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