Pileus Git - ~andy/linux/blob - 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 trace_uprobe_filter      filter;
  55         struct uprobe_consumer          consumer;
  56         struct inode                    *inode;
  57         char                            *filename;
  58         unsigned long                   offset;
  59         unsigned long                   nhit;
  60         struct trace_probe              tp;
  61 };
  62
  63 #define SIZEOF_TRACE_UPROBE(n)                          \
  64         (offsetof(struct trace_uprobe, tp.args) +       \
  65         (sizeof(struct probe_arg) * (n)))
  66
  67 static int register_uprobe_event(struct trace_uprobe *tu);
  68 static int unregister_uprobe_event(struct trace_uprobe *tu);
  69
  70 static DEFINE_MUTEX(uprobe_lock);
  71 static LIST_HEAD(uprobe_list);
  72
  73 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  74 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  75                                 unsigned long func, struct pt_regs *regs);
  76
  77 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
  78 {
  79         rwlock_init(&filter->rwlock);
  80         filter->nr_systemwide = 0;
  81         INIT_LIST_HEAD(&filter->perf_events);
  82 }
  83
  84 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
  85 {
  86         return !filter->nr_systemwide && list_empty(&filter->perf_events);
  87 }
  88
  89 static inline bool is_ret_probe(struct trace_uprobe *tu)
  90 {
  91         return tu->consumer.ret_handler != NULL;
  92 }
  93
  94 /*
  95  * Allocate new trace_uprobe and initialize it (including uprobes).
  96  */
  97 static struct trace_uprobe *
  98 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
  99 {
 100         struct trace_uprobe *tu;
 101
 102         if (!event || !is_good_name(event))
 103                 return ERR_PTR(-EINVAL);
 104
 105         if (!group || !is_good_name(group))
 106                 return ERR_PTR(-EINVAL);
 107
 108         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 109         if (!tu)
 110                 return ERR_PTR(-ENOMEM);
 111
 112         tu->tp.call.class = &tu->tp.class;
 113         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 114         if (!tu->tp.call.name)
 115                 goto error;
 116
 117         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 118         if (!tu->tp.class.system)
 119                 goto error;
 120
 121         INIT_LIST_HEAD(&tu->list);
 122         tu->consumer.handler = uprobe_dispatcher;
 123         if (is_ret)
 124                 tu->consumer.ret_handler = uretprobe_dispatcher;
 125         init_trace_uprobe_filter(&tu->filter);
 126         tu->tp.call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
 127         return tu;
 128
 129 error:
 130         kfree(tu->tp.call.name);
 131         kfree(tu);
 132
 133         return ERR_PTR(-ENOMEM);
 134 }
 135
 136 static void free_trace_uprobe(struct trace_uprobe *tu)
 137 {
 138         int i;
 139
 140         for (i = 0; i < tu->tp.nr_args; i++)
 141                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 142
 143         iput(tu->inode);
 144         kfree(tu->tp.call.class->system);
 145         kfree(tu->tp.call.name);
 146         kfree(tu->filename);
 147         kfree(tu);
 148 }
 149
 150 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 151 {
 152         struct trace_uprobe *tu;
 153
 154         list_for_each_entry(tu, &uprobe_list, list)
 155                 if (strcmp(tu->tp.call.name, event) == 0 &&
 156                     strcmp(tu->tp.call.class->system, group) == 0)
 157                         return tu;
 158
 159         return NULL;
 160 }
 161
 162 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 163 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 164 {
 165         int ret;
 166
 167         ret = unregister_uprobe_event(tu);
 168         if (ret)
 169                 return ret;
 170
 171         list_del(&tu->list);
 172         free_trace_uprobe(tu);
 173         return 0;
 174 }
 175
 176 /* Register a trace_uprobe and probe_event */
 177 static int register_trace_uprobe(struct trace_uprobe *tu)
 178 {
 179         struct trace_uprobe *old_tu;
 180         int ret;
 181
 182         mutex_lock(&uprobe_lock);
 183
 184         /* register as an event */
 185         old_tu = find_probe_event(tu->tp.call.name, tu->tp.call.class->system);
 186         if (old_tu) {
 187                 /* delete old event */
 188                 ret = unregister_trace_uprobe(old_tu);
 189                 if (ret)
 190                         goto end;
 191         }
 192
 193         ret = register_uprobe_event(tu);
 194         if (ret) {
 195                 pr_warning("Failed to register probe event(%d)\n", ret);
 196                 goto end;
 197         }
 198
 199         list_add_tail(&tu->list, &uprobe_list);
 200
 201 end:
 202         mutex_unlock(&uprobe_lock);
 203
 204         return ret;
 205 }
 206
 207 /*
 208  * Argument syntax:
 209  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 210  *
 211  *  - Remove uprobe: -:[GRP/]EVENT
 212  */
 213 static int create_trace_uprobe(int argc, char **argv)
 214 {
 215         struct trace_uprobe *tu;
 216         struct inode *inode;
 217         char *arg, *event, *group, *filename;
 218         char buf[MAX_EVENT_NAME_LEN];
 219         struct path path;
 220         unsigned long offset;
 221         bool is_delete, is_return;
 222         int i, ret;
 223
 224         inode = NULL;
 225         ret = 0;
 226         is_delete = false;
 227         is_return = false;
 228         event = NULL;
 229         group = NULL;
 230
 231         /* argc must be >= 1 */
 232         if (argv[0][0] == '-')
 233                 is_delete = true;
 234         else if (argv[0][0] == 'r')
 235                 is_return = true;
 236         else if (argv[0][0] != 'p') {
 237                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 238                 return -EINVAL;
 239         }
 240
 241         if (argv[0][1] == ':') {
 242                 event = &argv[0][2];
 243                 arg = strchr(event, '/');
 244
 245                 if (arg) {
 246                         group = event;
 247                         event = arg + 1;
 248                         event[-1] = '\0';
 249
 250                         if (strlen(group) == 0) {
 251                                 pr_info("Group name is not specified\n");
 252                                 return -EINVAL;
 253                         }
 254                 }
 255                 if (strlen(event) == 0) {
 256                         pr_info("Event name is not specified\n");
 257                         return -EINVAL;
 258                 }
 259         }
 260         if (!group)
 261                 group = UPROBE_EVENT_SYSTEM;
 262
 263         if (is_delete) {
 264                 int ret;
 265
 266                 if (!event) {
 267                         pr_info("Delete command needs an event name.\n");
 268                         return -EINVAL;
 269                 }
 270                 mutex_lock(&uprobe_lock);
 271                 tu = find_probe_event(event, group);
 272
 273                 if (!tu) {
 274                         mutex_unlock(&uprobe_lock);
 275                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 276                         return -ENOENT;
 277                 }
 278                 /* delete an event */
 279                 ret = unregister_trace_uprobe(tu);
 280                 mutex_unlock(&uprobe_lock);
 281                 return ret;
 282         }
 283
 284         if (argc < 2) {
 285                 pr_info("Probe point is not specified.\n");
 286                 return -EINVAL;
 287         }
 288         if (isdigit(argv[1][0])) {
 289                 pr_info("probe point must be have a filename.\n");
 290                 return -EINVAL;
 291         }
 292         arg = strchr(argv[1], ':');
 293         if (!arg) {
 294                 ret = -EINVAL;
 295                 goto fail_address_parse;
 296         }
 297
 298         *arg++ = '\0';
 299         filename = argv[1];
 300         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 301         if (ret)
 302                 goto fail_address_parse;
 303
 304         inode = igrab(path.dentry->d_inode);
 305         path_put(&path);
 306
 307         if (!inode || !S_ISREG(inode->i_mode)) {
 308                 ret = -EINVAL;
 309                 goto fail_address_parse;
 310         }
 311
 312         ret = kstrtoul(arg, 0, &offset);
 313         if (ret)
 314                 goto fail_address_parse;
 315
 316         argc -= 2;
 317         argv += 2;
 318
 319         /* setup a probe */
 320         if (!event) {
 321                 char *tail;
 322                 char *ptr;
 323
 324                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 325                 if (!tail) {
 326                         ret = -ENOMEM;
 327                         goto fail_address_parse;
 328                 }
 329
 330                 ptr = strpbrk(tail, ".-_");
 331                 if (ptr)
 332                         *ptr = '\0';
 333
 334                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 335                 event = buf;
 336                 kfree(tail);
 337         }
 338
 339         tu = alloc_trace_uprobe(group, event, argc, is_return);
 340         if (IS_ERR(tu)) {
 341                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 342                 ret = PTR_ERR(tu);
 343                 goto fail_address_parse;
 344         }
 345         tu->offset = offset;
 346         tu->inode = inode;
 347         tu->filename = kstrdup(filename, GFP_KERNEL);
 348
 349         if (!tu->filename) {
 350                 pr_info("Failed to allocate filename.\n");
 351                 ret = -ENOMEM;
 352                 goto error;
 353         }
 354
 355         /* parse arguments */
 356         ret = 0;
 357         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 358                 struct probe_arg *parg = &tu->tp.args[i];
 359
 360                 /* Increment count for freeing args in error case */
 361                 tu->tp.nr_args++;
 362
 363                 /* Parse argument name */
 364                 arg = strchr(argv[i], '=');
 365                 if (arg) {
 366                         *arg++ = '\0';
 367                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 368                 } else {
 369                         arg = argv[i];
 370                         /* If argument name is omitted, set "argN" */
 371                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 372                         parg->name = kstrdup(buf, GFP_KERNEL);
 373                 }
 374
 375                 if (!parg->name) {
 376                         pr_info("Failed to allocate argument[%d] name.\n", i);
 377                         ret = -ENOMEM;
 378                         goto error;
 379                 }
 380
 381                 if (!is_good_name(parg->name)) {
 382                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 383                         ret = -EINVAL;
 384                         goto error;
 385                 }
 386
 387                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 388                         pr_info("Argument[%d] name '%s' conflicts with "
 389                                 "another field.\n", i, argv[i]);
 390                         ret = -EINVAL;
 391                         goto error;
 392                 }
 393
 394                 /* Parse fetch argument */
 395                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 396                                                  false, false);
 397                 if (ret) {
 398                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 399                         goto error;
 400                 }
 401         }
 402
 403         ret = register_trace_uprobe(tu);
 404         if (ret)
 405                 goto error;
 406         return 0;
 407
 408 error:
 409         free_trace_uprobe(tu);
 410         return ret;
 411
 412 fail_address_parse:
 413         if (inode)
 414                 iput(inode);
 415
 416         pr_info("Failed to parse address or file.\n");
 417
 418         return ret;
 419 }
 420
 421 static int cleanup_all_probes(void)
 422 {
 423         struct trace_uprobe *tu;
 424         int ret = 0;
 425
 426         mutex_lock(&uprobe_lock);
 427         while (!list_empty(&uprobe_list)) {
 428                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 429                 ret = unregister_trace_uprobe(tu);
 430                 if (ret)
 431                         break;
 432         }
 433         mutex_unlock(&uprobe_lock);
 434         return ret;
 435 }
 436
 437 /* Probes listing interfaces */
 438 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 439 {
 440         mutex_lock(&uprobe_lock);
 441         return seq_list_start(&uprobe_list, *pos);
 442 }
 443
 444 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 445 {
 446         return seq_list_next(v, &uprobe_list, pos);
 447 }
 448
 449 static void probes_seq_stop(struct seq_file *m, void *v)
 450 {
 451         mutex_unlock(&uprobe_lock);
 452 }
 453
 454 static int probes_seq_show(struct seq_file *m, void *v)
 455 {
 456         struct trace_uprobe *tu = v;
 457         char c = is_ret_probe(tu) ? 'r' : 'p';
 458         int i;
 459
 460         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system, tu->tp.call.name);
 461         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 462
 463         for (i = 0; i < tu->tp.nr_args; i++)
 464                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 465
 466         seq_printf(m, "\n");
 467         return 0;
 468 }
 469
 470 static const struct seq_operations probes_seq_op = {
 471         .start  = probes_seq_start,
 472         .next   = probes_seq_next,
 473         .stop   = probes_seq_stop,
 474         .show   = probes_seq_show
 475 };
 476
 477 static int probes_open(struct inode *inode, struct file *file)
 478 {
 479         int ret;
 480
 481         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 482                 ret = cleanup_all_probes();
 483                 if (ret)
 484                         return ret;
 485         }
 486
 487         return seq_open(file, &probes_seq_op);
 488 }
 489
 490 static ssize_t probes_write(struct file *file, const char __user *buffer,
 491                             size_t count, loff_t *ppos)
 492 {
 493         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 494 }
 495
 496 static const struct file_operations uprobe_events_ops = {
 497         .owner          = THIS_MODULE,
 498         .open           = probes_open,
 499         .read           = seq_read,
 500         .llseek         = seq_lseek,
 501         .release        = seq_release,
 502         .write          = probes_write,
 503 };
 504
 505 /* Probes profiling interfaces */
 506 static int probes_profile_seq_show(struct seq_file *m, void *v)
 507 {
 508         struct trace_uprobe *tu = v;
 509
 510         seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->tp.call.name, tu->nhit);
 511         return 0;
 512 }
 513
 514 static const struct seq_operations profile_seq_op = {
 515         .start  = probes_seq_start,
 516         .next   = probes_seq_next,
 517         .stop   = probes_seq_stop,
 518         .show   = probes_profile_seq_show
 519 };
 520
 521 static int profile_open(struct inode *inode, struct file *file)
 522 {
 523         return seq_open(file, &profile_seq_op);
 524 }
 525
 526 static const struct file_operations uprobe_profile_ops = {
 527         .owner          = THIS_MODULE,
 528         .open           = profile_open,
 529         .read           = seq_read,
 530         .llseek         = seq_lseek,
 531         .release        = seq_release,
 532 };
 533
 534 static void uprobe_trace_print(struct trace_uprobe *tu,
 535                                 unsigned long func, struct pt_regs *regs)
 536 {
 537         struct uprobe_trace_entry_head *entry;
 538         struct ring_buffer_event *event;
 539         struct ring_buffer *buffer;
 540         void *data;
 541         int size, i;
 542         struct ftrace_event_call *call = &tu->tp.call;
 543
 544         size = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 545         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 546                                                   size + tu->tp.size, 0, 0);
 547         if (!event)
 548                 return;
 549
 550         entry = ring_buffer_event_data(event);
 551         if (is_ret_probe(tu)) {
 552                 entry->vaddr[0] = func;
 553                 entry->vaddr[1] = instruction_pointer(regs);
 554                 data = DATAOF_TRACE_ENTRY(entry, true);
 555         } else {
 556                 entry->vaddr[0] = instruction_pointer(regs);
 557                 data = DATAOF_TRACE_ENTRY(entry, false);
 558         }
 559
 560         for (i = 0; i < tu->tp.nr_args; i++) {
 561                 call_fetch(&tu->tp.args[i].fetch, regs,
 562                            data + tu->tp.args[i].offset);
 563         }
 564
 565         if (!call_filter_check_discard(call, entry, buffer, event))
 566                 trace_buffer_unlock_commit(buffer, event, 0, 0);
 567 }
 568
 569 /* uprobe handler */
 570 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
 571 {
 572         if (!is_ret_probe(tu))
 573                 uprobe_trace_print(tu, 0, regs);
 574         return 0;
 575 }
 576
 577 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 578                                 struct pt_regs *regs)
 579 {
 580         uprobe_trace_print(tu, func, regs);
 581 }
 582
 583 /* Event entry printers */
 584 static enum print_line_t
 585 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 586 {
 587         struct uprobe_trace_entry_head *entry;
 588         struct trace_seq *s = &iter->seq;
 589         struct trace_uprobe *tu;
 590         u8 *data;
 591         int i;
 592
 593         entry = (struct uprobe_trace_entry_head *)iter->ent;
 594         tu = container_of(event, struct trace_uprobe, tp.call.event);
 595
 596         if (is_ret_probe(tu)) {
 597                 if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)", tu->tp.call.name,
 598                                         entry->vaddr[1], entry->vaddr[0]))
 599                         goto partial;
 600                 data = DATAOF_TRACE_ENTRY(entry, true);
 601         } else {
 602                 if (!trace_seq_printf(s, "%s: (0x%lx)", tu->tp.call.name,
 603                                         entry->vaddr[0]))
 604                         goto partial;
 605                 data = DATAOF_TRACE_ENTRY(entry, false);
 606         }
 607
 608         for (i = 0; i < tu->tp.nr_args; i++) {
 609                 struct probe_arg *parg = &tu->tp.args[i];
 610
 611                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 612                         goto partial;
 613         }
 614
 615         if (trace_seq_puts(s, "\n"))
 616                 return TRACE_TYPE_HANDLED;
 617
 618 partial:
 619         return TRACE_TYPE_PARTIAL_LINE;
 620 }
 621
 622 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 623                                 enum uprobe_filter_ctx ctx,
 624                                 struct mm_struct *mm);
 625
 626 static int
 627 probe_event_enable(struct trace_uprobe *tu, int flag, filter_func_t filter)
 628 {
 629         int ret = 0;
 630
 631         if (trace_probe_is_enabled(&tu->tp))
 632                 return -EINTR;
 633
 634         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 635
 636         tu->tp.flags |= flag;
 637         tu->consumer.filter = filter;
 638         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 639         if (ret)
 640                 tu->tp.flags &= ~flag;
 641
 642         return ret;
 643 }
 644
 645 static void probe_event_disable(struct trace_uprobe *tu, int flag)
 646 {
 647         if (!trace_probe_is_enabled(&tu->tp))
 648                 return;
 649
 650         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 651
 652         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 653         tu->tp.flags &= ~flag;
 654 }
 655
 656 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 657 {
 658         int ret, i, size;
 659         struct uprobe_trace_entry_head field;
 660         struct trace_uprobe *tu = event_call->data;
 661
 662         if (is_ret_probe(tu)) {
 663                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 664                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 665                 size = SIZEOF_TRACE_ENTRY(true);
 666         } else {
 667                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 668                 size = SIZEOF_TRACE_ENTRY(false);
 669         }
 670         /* Set argument names as fields */
 671         for (i = 0; i < tu->tp.nr_args; i++) {
 672                 struct probe_arg *parg = &tu->tp.args[i];
 673
 674                 ret = trace_define_field(event_call, parg->type->fmttype,
 675                                          parg->name, size + parg->offset,
 676                                          parg->type->size, parg->type->is_signed,
 677                                          FILTER_OTHER);
 678
 679                 if (ret)
 680                         return ret;
 681         }
 682         return 0;
 683 }
 684
 685 #define LEN_OR_ZERO             (len ? len - pos : 0)
 686 static int __set_print_fmt(struct trace_uprobe *tu, char *buf, int len)
 687 {
 688         const char *fmt, *arg;
 689         int i;
 690         int pos = 0;
 691
 692         if (is_ret_probe(tu)) {
 693                 fmt = "(%lx <- %lx)";
 694                 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
 695         } else {
 696                 fmt = "(%lx)";
 697                 arg = "REC->" FIELD_STRING_IP;
 698         }
 699
 700         /* When len=0, we just calculate the needed length */
 701
 702         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 703
 704         for (i = 0; i < tu->tp.nr_args; i++) {
 705                 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
 706                                 tu->tp.args[i].name, tu->tp.args[i].type->fmt);
 707         }
 708
 709         pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 710
 711         for (i = 0; i < tu->tp.nr_args; i++) {
 712                 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
 713                                 tu->tp.args[i].name);
 714         }
 715
 716         return pos;     /* return the length of print_fmt */
 717 }
 718 #undef LEN_OR_ZERO
 719
 720 static int set_print_fmt(struct trace_uprobe *tu)
 721 {
 722         char *print_fmt;
 723         int len;
 724
 725         /* First: called with 0 length to calculate the needed length */
 726         len = __set_print_fmt(tu, NULL, 0);
 727         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 728         if (!print_fmt)
 729                 return -ENOMEM;
 730
 731         /* Second: actually write the @print_fmt */
 732         __set_print_fmt(tu, print_fmt, len + 1);
 733         tu->tp.call.print_fmt = print_fmt;
 734
 735         return 0;
 736 }
 737
 738 #ifdef CONFIG_PERF_EVENTS
 739 static bool
 740 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 741 {
 742         struct perf_event *event;
 743
 744         if (filter->nr_systemwide)
 745                 return true;
 746
 747         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
 748                 if (event->hw.tp_target->mm == mm)
 749                         return true;
 750         }
 751
 752         return false;
 753 }
 754
 755 static inline bool
 756 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 757 {
 758         return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 759 }
 760
 761 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 762 {
 763         bool done;
 764
 765         write_lock(&tu->filter.rwlock);
 766         if (event->hw.tp_target) {
 767                 /*
 768                  * event->parent != NULL means copy_process(), we can avoid
 769                  * uprobe_apply(). current->mm must be probed and we can rely
 770                  * on dup_mmap() which preserves the already installed bp's.
 771                  *
 772                  * attr.enable_on_exec means that exec/mmap will install the
 773                  * breakpoints we need.
 774                  */
 775                 done = tu->filter.nr_systemwide ||
 776                         event->parent || event->attr.enable_on_exec ||
 777                         uprobe_filter_event(tu, event);
 778                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
 779         } else {
 780                 done = tu->filter.nr_systemwide;
 781                 tu->filter.nr_systemwide++;
 782         }
 783         write_unlock(&tu->filter.rwlock);
 784
 785         if (!done)
 786                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
 787
 788         return 0;
 789 }
 790
 791 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 792 {
 793         bool done;
 794
 795         write_lock(&tu->filter.rwlock);
 796         if (event->hw.tp_target) {
 797                 list_del(&event->hw.tp_list);
 798                 done = tu->filter.nr_systemwide ||
 799                         (event->hw.tp_target->flags & PF_EXITING) ||
 800                         uprobe_filter_event(tu, event);
 801         } else {
 802                 tu->filter.nr_systemwide--;
 803                 done = tu->filter.nr_systemwide;
 804         }
 805         write_unlock(&tu->filter.rwlock);
 806
 807         if (!done)
 808                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 809
 810         return 0;
 811 }
 812
 813 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 814                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
 815 {
 816         struct trace_uprobe *tu;
 817         int ret;
 818
 819         tu = container_of(uc, struct trace_uprobe, consumer);
 820         read_lock(&tu->filter.rwlock);
 821         ret = __uprobe_perf_filter(&tu->filter, mm);
 822         read_unlock(&tu->filter.rwlock);
 823
 824         return ret;
 825 }
 826
 827 static void uprobe_perf_print(struct trace_uprobe *tu,
 828                                 unsigned long func, struct pt_regs *regs)
 829 {
 830         struct ftrace_event_call *call = &tu->tp.call;
 831         struct uprobe_trace_entry_head *entry;
 832         struct hlist_head *head;
 833         void *data;
 834         int size, rctx, i;
 835
 836         size = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 837         size = ALIGN(size + tu->tp.size + sizeof(u32), sizeof(u64)) - sizeof(u32);
 838
 839         preempt_disable();
 840         head = this_cpu_ptr(call->perf_events);
 841         if (hlist_empty(head))
 842                 goto out;
 843
 844         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 845         if (!entry)
 846                 goto out;
 847
 848         if (is_ret_probe(tu)) {
 849                 entry->vaddr[0] = func;
 850                 entry->vaddr[1] = instruction_pointer(regs);
 851                 data = DATAOF_TRACE_ENTRY(entry, true);
 852         } else {
 853                 entry->vaddr[0] = instruction_pointer(regs);
 854                 data = DATAOF_TRACE_ENTRY(entry, false);
 855         }
 856
 857         for (i = 0; i < tu->tp.nr_args; i++) {
 858                 struct probe_arg *parg = &tu->tp.args[i];
 859
 860                 call_fetch(&parg->fetch, regs, data + parg->offset);
 861         }
 862
 863         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 864  out:
 865         preempt_enable();
 866 }
 867
 868 /* uprobe profile handler */
 869 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
 870 {
 871         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
 872                 return UPROBE_HANDLER_REMOVE;
 873
 874         if (!is_ret_probe(tu))
 875                 uprobe_perf_print(tu, 0, regs);
 876         return 0;
 877 }
 878
 879 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
 880                                 struct pt_regs *regs)
 881 {
 882         uprobe_perf_print(tu, func, regs);
 883 }
 884 #endif  /* CONFIG_PERF_EVENTS */
 885
 886 static
 887 int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
 888 {
 889         struct trace_uprobe *tu = event->data;
 890
 891         switch (type) {
 892         case TRACE_REG_REGISTER:
 893                 return probe_event_enable(tu, TP_FLAG_TRACE, NULL);
 894
 895         case TRACE_REG_UNREGISTER:
 896                 probe_event_disable(tu, TP_FLAG_TRACE);
 897                 return 0;
 898
 899 #ifdef CONFIG_PERF_EVENTS
 900         case TRACE_REG_PERF_REGISTER:
 901                 return probe_event_enable(tu, TP_FLAG_PROFILE, uprobe_perf_filter);
 902
 903         case TRACE_REG_PERF_UNREGISTER:
 904                 probe_event_disable(tu, TP_FLAG_PROFILE);
 905                 return 0;
 906
 907         case TRACE_REG_PERF_OPEN:
 908                 return uprobe_perf_open(tu, data);
 909
 910         case TRACE_REG_PERF_CLOSE:
 911                 return uprobe_perf_close(tu, data);
 912
 913 #endif
 914         default:
 915                 return 0;
 916         }
 917         return 0;
 918 }
 919
 920 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
 921 {
 922         struct trace_uprobe *tu;
 923         int ret = 0;
 924
 925         tu = container_of(con, struct trace_uprobe, consumer);
 926         tu->nhit++;
 927
 928         if (tu->tp.flags & TP_FLAG_TRACE)
 929                 ret |= uprobe_trace_func(tu, regs);
 930
 931 #ifdef CONFIG_PERF_EVENTS
 932         if (tu->tp.flags & TP_FLAG_PROFILE)
 933                 ret |= uprobe_perf_func(tu, regs);
 934 #endif
 935         return ret;
 936 }
 937
 938 static int uretprobe_dispatcher(struct uprobe_consumer *con,
 939                                 unsigned long func, struct pt_regs *regs)
 940 {
 941         struct trace_uprobe *tu;
 942
 943         tu = container_of(con, struct trace_uprobe, consumer);
 944
 945         if (tu->tp.flags & TP_FLAG_TRACE)
 946                 uretprobe_trace_func(tu, func, regs);
 947
 948 #ifdef CONFIG_PERF_EVENTS
 949         if (tu->tp.flags & TP_FLAG_PROFILE)
 950                 uretprobe_perf_func(tu, func, regs);
 951 #endif
 952         return 0;
 953 }
 954
 955 static struct trace_event_functions uprobe_funcs = {
 956         .trace          = print_uprobe_event
 957 };
 958
 959 static int register_uprobe_event(struct trace_uprobe *tu)
 960 {
 961         struct ftrace_event_call *call = &tu->tp.call;
 962         int ret;
 963
 964         /* Initialize ftrace_event_call */
 965         INIT_LIST_HEAD(&call->class->fields);
 966         call->event.funcs = &uprobe_funcs;
 967         call->class->define_fields = uprobe_event_define_fields;
 968
 969         if (set_print_fmt(tu) < 0)
 970                 return -ENOMEM;
 971
 972         ret = register_ftrace_event(&call->event);
 973         if (!ret) {
 974                 kfree(call->print_fmt);
 975                 return -ENODEV;
 976         }
 977         call->flags = 0;
 978         call->class->reg = trace_uprobe_register;
 979         call->data = tu;
 980         ret = trace_add_event_call(call);
 981
 982         if (ret) {
 983                 pr_info("Failed to register uprobe event: %s\n", call->name);
 984                 kfree(call->print_fmt);
 985                 unregister_ftrace_event(&call->event);
 986         }
 987
 988         return ret;
 989 }
 990
 991 static int unregister_uprobe_event(struct trace_uprobe *tu)
 992 {
 993         int ret;
 994
 995         /* tu->event is unregistered in trace_remove_event_call() */
 996         ret = trace_remove_event_call(&tu->tp.call);
 997         if (ret)
 998                 return ret;
 999         kfree(tu->tp.call.print_fmt);
1000         tu->tp.call.print_fmt = NULL;
1001         return 0;
1002 }
1003
1004 /* Make a trace interface for controling probe points */
1005 static __init int init_uprobe_trace(void)
1006 {
1007         struct dentry *d_tracer;
1008
1009         d_tracer = tracing_init_dentry();
1010         if (!d_tracer)
1011                 return 0;
1012
1013         trace_create_file("uprobe_events", 0644, d_tracer,
1014                                     NULL, &uprobe_events_ops);
1015         /* Profile interface */
1016         trace_create_file("uprobe_profile", 0444, d_tracer,
1017                                     NULL, &uprobe_profile_ops);
1018         return 0;
1019 }
1020
1021 fs_initcall(init_uprobe_trace);