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