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          *  system only                 - match all system probes
 129          *
 130          * The below has the above satisfied with more arguments:
 131          *
 132          *  attached system/event       - If the arg has the system and event
 133          *                                the probe is attached to, match
 134          *                                probes with the attachment.
 135          *
 136          *  If any more args are given, then it requires a full match.
 137          */
 138
 139         /*
 140          * If system exists, but this probe is not part of that system
 141          * do not match.
 142          */
 143         if (system && strcmp(trace_probe_group_name(&ep->tp), system) != 0)
 144                 return false;
 145
 146         /* Must match the event name */
 147         if (event[0] != '\0' && strcmp(trace_probe_name(&ep->tp), event) != 0)
 148                 return false;
 149
 150         /* No arguments match all */
 151         if (argc < 1)
 152                 return true;
 153
 154         /* First argument is the system/event the probe is attached to */
 155
 156         slash = strchr(argv[0], '/');
 157         if (!slash)
 158                 slash = strchr(argv[0], '.');
 159         if (!slash)
 160                 return false;
 161
 162         if (strncmp(ep->event_system, argv[0], slash - argv[0]))
 163                 return false;
 164         if (strcmp(ep->event_name, slash + 1))
 165                 return false;
 166
 167         argc--;
 168         argv++;
 169
 170         /* If there are no other args, then match */
 171         if (argc < 1)
 172                 return true;
 173
 174         return trace_probe_match_command_args(&ep->tp, argc, argv);
 175 }
 176
 177 static struct dyn_event_operations eprobe_dyn_event_ops = {
 178         .create = eprobe_dyn_event_create,
 179         .show = eprobe_dyn_event_show,
 180         .is_busy = eprobe_dyn_event_is_busy,
 181         .free = eprobe_dyn_event_release,
 182         .match = eprobe_dyn_event_match,
 183 };
 184
 185 static struct trace_eprobe *alloc_event_probe(const char *group,
 186                                               const char *this_event,
 187                                               struct trace_event_call *event,
 188                                               int nargs)
 189 {
 190         struct trace_eprobe *ep;
 191         const char *event_name;
 192         const char *sys_name;
 193         int ret = -ENOMEM;
 194
 195         if (!event)
 196                 return ERR_PTR(-ENODEV);
 197
 198         sys_name = event->class->system;
 199         event_name = trace_event_name(event);
 200
 201         ep = kzalloc(struct_size(ep, tp.args, nargs), GFP_KERNEL);
 202         if (!ep) {
 203                 trace_event_put_ref(event);
 204                 goto error;
 205         }
 206         ep->event = event;
 207         ep->event_name = kstrdup(event_name, GFP_KERNEL);
 208         if (!ep->event_name)
 209                 goto error;
 210         ep->event_system = kstrdup(sys_name, GFP_KERNEL);
 211         if (!ep->event_system)
 212                 goto error;
 213
 214         ret = trace_probe_init(&ep->tp, this_event, group, false);
 215         if (ret < 0)
 216                 goto error;
 217
 218         dyn_event_init(&ep->devent, &eprobe_dyn_event_ops);
 219         return ep;
 220 error:
 221         trace_event_probe_cleanup(ep);
 222         return ERR_PTR(ret);
 223 }
 224
 225 static int trace_eprobe_tp_arg_update(struct trace_eprobe *ep, int i)
 226 {
 227         struct probe_arg *parg = &ep->tp.args[i];
 228         struct ftrace_event_field *field;
 229         struct list_head *head;
 230
 231         head = trace_get_fields(ep->event);
 232         list_for_each_entry(field, head, link) {
 233                 if (!strcmp(parg->code->data, field->name)) {
 234                         kfree(parg->code->data);
 235                         parg->code->data = field;
 236                         return 0;
 237                 }
 238         }
 239         kfree(parg->code->data);
 240         parg->code->data = NULL;
 241         return -ENOENT;
 242 }
 243
 244 static int eprobe_event_define_fields(struct trace_event_call *event_call)
 245 {
 246         struct eprobe_trace_entry_head field;
 247         struct trace_probe *tp;
 248
 249         tp = trace_probe_primary_from_call(event_call);
 250         if (WARN_ON_ONCE(!tp))
 251                 return -ENOENT;
 252
 253         return traceprobe_define_arg_fields(event_call, sizeof(field), tp);
 254 }
 255
 256 static struct trace_event_fields eprobe_fields_array[] = {
 257         { .type = TRACE_FUNCTION_TYPE,
 258           .define_fields = eprobe_event_define_fields },
 259         {}
 260 };
 261
 262 /* Event entry printers */
 263 static enum print_line_t
 264 print_eprobe_event(struct trace_iterator *iter, int flags,
 265                    struct trace_event *event)
 266 {
 267         struct eprobe_trace_entry_head *field;
 268         struct trace_event_call *pevent;
 269         struct trace_event *probed_event;
 270         struct trace_seq *s = &iter->seq;
 271         struct trace_eprobe *ep;
 272         struct trace_probe *tp;
 273         unsigned int type;
 274
 275         field = (struct eprobe_trace_entry_head *)iter->ent;
 276         tp = trace_probe_primary_from_call(
 277                 container_of(event, struct trace_event_call, event));
 278         if (WARN_ON_ONCE(!tp))
 279                 goto out;
 280
 281         ep = container_of(tp, struct trace_eprobe, tp);
 282         type = ep->event->event.type;
 283
 284         trace_seq_printf(s, "%s: (", trace_probe_name(tp));
 285
 286         probed_event = ftrace_find_event(type);
 287         if (probed_event) {
 288                 pevent = container_of(probed_event, struct trace_event_call, event);
 289                 trace_seq_printf(s, "%s.%s", pevent->class->system,
 290                                  trace_event_name(pevent));
 291         } else {
 292                 trace_seq_printf(s, "%u", type);
 293         }
 294
 295         trace_seq_putc(s, ')');
 296
 297         if (print_probe_args(s, tp->args, tp->nr_args,
 298                              (u8 *)&field[1], field) < 0)
 299                 goto out;
 300
 301         trace_seq_putc(s, '\n');
 302  out:
 303         return trace_handle_return(s);
 304 }
 305
 306 static unsigned long get_event_field(struct fetch_insn *code, void *rec)
 307 {
 308         struct ftrace_event_field *field = code->data;
 309         unsigned long val;
 310         void *addr;
 311
 312         addr = rec + field->offset;
 313
 314         switch (field->size) {
 315         case 1:
 316                 if (field->is_signed)
 317                         val = *(char *)addr;
 318                 else
 319                         val = *(unsigned char *)addr;
 320                 break;
 321         case 2:
 322                 if (field->is_signed)
 323                         val = *(short *)addr;
 324                 else
 325                         val = *(unsigned short *)addr;
 326                 break;
 327         case 4:
 328                 if (field->is_signed)
 329                         val = *(int *)addr;
 330                 else
 331                         val = *(unsigned int *)addr;
 332                 break;
 333         default:
 334                 if (field->is_signed)
 335                         val = *(long *)addr;
 336                 else
 337                         val = *(unsigned long *)addr;
 338                 break;
 339         }
 340         return val;
 341 }
 342
 343 static int get_eprobe_size(struct trace_probe *tp, void *rec)
 344 {
 345         struct probe_arg *arg;
 346         int i, len, ret = 0;
 347
 348         for (i = 0; i < tp->nr_args; i++) {
 349                 arg = tp->args + i;
 350                 if (unlikely(arg->dynamic)) {
 351                         unsigned long val;
 352
 353                         val = get_event_field(arg->code, rec);
 354                         len = process_fetch_insn_bottom(arg->code + 1, val, NULL, NULL);
 355                         if (len > 0)
 356                                 ret += len;
 357                 }
 358         }
 359
 360         return ret;
 361 }
 362
 363 /* Kprobe specific fetch functions */
 364
 365 /* Note that we don't verify it, since the code does not come from user space */
 366 static int
 367 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 368                    void *base)
 369 {
 370         unsigned long val;
 371
 372         val = get_event_field(code, rec);
 373         return process_fetch_insn_bottom(code + 1, val, dest, base);
 374 }
 375 NOKPROBE_SYMBOL(process_fetch_insn)
 376
 377 /* Return the length of string -- including null terminal byte */
 378 static nokprobe_inline int
 379 fetch_store_strlen_user(unsigned long addr)
 380 {
 381         const void __user *uaddr =  (__force const void __user *)addr;
 382
 383         return strnlen_user_nofault(uaddr, MAX_STRING_SIZE);
 384 }
 385
 386 /* Return the length of string -- including null terminal byte */
 387 static nokprobe_inline int
 388 fetch_store_strlen(unsigned long addr)
 389 {
 390         int ret, len = 0;
 391         u8 c;
 392
 393 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 394         if (addr < TASK_SIZE)
 395                 return fetch_store_strlen_user(addr);
 396 #endif
 397
 398         do {
 399                 ret = copy_from_kernel_nofault(&c, (u8 *)addr + len, 1);
 400                 len++;
 401         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 402
 403         return (ret < 0) ? ret : len;
 404 }
 405
 406 /*
 407  * Fetch a null-terminated string from user. Caller MUST set *(u32 *)buf
 408  * with max length and relative data location.
 409  */
 410 static nokprobe_inline int
 411 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 412 {
 413         const void __user *uaddr =  (__force const void __user *)addr;
 414         int maxlen = get_loc_len(*(u32 *)dest);
 415         void *__dest;
 416         long ret;
 417
 418         if (unlikely(!maxlen))
 419                 return -ENOMEM;
 420
 421         __dest = get_loc_data(dest, base);
 422
 423         ret = strncpy_from_user_nofault(__dest, uaddr, maxlen);
 424         if (ret >= 0)
 425                 *(u32 *)dest = make_data_loc(ret, __dest - base);
 426
 427         return ret;
 428 }
 429
 430 /*
 431  * Fetch a null-terminated string. Caller MUST set *(u32 *)buf with max
 432  * length and relative data location.
 433  */
 434 static nokprobe_inline int
 435 fetch_store_string(unsigned long addr, void *dest, void *base)
 436 {
 437         int maxlen = get_loc_len(*(u32 *)dest);
 438         void *__dest;
 439         long ret;
 440
 441 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 442         if ((unsigned long)addr < TASK_SIZE)
 443                 return fetch_store_string_user(addr, dest, base);
 444 #endif
 445
 446         if (unlikely(!maxlen))
 447                 return -ENOMEM;
 448
 449         __dest = get_loc_data(dest, base);
 450
 451         /*
 452          * Try to get string again, since the string can be changed while
 453          * probing.
 454          */
 455         ret = strncpy_from_kernel_nofault(__dest, (void *)addr, maxlen);
 456         if (ret >= 0)
 457                 *(u32 *)dest = make_data_loc(ret, __dest - base);
 458
 459         return ret;
 460 }
 461
 462 static nokprobe_inline int
 463 probe_mem_read_user(void *dest, void *src, size_t size)
 464 {
 465         const void __user *uaddr =  (__force const void __user *)src;
 466
 467         return copy_from_user_nofault(dest, uaddr, size);
 468 }
 469
 470 static nokprobe_inline int
 471 probe_mem_read(void *dest, void *src, size_t size)
 472 {
 473 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 474         if ((unsigned long)src < TASK_SIZE)
 475                 return probe_mem_read_user(dest, src, size);
 476 #endif
 477         return copy_from_kernel_nofault(dest, src, size);
 478 }
 479
 480 /* eprobe handler */
 481 static inline void
 482 __eprobe_trace_func(struct eprobe_data *edata, void *rec)
 483 {
 484         struct eprobe_trace_entry_head *entry;
 485         struct trace_event_call *call = trace_probe_event_call(&edata->ep->tp);
 486         struct trace_event_buffer fbuffer;
 487         int dsize;
 488
 489         if (WARN_ON_ONCE(call != edata->file->event_call))
 490                 return;
 491
 492         if (trace_trigger_soft_disabled(edata->file))
 493                 return;
 494
 495         dsize = get_eprobe_size(&edata->ep->tp, rec);
 496
 497         entry = trace_event_buffer_reserve(&fbuffer, edata->file,
 498                                            sizeof(*entry) + edata->ep->tp.size + dsize);
 499
 500         if (!entry)
 501                 return;
 502
 503         entry = fbuffer.entry = ring_buffer_event_data(fbuffer.event);
 504         store_trace_args(&entry[1], &edata->ep->tp, rec, sizeof(*entry), dsize);
 505
 506         trace_event_buffer_commit(&fbuffer);
 507 }
 508
 509 /*
 510  * The event probe implementation uses event triggers to get access to
 511  * the event it is attached to, but is not an actual trigger. The below
 512  * functions are just stubs to fulfill what is needed to use the trigger
 513  * infrastructure.
 514  */
 515 static int eprobe_trigger_init(struct event_trigger_data *data)
 516 {
 517         return 0;
 518 }
 519
 520 static void eprobe_trigger_free(struct event_trigger_data *data)
 521 {
 522
 523 }
 524
 525 static int eprobe_trigger_print(struct seq_file *m,
 526                                 struct event_trigger_data *data)
 527 {
 528         /* Do not print eprobe event triggers */
 529         return 0;
 530 }
 531
 532 static void eprobe_trigger_func(struct event_trigger_data *data,
 533                                 struct trace_buffer *buffer, void *rec,
 534                                 struct ring_buffer_event *rbe)
 535 {
 536         struct eprobe_data *edata = data->private_data;
 537
 538         __eprobe_trace_func(edata, rec);
 539 }
 540
 541 static struct event_trigger_ops eprobe_trigger_ops = {
 542         .trigger                = eprobe_trigger_func,
 543         .print                  = eprobe_trigger_print,
 544         .init                   = eprobe_trigger_init,
 545         .free                   = eprobe_trigger_free,
 546 };
 547
 548 static int eprobe_trigger_cmd_parse(struct event_command *cmd_ops,
 549                                     struct trace_event_file *file,
 550                                     char *glob, char *cmd,
 551                                     char *param_and_filter)
 552 {
 553         return -1;
 554 }
 555
 556 static int eprobe_trigger_reg_func(char *glob,
 557                                    struct event_trigger_data *data,
 558                                    struct trace_event_file *file)
 559 {
 560         return -1;
 561 }
 562
 563 static void eprobe_trigger_unreg_func(char *glob,
 564                                       struct event_trigger_data *data,
 565                                       struct trace_event_file *file)
 566 {
 567
 568 }
 569
 570 static struct event_trigger_ops *eprobe_trigger_get_ops(char *cmd,
 571                                                         char *param)
 572 {
 573         return &eprobe_trigger_ops;
 574 }
 575
 576 static struct event_command event_trigger_cmd = {
 577         .name                   = "eprobe",
 578         .trigger_type           = ETT_EVENT_EPROBE,
 579         .flags                  = EVENT_CMD_FL_NEEDS_REC,
 580         .parse                  = eprobe_trigger_cmd_parse,
 581         .reg                    = eprobe_trigger_reg_func,
 582         .unreg                  = eprobe_trigger_unreg_func,
 583         .unreg_all              = NULL,
 584         .get_trigger_ops        = eprobe_trigger_get_ops,
 585         .set_filter             = NULL,
 586 };
 587
 588 static struct event_trigger_data *
 589 new_eprobe_trigger(struct trace_eprobe *ep, struct trace_event_file *file)
 590 {
 591         struct event_trigger_data *trigger;
 592         struct eprobe_data *edata;
 593
 594         edata = kzalloc(sizeof(*edata), GFP_KERNEL);
 595         trigger = kzalloc(sizeof(*trigger), GFP_KERNEL);
 596         if (!trigger || !edata) {
 597                 kfree(edata);
 598                 kfree(trigger);
 599                 return ERR_PTR(-ENOMEM);
 600         }
 601
 602         trigger->flags = EVENT_TRIGGER_FL_PROBE;
 603         trigger->count = -1;
 604         trigger->ops = &eprobe_trigger_ops;
 605
 606         /*
 607          * EVENT PROBE triggers are not registered as commands with
 608          * register_event_command(), as they are not controlled by the user
 609          * from the trigger file
 610          */
 611         trigger->cmd_ops = &event_trigger_cmd;
 612
 613         INIT_LIST_HEAD(&trigger->list);
 614         RCU_INIT_POINTER(trigger->filter, NULL);
 615
 616         edata->file = file;
 617         edata->ep = ep;
 618         trigger->private_data = edata;
 619
 620         return trigger;
 621 }
 622
 623 static int enable_eprobe(struct trace_eprobe *ep,
 624                          struct trace_event_file *eprobe_file)
 625 {
 626         struct event_trigger_data *trigger;
 627         struct trace_event_file *file;
 628         struct trace_array *tr = eprobe_file->tr;
 629
 630         file = find_event_file(tr, ep->event_system, ep->event_name);
 631         if (!file)
 632                 return -ENOENT;
 633         trigger = new_eprobe_trigger(ep, eprobe_file);
 634         if (IS_ERR(trigger))
 635                 return PTR_ERR(trigger);
 636
 637         list_add_tail_rcu(&trigger->list, &file->triggers);
 638
 639         trace_event_trigger_enable_disable(file, 1);
 640         update_cond_flag(file);
 641
 642         return 0;
 643 }
 644
 645 static struct trace_event_functions eprobe_funcs = {
 646         .trace          = print_eprobe_event
 647 };
 648
 649 static int disable_eprobe(struct trace_eprobe *ep,
 650                           struct trace_array *tr)
 651 {
 652         struct event_trigger_data *trigger = NULL, *iter;
 653         struct trace_event_file *file;
 654         struct eprobe_data *edata;
 655
 656         file = find_event_file(tr, ep->event_system, ep->event_name);
 657         if (!file)
 658                 return -ENOENT;
 659
 660         list_for_each_entry(iter, &file->triggers, list) {
 661                 if (!(iter->flags & EVENT_TRIGGER_FL_PROBE))
 662                         continue;
 663                 edata = iter->private_data;
 664                 if (edata->ep == ep) {
 665                         trigger = iter;
 666                         break;
 667                 }
 668         }
 669         if (!trigger)
 670                 return -ENODEV;
 671
 672         list_del_rcu(&trigger->list);
 673
 674         trace_event_trigger_enable_disable(file, 0);
 675         update_cond_flag(file);
 676
 677         /* Make sure nothing is using the edata or trigger */
 678         tracepoint_synchronize_unregister();
 679
 680         kfree(edata);
 681         kfree(trigger);
 682
 683         return 0;
 684 }
 685
 686 static int enable_trace_eprobe(struct trace_event_call *call,
 687                                struct trace_event_file *file)
 688 {
 689         struct trace_probe *pos, *tp;
 690         struct trace_eprobe *ep;
 691         bool enabled;
 692         int ret = 0;
 693
 694         tp = trace_probe_primary_from_call(call);
 695         if (WARN_ON_ONCE(!tp))
 696                 return -ENODEV;
 697         enabled = trace_probe_is_enabled(tp);
 698
 699         /* This also changes "enabled" state */
 700         if (file) {
 701                 ret = trace_probe_add_file(tp, file);
 702                 if (ret)
 703                         return ret;
 704         } else
 705                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
 706
 707         if (enabled)
 708                 return 0;
 709
 710         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 711                 ep = container_of(pos, struct trace_eprobe, tp);
 712                 ret = enable_eprobe(ep, file);
 713                 if (ret)
 714                         break;
 715                 enabled = true;
 716         }
 717
 718         if (ret) {
 719                 /* Failed to enable one of them. Roll back all */
 720                 if (enabled)
 721                         disable_eprobe(ep, file->tr);
 722                 if (file)
 723                         trace_probe_remove_file(tp, file);
 724                 else
 725                         trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 726         }
 727
 728         return ret;
 729 }
 730
 731 static int disable_trace_eprobe(struct trace_event_call *call,
 732                                 struct trace_event_file *file)
 733 {
 734         struct trace_probe *pos, *tp;
 735         struct trace_eprobe *ep;
 736
 737         tp = trace_probe_primary_from_call(call);
 738         if (WARN_ON_ONCE(!tp))
 739                 return -ENODEV;
 740
 741         if (file) {
 742                 if (!trace_probe_get_file_link(tp, file))
 743                         return -ENOENT;
 744                 if (!trace_probe_has_single_file(tp))
 745                         goto out;
 746                 trace_probe_clear_flag(tp, TP_FLAG_TRACE);
 747         } else
 748                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
 749
 750         if (!trace_probe_is_enabled(tp)) {
 751                 list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 752                         ep = container_of(pos, struct trace_eprobe, tp);
 753                         disable_eprobe(ep, file->tr);
 754                 }
 755         }
 756
 757  out:
 758         if (file)
 759                 /*
 760                  * Synchronization is done in below function. For perf event,
 761                  * file == NULL and perf_trace_event_unreg() calls
 762                  * tracepoint_synchronize_unregister() to ensure synchronize
 763                  * event. We don't need to care about it.
 764                  */
 765                 trace_probe_remove_file(tp, file);
 766
 767         return 0;
 768 }
 769
 770 static int eprobe_register(struct trace_event_call *event,
 771                            enum trace_reg type, void *data)
 772 {
 773         struct trace_event_file *file = data;
 774
 775         switch (type) {
 776         case TRACE_REG_REGISTER:
 777                 return enable_trace_eprobe(event, file);
 778         case TRACE_REG_UNREGISTER:
 779                 return disable_trace_eprobe(event, file);
 780 #ifdef CONFIG_PERF_EVENTS
 781         case TRACE_REG_PERF_REGISTER:
 782         case TRACE_REG_PERF_UNREGISTER:
 783         case TRACE_REG_PERF_OPEN:
 784         case TRACE_REG_PERF_CLOSE:
 785         case TRACE_REG_PERF_ADD:
 786         case TRACE_REG_PERF_DEL:
 787                 return 0;
 788 #endif
 789         }
 790         return 0;
 791 }
 792
 793 static inline void init_trace_eprobe_call(struct trace_eprobe *ep)
 794 {
 795         struct trace_event_call *call = trace_probe_event_call(&ep->tp);
 796
 797         call->flags = TRACE_EVENT_FL_EPROBE;
 798         call->event.funcs = &eprobe_funcs;
 799         call->class->fields_array = eprobe_fields_array;
 800         call->class->reg = eprobe_register;
 801 }
 802
 803 static struct trace_event_call *
 804 find_and_get_event(const char *system, const char *event_name)
 805 {
 806         struct trace_event_call *tp_event;
 807         const char *name;
 808
 809         list_for_each_entry(tp_event, &ftrace_events, list) {
 810                 /* Skip other probes and ftrace events */
 811                 if (tp_event->flags &
 812                     (TRACE_EVENT_FL_IGNORE_ENABLE |
 813                      TRACE_EVENT_FL_KPROBE |
 814                      TRACE_EVENT_FL_UPROBE |
 815                      TRACE_EVENT_FL_EPROBE))
 816                         continue;
 817                 if (!tp_event->class->system ||
 818                     strcmp(system, tp_event->class->system))
 819                         continue;
 820                 name = trace_event_name(tp_event);
 821                 if (!name || strcmp(event_name, name))
 822                         continue;
 823                 if (!trace_event_try_get_ref(tp_event)) {
 824                         return NULL;
 825                         break;
 826                 }
 827                 return tp_event;
 828                 break;
 829         }
 830         return NULL;
 831 }
 832
 833 static int trace_eprobe_tp_update_arg(struct trace_eprobe *ep, const char *argv[], int i)
 834 {
 835         unsigned int flags = TPARG_FL_KERNEL | TPARG_FL_TPOINT;
 836         int ret;
 837
 838         ret = traceprobe_parse_probe_arg(&ep->tp, i, argv[i], flags);
 839         if (ret)
 840                 return ret;
 841
 842         if (ep->tp.args[i].code->op == FETCH_OP_TP_ARG) {
 843                 ret = trace_eprobe_tp_arg_update(ep, i);
 844                 if (ret)
 845                         trace_probe_log_err(0, BAD_ATTACH_ARG);
 846         }
 847
 848         return ret;
 849 }
 850
 851 static int __trace_eprobe_create(int argc, const char *argv[])
 852 {
 853         /*
 854          * Argument syntax:
 855          *      e[:[GRP/][ENAME]] SYSTEM.EVENT [FETCHARGS]
 856          * Fetch args:
 857          *  <name>=$<field>[:TYPE]
 858          */
 859         const char *event = NULL, *group = EPROBE_EVENT_SYSTEM;
 860         const char *sys_event = NULL, *sys_name = NULL;
 861         struct trace_event_call *event_call;
 862         struct trace_eprobe *ep = NULL;
 863         char buf1[MAX_EVENT_NAME_LEN];
 864         char buf2[MAX_EVENT_NAME_LEN];
 865         char gbuf[MAX_EVENT_NAME_LEN];
 866         int ret = 0;
 867         int i;
 868
 869         if (argc < 2 || argv[0][0] != 'e')
 870                 return -ECANCELED;
 871
 872         trace_probe_log_init("event_probe", argc, argv);
 873
 874         event = strchr(&argv[0][1], ':');
 875         if (event) {
 876                 event++;
 877                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
 878                                                   event - argv[0]);
 879                 if (ret)
 880                         goto parse_error;
 881         }
 882
 883         trace_probe_log_set_index(1);
 884         sys_event = argv[1];
 885         ret = traceprobe_parse_event_name(&sys_event, &sys_name, buf2, 0);
 886         if (!sys_event || !sys_name) {
 887                 trace_probe_log_err(0, NO_EVENT_INFO);
 888                 goto parse_error;
 889         }
 890
 891         if (!event) {
 892                 strscpy(buf1, argv[1], MAX_EVENT_NAME_LEN);
 893                 sanitize_event_name(buf1);
 894                 event = buf1;
 895         }
 896
 897         mutex_lock(&event_mutex);
 898         event_call = find_and_get_event(sys_name, sys_event);
 899         ep = alloc_event_probe(group, event, event_call, argc - 2);
 900         mutex_unlock(&event_mutex);
 901
 902         if (IS_ERR(ep)) {
 903                 ret = PTR_ERR(ep);
 904                 if (ret == -ENODEV)
 905                         trace_probe_log_err(0, BAD_ATTACH_EVENT);
 906                 /* This must return -ENOMEM or missing event, else there is a bug */
 907                 WARN_ON_ONCE(ret != -ENOMEM && ret != -ENODEV);
 908                 ep = NULL;
 909                 goto error;
 910         }
 911
 912         argc -= 2; argv += 2;
 913         /* parse arguments */
 914         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 915                 trace_probe_log_set_index(i + 2);
 916                 ret = trace_eprobe_tp_update_arg(ep, argv, i);
 917                 if (ret)
 918                         goto error;
 919         }
 920         ret = traceprobe_set_print_fmt(&ep->tp, PROBE_PRINT_EVENT);
 921         if (ret < 0)
 922                 goto error;
 923         init_trace_eprobe_call(ep);
 924         mutex_lock(&event_mutex);
 925         ret = trace_probe_register_event_call(&ep->tp);
 926         if (ret) {
 927                 if (ret == -EEXIST) {
 928                         trace_probe_log_set_index(0);
 929                         trace_probe_log_err(0, EVENT_EXIST);
 930                 }
 931                 mutex_unlock(&event_mutex);
 932                 goto error;
 933         }
 934         ret = dyn_event_add(&ep->devent, &ep->tp.event->call);
 935         mutex_unlock(&event_mutex);
 936         return ret;
 937 parse_error:
 938         ret = -EINVAL;
 939 error:
 940         trace_event_probe_cleanup(ep);
 941         return ret;
 942 }
 943
 944 /*
 945  * Register dynevent at core_initcall. This allows kernel to setup eprobe
 946  * events in postcore_initcall without tracefs.
 947  */
 948 static __init int trace_events_eprobe_init_early(void)
 949 {
 950         int err = 0;
 951
 952         err = dyn_event_register(&eprobe_dyn_event_ops);
 953         if (err)
 954                 pr_warn("Could not register eprobe_dyn_event_ops\n");
 955
 956         return err;
 957 }
 958 core_initcall(trace_events_eprobe_init_early);