kernel/trace/trace_uprobe.c

   1 /*
   2  * uprobes-based tracing events
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  16  *
  17  * Copyright (C) IBM Corporation, 2010-2012
  18  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/module.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/uprobes.h>
  24 #include <linux/namei.h>
  25 #include <linux/string.h>
  26
  27 #include "trace_probe.h"
  28
  29 #define UPROBE_EVENT_SYSTEM     "uprobes"
  30
  31 struct uprobe_trace_entry_head {
  32         struct trace_entry      ent;
  33         unsigned long           vaddr[];
  34 };
  35
  36 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  37         (sizeof(struct uprobe_trace_entry_head) +       \
  38          sizeof(unsigned long) * (is_return ? 2 : 1))
  39
  40 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  41         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  42
  43 struct trace_uprobe_filter {
  44         rwlock_t                rwlock;
  45         int                     nr_systemwide;
  46         struct list_head        perf_events;
  47 };
  48
  49 /*
  50  * uprobe event core functions
  51  */
  52 struct trace_uprobe {
  53         struct list_head                list;
  54         struct ftrace_event_class       class;
  55         struct ftrace_event_call        call;
  56         struct trace_uprobe_filter      filter;
  57         struct uprobe_consumer          consumer;
  58         struct inode                    *inode;
  59         char                            *filename;
  60         unsigned long                   offset;
  61         unsigned long                   nhit;
  62         unsigned int                    flags;  /* For TP_FLAG_* */
  63         ssize_t                         size;   /* trace entry size */
  64         unsigned int                    nr_args;
  65         struct probe_arg                args[];
  66 };
  67
  68 #define SIZEOF_TRACE_UPROBE(n)                  \
  69         (offsetof(struct trace_uprobe, args) +  \
  70         (sizeof(struct probe_arg) * (n)))
  71
  72 static int register_uprobe_event(struct trace_uprobe *tu);
  73 static void unregister_uprobe_event(struct trace_uprobe *tu);
  74
  75 static DEFINE_MUTEX(uprobe_lock);
  76 static LIST_HEAD(uprobe_list);
  77
  78 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  79
  80 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
  81 {
  82         rwlock_init(&filter->rwlock);
  83         filter->nr_systemwide = 0;
  84         INIT_LIST_HEAD(&filter->perf_events);
  85 }
  86
  87 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
  88 {
  89         return !filter->nr_systemwide && list_empty(&filter->perf_events);
  90 }
  91
  92 /*
  93  * Allocate new trace_uprobe and initialize it (including uprobes).
  94  */
  95 static struct trace_uprobe *
  96 alloc_trace_uprobe(const char *group, const char *event, int nargs)
  97 {
  98         struct trace_uprobe *tu;
  99
 100         if (!event || !is_good_name(event))
 101                 return ERR_PTR(-EINVAL);
 102
 103         if (!group || !is_good_name(group))
 104                 return ERR_PTR(-EINVAL);
 105
 106         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 107         if (!tu)
 108                 return ERR_PTR(-ENOMEM);
 109
 110         tu->call.class = &tu->class;
 111         tu->call.name = kstrdup(event, GFP_KERNEL);
 112         if (!tu->call.name)
 113                 goto error;
 114
 115         tu->class.system = kstrdup(group, GFP_KERNEL);
 116         if (!tu->class.system)
 117                 goto error;
 118
 119         INIT_LIST_HEAD(&tu->list);
 120         tu->consumer.handler = uprobe_dispatcher;
 121         init_trace_uprobe_filter(&tu->filter);
 122         return tu;
 123
 124 error:
 125         kfree(tu->call.name);
 126         kfree(tu);
 127
 128         return ERR_PTR(-ENOMEM);
 129 }
 130
 131 static void free_trace_uprobe(struct trace_uprobe *tu)
 132 {
 133         int i;
 134
 135         for (i = 0; i < tu->nr_args; i++)
 136                 traceprobe_free_probe_arg(&tu->args[i]);
 137
 138         iput(tu->inode);
 139         kfree(tu->call.class->system);
 140         kfree(tu->call.name);
 141         kfree(tu->filename);
 142         kfree(tu);
 143 }
 144
 145 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 146 {
 147         struct trace_uprobe *tu;
 148
 149         list_for_each_entry(tu, &uprobe_list, list)
 150                 if (strcmp(tu->call.name, event) == 0 &&
 151                     strcmp(tu->call.class->system, group) == 0)
 152                         return tu;
 153
 154         return NULL;
 155 }
 156
 157 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 158 static void unregister_trace_uprobe(struct trace_uprobe *tu)
 159 {
 160         list_del(&tu->list);
 161         unregister_uprobe_event(tu);
 162         free_trace_uprobe(tu);
 163 }
 164
 165 /* Register a trace_uprobe and probe_event */
 166 static int register_trace_uprobe(struct trace_uprobe *tu)
 167 {
 168         struct trace_uprobe *old_tp;
 169         int ret;
 170
 171         mutex_lock(&uprobe_lock);
 172
 173         /* register as an event */
 174         old_tp = find_probe_event(tu->call.name, tu->call.class->system);
 175         if (old_tp)
 176                 /* delete old event */
 177                 unregister_trace_uprobe(old_tp);
 178
 179         ret = register_uprobe_event(tu);
 180         if (ret) {
 181                 pr_warning("Failed to register probe event(%d)\n", ret);
 182                 goto end;
 183         }
 184
 185         list_add_tail(&tu->list, &uprobe_list);
 186
 187 end:
 188         mutex_unlock(&uprobe_lock);
 189
 190         return ret;
 191 }
 192
 193 /*
 194  * Argument syntax:
 195  *  - Add uprobe: p[:[GRP/]EVENT] PATH:SYMBOL[+offs] [FETCHARGS]
 196  *
 197  *  - Remove uprobe: -:[GRP/]EVENT
 198  */
 199 static int create_trace_uprobe(int argc, char **argv)
 200 {
 201         struct trace_uprobe *tu;
 202         struct inode *inode;
 203         char *arg, *event, *group, *filename;
 204         char buf[MAX_EVENT_NAME_LEN];
 205         struct path path;
 206         unsigned long offset;
 207         bool is_delete;
 208         int i, ret;
 209
 210         inode = NULL;
 211         ret = 0;
 212         is_delete = false;
 213         event = NULL;
 214         group = NULL;
 215
 216         /* argc must be >= 1 */
 217         if (argv[0][0] == '-')
 218                 is_delete = true;
 219         else if (argv[0][0] != 'p') {
 220                 pr_info("Probe definition must be started with 'p' or '-'.\n");
 221                 return -EINVAL;
 222         }
 223
 224         if (argv[0][1] == ':') {
 225                 event = &argv[0][2];
 226                 arg = strchr(event, '/');
 227
 228                 if (arg) {
 229                         group = event;
 230                         event = arg + 1;
 231                         event[-1] = '\0';
 232
 233                         if (strlen(group) == 0) {
 234                                 pr_info("Group name is not specified\n");
 235                                 return -EINVAL;
 236                         }
 237                 }
 238                 if (strlen(event) == 0) {
 239                         pr_info("Event name is not specified\n");
 240                         return -EINVAL;
 241                 }
 242         }
 243         if (!group)
 244                 group = UPROBE_EVENT_SYSTEM;
 245
 246         if (is_delete) {
 247                 if (!event) {
 248                         pr_info("Delete command needs an event name.\n");
 249                         return -EINVAL;
 250                 }
 251                 mutex_lock(&uprobe_lock);
 252                 tu = find_probe_event(event, group);
 253
 254                 if (!tu) {
 255                         mutex_unlock(&uprobe_lock);
 256                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 257                         return -ENOENT;
 258                 }
 259                 /* delete an event */
 260                 unregister_trace_uprobe(tu);
 261                 mutex_unlock(&uprobe_lock);
 262                 return 0;
 263         }
 264
 265         if (argc < 2) {
 266                 pr_info("Probe point is not specified.\n");
 267                 return -EINVAL;
 268         }
 269         if (isdigit(argv[1][0])) {
 270                 pr_info("probe point must be have a filename.\n");
 271                 return -EINVAL;
 272         }
 273         arg = strchr(argv[1], ':');
 274         if (!arg)
 275                 goto fail_address_parse;
 276
 277         *arg++ = '\0';
 278         filename = argv[1];
 279         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 280         if (ret)
 281                 goto fail_address_parse;
 282
 283         inode = igrab(path.dentry->d_inode);
 284         path_put(&path);
 285
 286         if (!inode || !S_ISREG(inode->i_mode)) {
 287                 ret = -EINVAL;
 288                 goto fail_address_parse;
 289         }
 290
 291         ret = kstrtoul(arg, 0, &offset);
 292         if (ret)
 293                 goto fail_address_parse;
 294
 295         argc -= 2;
 296         argv += 2;
 297
 298         /* setup a probe */
 299         if (!event) {
 300                 char *tail;
 301                 char *ptr;
 302
 303                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 304                 if (!tail) {
 305                         ret = -ENOMEM;
 306                         goto fail_address_parse;
 307                 }
 308
 309                 ptr = strpbrk(tail, ".-_");
 310                 if (ptr)
 311                         *ptr = '\0';
 312
 313                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 314                 event = buf;
 315                 kfree(tail);
 316         }
 317
 318         tu = alloc_trace_uprobe(group, event, argc);
 319         if (IS_ERR(tu)) {
 320                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 321                 ret = PTR_ERR(tu);
 322                 goto fail_address_parse;
 323         }
 324         tu->offset = offset;
 325         tu->inode = inode;
 326         tu->filename = kstrdup(filename, GFP_KERNEL);
 327
 328         if (!tu->filename) {
 329                 pr_info("Failed to allocate filename.\n");
 330                 ret = -ENOMEM;
 331                 goto error;
 332         }
 333
 334         /* parse arguments */
 335         ret = 0;
 336         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 337                 /* Increment count for freeing args in error case */
 338                 tu->nr_args++;
 339
 340                 /* Parse argument name */
 341                 arg = strchr(argv[i], '=');
 342                 if (arg) {
 343                         *arg++ = '\0';
 344                         tu->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 345                 } else {
 346                         arg = argv[i];
 347                         /* If argument name is omitted, set "argN" */
 348                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 349                         tu->args[i].name = kstrdup(buf, GFP_KERNEL);
 350                 }
 351
 352                 if (!tu->args[i].name) {
 353                         pr_info("Failed to allocate argument[%d] name.\n", i);
 354                         ret = -ENOMEM;
 355                         goto error;
 356                 }
 357
 358                 if (!is_good_name(tu->args[i].name)) {
 359                         pr_info("Invalid argument[%d] name: %s\n", i, tu->args[i].name);
 360                         ret = -EINVAL;
 361                         goto error;
 362                 }
 363
 364                 if (traceprobe_conflict_field_name(tu->args[i].name, tu->args, i)) {
 365                         pr_info("Argument[%d] name '%s' conflicts with "
 366                                 "another field.\n", i, argv[i]);
 367                         ret = -EINVAL;
 368                         goto error;
 369                 }
 370
 371                 /* Parse fetch argument */
 372                 ret = traceprobe_parse_probe_arg(arg, &tu->size, &tu->args[i], false, false);
 373                 if (ret) {
 374                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 375                         goto error;
 376                 }
 377         }
 378
 379         ret = register_trace_uprobe(tu);
 380         if (ret)
 381                 goto error;
 382         return 0;
 383
 384 error:
 385         free_trace_uprobe(tu);
 386         return ret;
 387
 388 fail_address_parse:
 389         if (inode)
 390                 iput(inode);
 391
 392         pr_info("Failed to parse address or file.\n");
 393
 394         return ret;
 395 }
 396
 397 static void cleanup_all_probes(void)
 398 {
 399         struct trace_uprobe *tu;
 400
 401         mutex_lock(&uprobe_lock);
 402         while (!list_empty(&uprobe_list)) {
 403                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 404                 unregister_trace_uprobe(tu);
 405         }
 406         mutex_unlock(&uprobe_lock);
 407 }
 408
 409 /* Probes listing interfaces */
 410 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 411 {
 412         mutex_lock(&uprobe_lock);
 413         return seq_list_start(&uprobe_list, *pos);
 414 }
 415
 416 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 417 {
 418         return seq_list_next(v, &uprobe_list, pos);
 419 }
 420
 421 static void probes_seq_stop(struct seq_file *m, void *v)
 422 {
 423         mutex_unlock(&uprobe_lock);
 424 }
 425
 426 static int probes_seq_show(struct seq_file *m, void *v)
 427 {
 428         struct trace_uprobe *tu = v;
 429         int i;
 430
 431         seq_printf(m, "p:%s/%s", tu->call.class->system, tu->call.name);
 432         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 433
 434         for (i = 0; i < tu->nr_args; i++)
 435                 seq_printf(m, " %s=%s", tu->args[i].name, tu->args[i].comm);
 436
 437         seq_printf(m, "\n");
 438         return 0;
 439 }
 440
 441 static const struct seq_operations probes_seq_op = {
 442         .start  = probes_seq_start,
 443         .next   = probes_seq_next,
 444         .stop   = probes_seq_stop,
 445         .show   = probes_seq_show
 446 };
 447
 448 static int probes_open(struct inode *inode, struct file *file)
 449 {
 450         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
 451                 cleanup_all_probes();
 452
 453         return seq_open(file, &probes_seq_op);
 454 }
 455
 456 static ssize_t probes_write(struct file *file, const char __user *buffer,
 457                             size_t count, loff_t *ppos)
 458 {
 459         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 460 }
 461
 462 static const struct file_operations uprobe_events_ops = {
 463         .owner          = THIS_MODULE,
 464         .open           = probes_open,
 465         .read           = seq_read,
 466         .llseek         = seq_lseek,
 467         .release        = seq_release,
 468         .write          = probes_write,
 469 };
 470
 471 /* Probes profiling interfaces */
 472 static int probes_profile_seq_show(struct seq_file *m, void *v)
 473 {
 474         struct trace_uprobe *tu = v;
 475
 476         seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->call.name, tu->nhit);
 477         return 0;
 478 }
 479
 480 static const struct seq_operations profile_seq_op = {
 481         .start  = probes_seq_start,
 482         .next   = probes_seq_next,
 483         .stop   = probes_seq_stop,
 484         .show   = probes_profile_seq_show
 485 };
 486
 487 static int profile_open(struct inode *inode, struct file *file)
 488 {
 489         return seq_open(file, &profile_seq_op);
 490 }
 491
 492 static const struct file_operations uprobe_profile_ops = {
 493         .owner          = THIS_MODULE,
 494         .open           = profile_open,
 495         .read           = seq_read,
 496         .llseek         = seq_lseek,
 497         .release        = seq_release,
 498 };
 499
 500 /* uprobe handler */
 501 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
 502 {
 503         struct uprobe_trace_entry_head *entry;
 504         struct ring_buffer_event *event;
 505         struct ring_buffer *buffer;
 506         void *data;
 507         int size, i;
 508         struct ftrace_event_call *call = &tu->call;
 509
 510         size = SIZEOF_TRACE_ENTRY(false) + tu->size;
 511         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 512                                                   size, 0, 0);
 513         if (!event)
 514                 return 0;
 515
 516         entry = ring_buffer_event_data(event);
 517         entry->vaddr[0] = instruction_pointer(regs);
 518         data = DATAOF_TRACE_ENTRY(entry, false);
 519         for (i = 0; i < tu->nr_args; i++)
 520                 call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
 521
 522         if (!filter_current_check_discard(buffer, call, entry, event))
 523                 trace_buffer_unlock_commit(buffer, event, 0, 0);
 524
 525         return 0;
 526 }
 527
 528 /* Event entry printers */
 529 static enum print_line_t
 530 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 531 {
 532         struct uprobe_trace_entry_head *entry;
 533         struct trace_seq *s = &iter->seq;
 534         struct trace_uprobe *tu;
 535         u8 *data;
 536         int i;
 537
 538         entry = (struct uprobe_trace_entry_head *)iter->ent;
 539         tu = container_of(event, struct trace_uprobe, call.event);
 540
 541         if (!trace_seq_printf(s, "%s: (0x%lx)", tu->call.name, entry->vaddr[0]))
 542                 goto partial;
 543
 544         data = DATAOF_TRACE_ENTRY(entry, false);
 545         for (i = 0; i < tu->nr_args; i++) {
 546                 if (!tu->args[i].type->print(s, tu->args[i].name,
 547                                              data + tu->args[i].offset, entry))
 548                         goto partial;
 549         }
 550
 551         if (trace_seq_puts(s, "\n"))
 552                 return TRACE_TYPE_HANDLED;
 553
 554 partial:
 555         return TRACE_TYPE_PARTIAL_LINE;
 556 }
 557
 558 static inline bool is_trace_uprobe_enabled(struct trace_uprobe *tu)
 559 {
 560         return tu->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE);
 561 }
 562
 563 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 564                                 enum uprobe_filter_ctx ctx,
 565                                 struct mm_struct *mm);
 566
 567 static int
 568 probe_event_enable(struct trace_uprobe *tu, int flag, filter_func_t filter)
 569 {
 570         int ret = 0;
 571
 572         if (is_trace_uprobe_enabled(tu))
 573                 return -EINTR;
 574
 575         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 576
 577         tu->flags |= flag;
 578         tu->consumer.filter = filter;
 579         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 580         if (ret)
 581                 tu->flags &= ~flag;
 582
 583         return ret;
 584 }
 585
 586 static void probe_event_disable(struct trace_uprobe *tu, int flag)
 587 {
 588         if (!is_trace_uprobe_enabled(tu))
 589                 return;
 590
 591         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 592
 593         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 594         tu->flags &= ~flag;
 595 }
 596
 597 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 598 {
 599         int ret, i, size;
 600         struct uprobe_trace_entry_head field;
 601         struct trace_uprobe *tu = event_call->data;
 602
 603         DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 604         size = SIZEOF_TRACE_ENTRY(false);
 605         /* Set argument names as fields */
 606         for (i = 0; i < tu->nr_args; i++) {
 607                 ret = trace_define_field(event_call, tu->args[i].type->fmttype,
 608                                          tu->args[i].name,
 609                                          size + tu->args[i].offset,
 610                                          tu->args[i].type->size,
 611                                          tu->args[i].type->is_signed,
 612                                          FILTER_OTHER);
 613
 614                 if (ret)
 615                         return ret;
 616         }
 617         return 0;
 618 }
 619
 620 #define LEN_OR_ZERO             (len ? len - pos : 0)
 621 static int __set_print_fmt(struct trace_uprobe *tu, char *buf, int len)
 622 {
 623         const char *fmt, *arg;
 624         int i;
 625         int pos = 0;
 626
 627         fmt = "(%lx)";
 628         arg = "REC->" FIELD_STRING_IP;
 629
 630         /* When len=0, we just calculate the needed length */
 631
 632         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 633
 634         for (i = 0; i < tu->nr_args; i++) {
 635                 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
 636                                 tu->args[i].name, tu->args[i].type->fmt);
 637         }
 638
 639         pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 640
 641         for (i = 0; i < tu->nr_args; i++) {
 642                 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
 643                                 tu->args[i].name);
 644         }
 645
 646         return pos;     /* return the length of print_fmt */
 647 }
 648 #undef LEN_OR_ZERO
 649
 650 static int set_print_fmt(struct trace_uprobe *tu)
 651 {
 652         char *print_fmt;
 653         int len;
 654
 655         /* First: called with 0 length to calculate the needed length */
 656         len = __set_print_fmt(tu, NULL, 0);
 657         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 658         if (!print_fmt)
 659                 return -ENOMEM;
 660
 661         /* Second: actually write the @print_fmt */
 662         __set_print_fmt(tu, print_fmt, len + 1);
 663         tu->call.print_fmt = print_fmt;
 664
 665         return 0;
 666 }
 667
 668 #ifdef CONFIG_PERF_EVENTS
 669 static bool
 670 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 671 {
 672         struct perf_event *event;
 673
 674         if (filter->nr_systemwide)
 675                 return true;
 676
 677         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
 678                 if (event->hw.tp_target->mm == mm)
 679                         return true;
 680         }
 681
 682         return false;
 683 }
 684
 685 static inline bool
 686 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 687 {
 688         return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 689 }
 690
 691 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 692 {
 693         bool done;
 694
 695         write_lock(&tu->filter.rwlock);
 696         if (event->hw.tp_target) {
 697                 /*
 698                  * event->parent != NULL means copy_process(), we can avoid
 699                  * uprobe_apply(). current->mm must be probed and we can rely
 700                  * on dup_mmap() which preserves the already installed bp's.
 701                  *
 702                  * attr.enable_on_exec means that exec/mmap will install the
 703                  * breakpoints we need.
 704                  */
 705                 done = tu->filter.nr_systemwide ||
 706                         event->parent || event->attr.enable_on_exec ||
 707                         uprobe_filter_event(tu, event);
 708                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
 709         } else {
 710                 done = tu->filter.nr_systemwide;
 711                 tu->filter.nr_systemwide++;
 712         }
 713         write_unlock(&tu->filter.rwlock);
 714
 715         if (!done)
 716                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
 717
 718         return 0;
 719 }
 720
 721 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 722 {
 723         bool done;
 724
 725         write_lock(&tu->filter.rwlock);
 726         if (event->hw.tp_target) {
 727                 list_del(&event->hw.tp_list);
 728                 done = tu->filter.nr_systemwide ||
 729                         (event->hw.tp_target->flags & PF_EXITING) ||
 730                         uprobe_filter_event(tu, event);
 731         } else {
 732                 tu->filter.nr_systemwide--;
 733                 done = tu->filter.nr_systemwide;
 734         }
 735         write_unlock(&tu->filter.rwlock);
 736
 737         if (!done)
 738                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 739
 740         return 0;
 741 }
 742
 743 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 744                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
 745 {
 746         struct trace_uprobe *tu;
 747         int ret;
 748
 749         tu = container_of(uc, struct trace_uprobe, consumer);
 750         read_lock(&tu->filter.rwlock);
 751         ret = __uprobe_perf_filter(&tu->filter, mm);
 752         read_unlock(&tu->filter.rwlock);
 753
 754         return ret;
 755 }
 756
 757 /* uprobe profile handler */
 758 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
 759 {
 760         struct ftrace_event_call *call = &tu->call;
 761         struct uprobe_trace_entry_head *entry;
 762         struct hlist_head *head;
 763         unsigned long ip;
 764         void *data;
 765         int size, rctx, i;
 766
 767         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
 768                 return UPROBE_HANDLER_REMOVE;
 769
 770         size = SIZEOF_TRACE_ENTRY(false);
 771         size = ALIGN(size + tu->size + sizeof(u32), sizeof(u64)) - sizeof(u32);
 772         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
 773                 return 0;
 774
 775         preempt_disable();
 776         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 777         if (!entry)
 778                 goto out;
 779
 780         ip = instruction_pointer(regs);
 781         entry->vaddr[0] = ip;
 782         data = DATAOF_TRACE_ENTRY(entry, false);
 783         for (i = 0; i < tu->nr_args; i++)
 784                 call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
 785
 786         head = this_cpu_ptr(call->perf_events);
 787         perf_trace_buf_submit(entry, size, rctx, ip, 1, regs, head, NULL);
 788  out:
 789         preempt_enable();
 790         return 0;
 791 }
 792 #endif  /* CONFIG_PERF_EVENTS */
 793
 794 static
 795 int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
 796 {
 797         struct trace_uprobe *tu = event->data;
 798
 799         switch (type) {
 800         case TRACE_REG_REGISTER:
 801                 return probe_event_enable(tu, TP_FLAG_TRACE, NULL);
 802
 803         case TRACE_REG_UNREGISTER:
 804                 probe_event_disable(tu, TP_FLAG_TRACE);
 805                 return 0;
 806
 807 #ifdef CONFIG_PERF_EVENTS
 808         case TRACE_REG_PERF_REGISTER:
 809                 return probe_event_enable(tu, TP_FLAG_PROFILE, uprobe_perf_filter);
 810
 811         case TRACE_REG_PERF_UNREGISTER:
 812                 probe_event_disable(tu, TP_FLAG_PROFILE);
 813                 return 0;
 814
 815         case TRACE_REG_PERF_OPEN:
 816                 return uprobe_perf_open(tu, data);
 817
 818         case TRACE_REG_PERF_CLOSE:
 819                 return uprobe_perf_close(tu, data);
 820
 821 #endif
 822         default:
 823                 return 0;
 824         }
 825         return 0;
 826 }
 827
 828 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
 829 {
 830         struct trace_uprobe *tu;
 831         int ret = 0;
 832
 833         tu = container_of(con, struct trace_uprobe, consumer);
 834         tu->nhit++;
 835
 836         if (tu->flags & TP_FLAG_TRACE)
 837                 ret |= uprobe_trace_func(tu, regs);
 838
 839 #ifdef CONFIG_PERF_EVENTS
 840         if (tu->flags & TP_FLAG_PROFILE)
 841                 ret |= uprobe_perf_func(tu, regs);
 842 #endif
 843         return ret;
 844 }
 845
 846 static struct trace_event_functions uprobe_funcs = {
 847         .trace          = print_uprobe_event
 848 };
 849
 850 static int register_uprobe_event(struct trace_uprobe *tu)
 851 {
 852         struct ftrace_event_call *call = &tu->call;
 853         int ret;
 854
 855         /* Initialize ftrace_event_call */
 856         INIT_LIST_HEAD(&call->class->fields);
 857         call->event.funcs = &uprobe_funcs;
 858         call->class->define_fields = uprobe_event_define_fields;
 859
 860         if (set_print_fmt(tu) < 0)
 861                 return -ENOMEM;
 862
 863         ret = register_ftrace_event(&call->event);
 864         if (!ret) {
 865                 kfree(call->print_fmt);
 866                 return -ENODEV;
 867         }
 868         call->flags = 0;
 869         call->class->reg = trace_uprobe_register;
 870         call->data = tu;
 871         ret = trace_add_event_call(call);
 872
 873         if (ret) {
 874                 pr_info("Failed to register uprobe event: %s\n", call->name);
 875                 kfree(call->print_fmt);
 876                 unregister_ftrace_event(&call->event);
 877         }
 878
 879         return ret;
 880 }
 881
 882 static void unregister_uprobe_event(struct trace_uprobe *tu)
 883 {
 884         /* tu->event is unregistered in trace_remove_event_call() */
 885         trace_remove_event_call(&tu->call);
 886         kfree(tu->call.print_fmt);
 887         tu->call.print_fmt = NULL;
 888 }
 889
 890 /* Make a trace interface for controling probe points */
 891 static __init int init_uprobe_trace(void)
 892 {
 893         struct dentry *d_tracer;
 894
 895         d_tracer = tracing_init_dentry();
 896         if (!d_tracer)
 897                 return 0;
 898
 899         trace_create_file("uprobe_events", 0644, d_tracer,
 900                                     NULL, &uprobe_events_ops);
 901         /* Profile interface */
 902         trace_create_file("uprobe_profile", 0444, d_tracer,
 903                                     NULL, &uprobe_profile_ops);
 904         return 0;
 905 }
 906
 907 fs_initcall(init_uprobe_trace);