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