Pileus Git - ~andy/linux/blob - kernel/trace/trace_uprobe.c

   1 /*
   2  * uprobes-based tracing events
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  16  *
  17  * Copyright (C) IBM Corporation, 2010-2012
  18  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/module.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/uprobes.h>
  24 #include <linux/namei.h>
  25 #include <linux/string.h>
  26
  27 #include "trace_probe.h"
  28
  29 #define UPROBE_EVENT_SYSTEM     "uprobes"
  30
  31 struct uprobe_trace_entry_head {
  32         struct trace_entry      ent;
  33         unsigned long           vaddr[];
  34 };
  35
  36 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  37         (sizeof(struct uprobe_trace_entry_head) +       \
  38          sizeof(unsigned long) * (is_return ? 2 : 1))
  39
  40 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  41         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  42
  43 struct trace_uprobe_filter {
  44         rwlock_t                rwlock;
  45         int                     nr_systemwide;
  46         struct list_head        perf_events;
  47 };
  48
  49 /*
  50  * uprobe event core functions
  51  */
  52 struct trace_uprobe {
  53         struct list_head                list;
  54         struct 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         tu->call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
 132         return tu;
 133
 134 error:
 135         kfree(tu->call.name);
 136         kfree(tu);
 137
 138         return ERR_PTR(-ENOMEM);
 139 }
 140
 141 static void free_trace_uprobe(struct trace_uprobe *tu)
 142 {
 143         int i;
 144
 145         for (i = 0; i < tu->nr_args; i++)
 146                 traceprobe_free_probe_arg(&tu->args[i]);
 147
 148         iput(tu->inode);
 149         kfree(tu->call.class->system);
 150         kfree(tu->call.name);
 151         kfree(tu->filename);
 152         kfree(tu);
 153 }
 154
 155 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 156 {
 157         struct trace_uprobe *tu;
 158
 159         list_for_each_entry(tu, &uprobe_list, list)
 160                 if (strcmp(tu->call.name, event) == 0 &&
 161                     strcmp(tu->call.class->system, group) == 0)
 162                         return tu;
 163
 164         return NULL;
 165 }
 166
 167 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 168 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 169 {
 170         int ret;
 171
 172         ret = unregister_uprobe_event(tu);
 173         if (ret)
 174                 return ret;
 175
 176         list_del(&tu->list);
 177         free_trace_uprobe(tu);
 178         return 0;
 179 }
 180
 181 /* Register a trace_uprobe and probe_event */
 182 static int register_trace_uprobe(struct trace_uprobe *tu)
 183 {
 184         struct trace_uprobe *old_tp;
 185         int ret;
 186
 187         mutex_lock(&uprobe_lock);
 188
 189         /* register as an event */
 190         old_tp = find_probe_event(tu->call.name, tu->call.class->system);
 191         if (old_tp) {
 192                 /* delete old event */
 193                 ret = unregister_trace_uprobe(old_tp);
 194                 if (ret)
 195                         goto end;
 196         }
 197
 198         ret = register_uprobe_event(tu);
 199         if (ret) {
 200                 pr_warning("Failed to register probe event(%d)\n", ret);
 201                 goto end;
 202         }
 203
 204         list_add_tail(&tu->list, &uprobe_list);
 205
 206 end:
 207         mutex_unlock(&uprobe_lock);
 208
 209         return ret;
 210 }
 211
 212 /*
 213  * Argument syntax:
 214  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:SYMBOL [FETCHARGS]
 215  *
 216  *  - Remove uprobe: -:[GRP/]EVENT
 217  */
 218 static int create_trace_uprobe(int argc, char **argv)
 219 {
 220         struct trace_uprobe *tu;
 221         struct inode *inode;
 222         char *arg, *event, *group, *filename;
 223         char buf[MAX_EVENT_NAME_LEN];
 224         struct path path;
 225         unsigned long offset;
 226         bool is_delete, is_return;
 227         int i, ret;
 228
 229         inode = NULL;
 230         ret = 0;
 231         is_delete = false;
 232         is_return = false;
 233         event = NULL;
 234         group = NULL;
 235
 236         /* argc must be >= 1 */
 237         if (argv[0][0] == '-')
 238                 is_delete = true;
 239         else if (argv[0][0] == 'r')
 240                 is_return = true;
 241         else if (argv[0][0] != 'p') {
 242                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 243                 return -EINVAL;
 244         }
 245
 246         if (argv[0][1] == ':') {
 247                 event = &argv[0][2];
 248                 arg = strchr(event, '/');
 249
 250                 if (arg) {
 251                         group = event;
 252                         event = arg + 1;
 253                         event[-1] = '\0';
 254
 255                         if (strlen(group) == 0) {
 256                                 pr_info("Group name is not specified\n");
 257                                 return -EINVAL;
 258                         }
 259                 }
 260                 if (strlen(event) == 0) {
 261                         pr_info("Event name is not specified\n");
 262                         return -EINVAL;
 263                 }
 264         }
 265         if (!group)
 266                 group = UPROBE_EVENT_SYSTEM;
 267
 268         if (is_delete) {
 269                 int ret;
 270
 271                 if (!event) {
 272                         pr_info("Delete command needs an event name.\n");
 273                         return -EINVAL;
 274                 }
 275                 mutex_lock(&uprobe_lock);
 276                 tu = find_probe_event(event, group);
 277
 278                 if (!tu) {
 279                         mutex_unlock(&uprobe_lock);
 280                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 281                         return -ENOENT;
 282                 }
 283                 /* delete an event */
 284                 ret = unregister_trace_uprobe(tu);
 285                 mutex_unlock(&uprobe_lock);
 286                 return ret;
 287         }
 288
 289         if (argc < 2) {
 290                 pr_info("Probe point is not specified.\n");
 291                 return -EINVAL;
 292         }
 293         if (isdigit(argv[1][0])) {
 294                 pr_info("probe point must be have a filename.\n");
 295                 return -EINVAL;
 296         }
 297         arg = strchr(argv[1], ':');
 298         if (!arg) {
 299                 ret = -EINVAL;
 300                 goto fail_address_parse;
 301         }
 302
 303         *arg++ = '\0';
 304         filename = argv[1];
 305         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 306         if (ret)
 307                 goto fail_address_parse;
 308
 309         inode = igrab(path.dentry->d_inode);
 310         path_put(&path);
 311
 312         if (!inode || !S_ISREG(inode->i_mode)) {
 313                 ret = -EINVAL;
 314                 goto fail_address_parse;
 315         }
 316
 317         ret = kstrtoul(arg, 0, &offset);
 318         if (ret)
 319                 goto fail_address_parse;
 320
 321         argc -= 2;
 322         argv += 2;
 323
 324         /* setup a probe */
 325         if (!event) {
 326                 char *tail;
 327                 char *ptr;
 328
 329                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 330                 if (!tail) {
 331                         ret = -ENOMEM;
 332                         goto fail_address_parse;
 333                 }
 334
 335                 ptr = strpbrk(tail, ".-_");
 336                 if (ptr)
 337                         *ptr = '\0';
 338
 339                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 340                 event = buf;
 341                 kfree(tail);
 342         }
 343
 344         tu = alloc_trace_uprobe(group, event, argc, is_return);
 345         if (IS_ERR(tu)) {
 346                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 347                 ret = PTR_ERR(tu);
 348                 goto fail_address_parse;
 349         }
 350         tu->offset = offset;
 351         tu->inode = inode;
 352         tu->filename = kstrdup(filename, GFP_KERNEL);
 353
 354         if (!tu->filename) {
 355                 pr_info("Failed to allocate filename.\n");
 356                 ret = -ENOMEM;
 357                 goto error;
 358         }
 359
 360         /* parse arguments */
 361         ret = 0;
 362         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 363                 /* Increment count for freeing args in error case */
 364                 tu->nr_args++;
 365
 366                 /* Parse argument name */
 367                 arg = strchr(argv[i], '=');
 368                 if (arg) {
 369                         *arg++ = '\0';
 370                         tu->args[i].name = kstrdup(argv[i], GFP_KERNEL);
 371                 } else {
 372                         arg = argv[i];
 373                         /* If argument name is omitted, set "argN" */
 374                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 375                         tu->args[i].name = kstrdup(buf, GFP_KERNEL);
 376                 }
 377
 378                 if (!tu->args[i].name) {
 379                         pr_info("Failed to allocate argument[%d] name.\n", i);
 380                         ret = -ENOMEM;
 381                         goto error;
 382                 }
 383
 384                 if (!is_good_name(tu->args[i].name)) {
 385                         pr_info("Invalid argument[%d] name: %s\n", i, tu->args[i].name);
 386                         ret = -EINVAL;
 387                         goto error;
 388                 }
 389
 390                 if (traceprobe_conflict_field_name(tu->args[i].name, tu->args, i)) {
 391                         pr_info("Argument[%d] name '%s' conflicts with "
 392                                 "another field.\n", i, argv[i]);
 393                         ret = -EINVAL;
 394                         goto error;
 395                 }
 396
 397                 /* Parse fetch argument */
 398                 ret = traceprobe_parse_probe_arg(arg, &tu->size, &tu->args[i], false, false);
 399                 if (ret) {
 400                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 401                         goto error;
 402                 }
 403         }
 404
 405         ret = register_trace_uprobe(tu);
 406         if (ret)
 407                 goto error;
 408         return 0;
 409
 410 error:
 411         free_trace_uprobe(tu);
 412         return ret;
 413
 414 fail_address_parse:
 415         if (inode)
 416                 iput(inode);
 417
 418         pr_info("Failed to parse address or file.\n");
 419
 420         return ret;
 421 }
 422
 423 static int cleanup_all_probes(void)
 424 {
 425         struct trace_uprobe *tu;
 426         int ret = 0;
 427
 428         mutex_lock(&uprobe_lock);
 429         while (!list_empty(&uprobe_list)) {
 430                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 431                 ret = unregister_trace_uprobe(tu);
 432                 if (ret)
 433                         break;
 434         }
 435         mutex_unlock(&uprobe_lock);
 436         return ret;
 437 }
 438
 439 /* Probes listing interfaces */
 440 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 441 {
 442         mutex_lock(&uprobe_lock);
 443         return seq_list_start(&uprobe_list, *pos);
 444 }
 445
 446 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 447 {
 448         return seq_list_next(v, &uprobe_list, pos);
 449 }
 450
 451 static void probes_seq_stop(struct seq_file *m, void *v)
 452 {
 453         mutex_unlock(&uprobe_lock);
 454 }
 455
 456 static int probes_seq_show(struct seq_file *m, void *v)
 457 {
 458         struct trace_uprobe *tu = v;
 459         char c = is_ret_probe(tu) ? 'r' : 'p';
 460         int i;
 461
 462         seq_printf(m, "%c:%s/%s", c, tu->call.class->system, tu->call.name);
 463         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 464
 465         for (i = 0; i < tu->nr_args; i++)
 466                 seq_printf(m, " %s=%s", tu->args[i].name, tu->args[i].comm);
 467
 468         seq_printf(m, "\n");
 469         return 0;
 470 }
 471
 472 static const struct seq_operations probes_seq_op = {
 473         .start  = probes_seq_start,
 474         .next   = probes_seq_next,
 475         .stop   = probes_seq_stop,
 476         .show   = probes_seq_show
 477 };
 478
 479 static int probes_open(struct inode *inode, struct file *file)
 480 {
 481         int ret;
 482
 483         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 484                 ret = cleanup_all_probes();
 485                 if (ret)
 486                         return ret;
 487         }
 488
 489         return seq_open(file, &probes_seq_op);
 490 }
 491
 492 static ssize_t probes_write(struct file *file, const char __user *buffer,
 493                             size_t count, loff_t *ppos)
 494 {
 495         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 496 }
 497
 498 static const struct file_operations uprobe_events_ops = {
 499         .owner          = THIS_MODULE,
 500         .open           = probes_open,
 501         .read           = seq_read,
 502         .llseek         = seq_lseek,
 503         .release        = seq_release,
 504         .write          = probes_write,
 505 };
 506
 507 /* Probes profiling interfaces */
 508 static int probes_profile_seq_show(struct seq_file *m, void *v)
 509 {
 510         struct trace_uprobe *tu = v;
 511
 512         seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->call.name, tu->nhit);
 513         return 0;
 514 }
 515
 516 static const struct seq_operations profile_seq_op = {
 517         .start  = probes_seq_start,
 518         .next   = probes_seq_next,
 519         .stop   = probes_seq_stop,
 520         .show   = probes_profile_seq_show
 521 };
 522
 523 static int profile_open(struct inode *inode, struct file *file)
 524 {
 525         return seq_open(file, &profile_seq_op);
 526 }
 527
 528 static const struct file_operations uprobe_profile_ops = {
 529         .owner          = THIS_MODULE,
 530         .open           = profile_open,
 531         .read           = seq_read,
 532         .llseek         = seq_lseek,
 533         .release        = seq_release,
 534 };
 535
 536 static void uprobe_trace_print(struct trace_uprobe *tu,
 537                                 unsigned long func, struct pt_regs *regs)
 538 {
 539         struct uprobe_trace_entry_head *entry;
 540         struct ring_buffer_event *event;
 541         struct ring_buffer *buffer;
 542         void *data;
 543         int size, i;
 544         struct ftrace_event_call *call = &tu->call;
 545
 546         size = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 547         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 548                                                   size + tu->size, 0, 0);
 549         if (!event)
 550                 return;
 551
 552         entry = ring_buffer_event_data(event);
 553         if (is_ret_probe(tu)) {
 554                 entry->vaddr[0] = func;
 555                 entry->vaddr[1] = instruction_pointer(regs);
 556                 data = DATAOF_TRACE_ENTRY(entry, true);
 557         } else {
 558                 entry->vaddr[0] = instruction_pointer(regs);
 559                 data = DATAOF_TRACE_ENTRY(entry, false);
 560         }
 561
 562         for (i = 0; i < tu->nr_args; i++)
 563                 call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
 564
 565         if (!call_filter_check_discard(call, entry, buffer, event))
 566                 trace_buffer_unlock_commit(buffer, event, 0, 0);
 567 }
 568
 569 /* uprobe handler */
 570 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
 571 {
 572         if (!is_ret_probe(tu))
 573                 uprobe_trace_print(tu, 0, regs);
 574         return 0;
 575 }
 576
 577 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 578                                 struct pt_regs *regs)
 579 {
 580         uprobe_trace_print(tu, func, regs);
 581 }
 582
 583 /* Event entry printers */
 584 static enum print_line_t
 585 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 586 {
 587         struct uprobe_trace_entry_head *entry;
 588         struct trace_seq *s = &iter->seq;
 589         struct trace_uprobe *tu;
 590         u8 *data;
 591         int i;
 592
 593         entry = (struct uprobe_trace_entry_head *)iter->ent;
 594         tu = container_of(event, struct trace_uprobe, call.event);
 595
 596         if (is_ret_probe(tu)) {
 597                 if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)", tu->call.name,
 598                                         entry->vaddr[1], entry->vaddr[0]))
 599                         goto partial;
 600                 data = DATAOF_TRACE_ENTRY(entry, true);
 601         } else {
 602                 if (!trace_seq_printf(s, "%s: (0x%lx)", tu->call.name,
 603                                         entry->vaddr[0]))
 604                         goto partial;
 605                 data = DATAOF_TRACE_ENTRY(entry, false);
 606         }
 607
 608         for (i = 0; i < tu->nr_args; i++) {
 609                 if (!tu->args[i].type->print(s, tu->args[i].name,
 610                                              data + tu->args[i].offset, entry))
 611                         goto partial;
 612         }
 613
 614         if (trace_seq_puts(s, "\n"))
 615                 return TRACE_TYPE_HANDLED;
 616
 617 partial:
 618         return TRACE_TYPE_PARTIAL_LINE;
 619 }
 620
 621 static inline bool is_trace_uprobe_enabled(struct trace_uprobe *tu)
 622 {
 623         return tu->flags & (TP_FLAG_TRACE | TP_FLAG_PROFILE);
 624 }
 625
 626 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 627                                 enum uprobe_filter_ctx ctx,
 628                                 struct mm_struct *mm);
 629
 630 static int
 631 probe_event_enable(struct trace_uprobe *tu, int flag, filter_func_t filter)
 632 {
 633         int ret = 0;
 634
 635         if (is_trace_uprobe_enabled(tu))
 636                 return -EINTR;
 637
 638         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 639
 640         tu->flags |= flag;
 641         tu->consumer.filter = filter;
 642         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 643         if (ret)
 644                 tu->flags &= ~flag;
 645
 646         return ret;
 647 }
 648
 649 static void probe_event_disable(struct trace_uprobe *tu, int flag)
 650 {
 651         if (!is_trace_uprobe_enabled(tu))
 652                 return;
 653
 654         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 655
 656         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 657         tu->flags &= ~flag;
 658 }
 659
 660 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 661 {
 662         int ret, i, size;
 663         struct uprobe_trace_entry_head field;
 664         struct trace_uprobe *tu = event_call->data;
 665
 666         if (is_ret_probe(tu)) {
 667                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 668                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 669                 size = SIZEOF_TRACE_ENTRY(true);
 670         } else {
 671                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 672                 size = SIZEOF_TRACE_ENTRY(false);
 673         }
 674         /* Set argument names as fields */
 675         for (i = 0; i < tu->nr_args; i++) {
 676                 ret = trace_define_field(event_call, tu->args[i].type->fmttype,
 677                                          tu->args[i].name,
 678                                          size + tu->args[i].offset,
 679                                          tu->args[i].type->size,
 680                                          tu->args[i].type->is_signed,
 681                                          FILTER_OTHER);
 682
 683                 if (ret)
 684                         return ret;
 685         }
 686         return 0;
 687 }
 688
 689 #define LEN_OR_ZERO             (len ? len - pos : 0)
 690 static int __set_print_fmt(struct trace_uprobe *tu, char *buf, int len)
 691 {
 692         const char *fmt, *arg;
 693         int i;
 694         int pos = 0;
 695
 696         if (is_ret_probe(tu)) {
 697                 fmt = "(%lx <- %lx)";
 698                 arg = "REC->" FIELD_STRING_FUNC ", REC->" FIELD_STRING_RETIP;
 699         } else {
 700                 fmt = "(%lx)";
 701                 arg = "REC->" FIELD_STRING_IP;
 702         }
 703
 704         /* When len=0, we just calculate the needed length */
 705
 706         pos += snprintf(buf + pos, LEN_OR_ZERO, "\"%s", fmt);
 707
 708         for (i = 0; i < tu->nr_args; i++) {
 709                 pos += snprintf(buf + pos, LEN_OR_ZERO, " %s=%s",
 710                                 tu->args[i].name, tu->args[i].type->fmt);
 711         }
 712
 713         pos += snprintf(buf + pos, LEN_OR_ZERO, "\", %s", arg);
 714
 715         for (i = 0; i < tu->nr_args; i++) {
 716                 pos += snprintf(buf + pos, LEN_OR_ZERO, ", REC->%s",
 717                                 tu->args[i].name);
 718         }
 719
 720         return pos;     /* return the length of print_fmt */
 721 }
 722 #undef LEN_OR_ZERO
 723
 724 static int set_print_fmt(struct trace_uprobe *tu)
 725 {
 726         char *print_fmt;
 727         int len;
 728
 729         /* First: called with 0 length to calculate the needed length */
 730         len = __set_print_fmt(tu, NULL, 0);
 731         print_fmt = kmalloc(len + 1, GFP_KERNEL);
 732         if (!print_fmt)
 733                 return -ENOMEM;
 734
 735         /* Second: actually write the @print_fmt */
 736         __set_print_fmt(tu, print_fmt, len + 1);
 737         tu->call.print_fmt = print_fmt;
 738
 739         return 0;
 740 }
 741
 742 #ifdef CONFIG_PERF_EVENTS
 743 static bool
 744 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 745 {
 746         struct perf_event *event;
 747
 748         if (filter->nr_systemwide)
 749                 return true;
 750
 751         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
 752                 if (event->hw.tp_target->mm == mm)
 753                         return true;
 754         }
 755
 756         return false;
 757 }
 758
 759 static inline bool
 760 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 761 {
 762         return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 763 }
 764
 765 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 766 {
 767         bool done;
 768
 769         write_lock(&tu->filter.rwlock);
 770         if (event->hw.tp_target) {
 771                 /*
 772                  * event->parent != NULL means copy_process(), we can avoid
 773                  * uprobe_apply(). current->mm must be probed and we can rely
 774                  * on dup_mmap() which preserves the already installed bp's.
 775                  *
 776                  * attr.enable_on_exec means that exec/mmap will install the
 777                  * breakpoints we need.
 778                  */
 779                 done = tu->filter.nr_systemwide ||
 780                         event->parent || event->attr.enable_on_exec ||
 781                         uprobe_filter_event(tu, event);
 782                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
 783         } else {
 784                 done = tu->filter.nr_systemwide;
 785                 tu->filter.nr_systemwide++;
 786         }
 787         write_unlock(&tu->filter.rwlock);
 788
 789         if (!done)
 790                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
 791
 792         return 0;
 793 }
 794
 795 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 796 {
 797         bool done;
 798
 799         write_lock(&tu->filter.rwlock);
 800         if (event->hw.tp_target) {
 801                 list_del(&event->hw.tp_list);
 802                 done = tu->filter.nr_systemwide ||
 803                         (event->hw.tp_target->flags & PF_EXITING) ||
 804                         uprobe_filter_event(tu, event);
 805         } else {
 806                 tu->filter.nr_systemwide--;
 807                 done = tu->filter.nr_systemwide;
 808         }
 809         write_unlock(&tu->filter.rwlock);
 810
 811         if (!done)
 812                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 813
 814         return 0;
 815 }
 816
 817 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 818                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
 819 {
 820         struct trace_uprobe *tu;
 821         int ret;
 822
 823         tu = container_of(uc, struct trace_uprobe, consumer);
 824         read_lock(&tu->filter.rwlock);
 825         ret = __uprobe_perf_filter(&tu->filter, mm);
 826         read_unlock(&tu->filter.rwlock);
 827
 828         return ret;
 829 }
 830
 831 static void uprobe_perf_print(struct trace_uprobe *tu,
 832                                 unsigned long func, struct pt_regs *regs)
 833 {
 834         struct ftrace_event_call *call = &tu->call;
 835         struct uprobe_trace_entry_head *entry;
 836         struct hlist_head *head;
 837         void *data;
 838         int size, rctx, i;
 839
 840         size = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 841         size = ALIGN(size + tu->size + sizeof(u32), sizeof(u64)) - sizeof(u32);
 842
 843         preempt_disable();
 844         head = this_cpu_ptr(call->perf_events);
 845         if (hlist_empty(head))
 846                 goto out;
 847
 848         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 849         if (!entry)
 850                 goto out;
 851
 852         if (is_ret_probe(tu)) {
 853                 entry->vaddr[0] = func;
 854                 entry->vaddr[1] = instruction_pointer(regs);
 855                 data = DATAOF_TRACE_ENTRY(entry, true);
 856         } else {
 857                 entry->vaddr[0] = instruction_pointer(regs);
 858                 data = DATAOF_TRACE_ENTRY(entry, false);
 859         }
 860
 861         for (i = 0; i < tu->nr_args; i++)
 862                 call_fetch(&tu->args[i].fetch, regs, data + tu->args[i].offset);
 863
 864         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 865  out:
 866         preempt_enable();
 867 }
 868
 869 /* uprobe profile handler */
 870 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
 871 {
 872         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
 873                 return UPROBE_HANDLER_REMOVE;
 874
 875         if (!is_ret_probe(tu))
 876                 uprobe_perf_print(tu, 0, regs);
 877         return 0;
 878 }
 879
 880 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
 881                                 struct pt_regs *regs)
 882 {
 883         uprobe_perf_print(tu, func, regs);
 884 }
 885 #endif  /* CONFIG_PERF_EVENTS */
 886
 887 static
 888 int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
 889 {
 890         struct trace_uprobe *tu = event->data;
 891
 892         switch (type) {
 893         case TRACE_REG_REGISTER:
 894                 return probe_event_enable(tu, TP_FLAG_TRACE, NULL);
 895
 896         case TRACE_REG_UNREGISTER:
 897                 probe_event_disable(tu, TP_FLAG_TRACE);
 898                 return 0;
 899
 900 #ifdef CONFIG_PERF_EVENTS
 901         case TRACE_REG_PERF_REGISTER:
 902                 return probe_event_enable(tu, TP_FLAG_PROFILE, uprobe_perf_filter);
 903
 904         case TRACE_REG_PERF_UNREGISTER:
 905                 probe_event_disable(tu, TP_FLAG_PROFILE);
 906                 return 0;
 907
 908         case TRACE_REG_PERF_OPEN:
 909                 return uprobe_perf_open(tu, data);
 910
 911         case TRACE_REG_PERF_CLOSE:
 912                 return uprobe_perf_close(tu, data);
 913
 914 #endif
 915         default:
 916                 return 0;
 917         }
 918         return 0;
 919 }
 920
 921 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
 922 {
 923         struct trace_uprobe *tu;
 924         int ret = 0;
 925
 926         tu = container_of(con, struct trace_uprobe, consumer);
 927         tu->nhit++;
 928
 929         if (tu->flags & TP_FLAG_TRACE)
 930                 ret |= uprobe_trace_func(tu, regs);
 931
 932 #ifdef CONFIG_PERF_EVENTS
 933         if (tu->flags & TP_FLAG_PROFILE)
 934                 ret |= uprobe_perf_func(tu, regs);
 935 #endif
 936         return ret;
 937 }
 938
 939 static int uretprobe_dispatcher(struct uprobe_consumer *con,
 940                                 unsigned long func, struct pt_regs *regs)
 941 {
 942         struct trace_uprobe *tu;
 943
 944         tu = container_of(con, struct trace_uprobe, consumer);
 945
 946         if (tu->flags & TP_FLAG_TRACE)
 947                 uretprobe_trace_func(tu, func, regs);
 948
 949 #ifdef CONFIG_PERF_EVENTS
 950         if (tu->flags & TP_FLAG_PROFILE)
 951                 uretprobe_perf_func(tu, func, regs);
 952 #endif
 953         return 0;
 954 }
 955
 956 static struct trace_event_functions uprobe_funcs = {
 957         .trace          = print_uprobe_event
 958 };
 959
 960 static int register_uprobe_event(struct trace_uprobe *tu)
 961 {
 962         struct ftrace_event_call *call = &tu->call;
 963         int ret;
 964
 965         /* Initialize ftrace_event_call */
 966         INIT_LIST_HEAD(&call->class->fields);
 967         call->event.funcs = &uprobe_funcs;
 968         call->class->define_fields = uprobe_event_define_fields;
 969
 970         if (set_print_fmt(tu) < 0)
 971                 return -ENOMEM;
 972
 973         ret = register_ftrace_event(&call->event);
 974         if (!ret) {
 975                 kfree(call->print_fmt);
 976                 return -ENODEV;
 977         }
 978         call->flags = 0;
 979         call->class->reg = trace_uprobe_register;
 980         call->data = tu;
 981         ret = trace_add_event_call(call);
 982
 983         if (ret) {
 984                 pr_info("Failed to register uprobe event: %s\n", call->name);
 985                 kfree(call->print_fmt);
 986                 unregister_ftrace_event(&call->event);
 987         }
 988
 989         return ret;
 990 }
 991
 992 static int unregister_uprobe_event(struct trace_uprobe *tu)
 993 {
 994         int ret;
 995
 996         /* tu->event is unregistered in trace_remove_event_call() */
 997         ret = trace_remove_event_call(&tu->call);
 998         if (ret)
 999                 return ret;
1000         kfree(tu->call.print_fmt);
1001         tu->call.print_fmt = NULL;
1002         return 0;
1003 }
1004
1005 /* Make a trace interface for controling probe points */
1006 static __init int init_uprobe_trace(void)
1007 {
1008         struct dentry *d_tracer;
1009
1010         d_tracer = tracing_init_dentry();
1011         if (!d_tracer)
1012                 return 0;
1013
1014         trace_create_file("uprobe_events", 0644, d_tracer,
1015                                     NULL, &uprobe_events_ops);
1016         /* Profile interface */
1017         trace_create_file("uprobe_profile", 0444, d_tracer,
1018                                     NULL, &uprobe_profile_ops);
1019         return 0;
1020 }
1021
1022 fs_initcall(init_uprobe_trace);