR_ERR(ep);
 899                 /* This must return -ENOMEM or missing event, else there is a bug */
 900                 WARN_ON_ONCE(ret != -ENOMEM && ret != -ENODEV);
 901                 ep = NULL;
 902                 goto error;
 903         }
 904
 905         argc -= 2; argv += 2;
 906         /* parse arguments */
 907         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 908                 trace_probe_log_set_index(i + 2);
 909                 ret = trace_eprobe_tp_update_arg(ep, argv, i);
 910                 if (ret)
 911                         goto error;
 912         }
 913         ret = traceprobe_set_print_fmt(&ep->tp, PROBE_PRINT_EVENT);
 914         if (ret < 0)
 915                 goto error;
 916         init_trace_eprobe_call(ep);
 917         mutex_lock(&event_mutex);
 918         ret = trace_probe_register_event_call(&ep->tp);
 919         if (ret) {
 920                 if (ret == -EEXIST) {
 921                         trace_probe_log_set_index(0);
 922                         trace_probe_log_err(0, EVENT_EXIST);
 923                 }
 924                 mutex_unlock(&event_mutex);
 925                 goto error;
 926         }
 927         ret = dyn_event_add(&ep->devent, &ep->tp.event->call);
 928         mutex_unlock(&event_mutex);
 929         return ret;
 930 parse_error:
 931         ret = -EINVAL;
 932 error:
 933         trace_event_probe_cleanup(ep);
 934         return ret;
 935 }
 936
 937 /*
 938  * Register dynevent at core_initcall. This allows kernel to setup eprobe
 939  * events in postcore_initcall without tracefs.
 940  */
 941 static __init int trace_events_eprobe_init_early(void)
 942 {
 943         int err = 0;
 944
 945         err = dyn_event_register(&eprobe_dyn_event_ops);
 946         if (err)
 947                 pr_warn("Could not register eprobe_dyn_event_ops\n");
 948
 949         return err;
 950 }
 951 core_initcall(trace_events_eprobe_init_early);