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