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 trace_uprobe_filter      filter;
  55         struct uprobe_consumer          consumer;
  56         struct inode                    *inode;
  57         char                            *filename;
  58         unsigned long                   offset;
  59         unsigned long                   nhit;
  60         struct trace_probe              tp;
  61 };
  62
  63 #define SIZEOF_TRACE_UPROBE(n)                          \
  64         (offsetof(struct trace_uprobe, tp.args) +       \
  65         (sizeof(struct probe_arg) * (n)))
  66
  67 static int register_uprobe_event(struct trace_uprobe *tu);
  68 static int unregister_uprobe_event(struct trace_uprobe *tu);
  69
  70 static DEFINE_MUTEX(uprobe_lock);
  71 static LIST_HEAD(uprobe_list);
  72
  73 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  74 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  75                                 unsigned long func, struct pt_regs *regs);
  76
  77 #ifdef CONFIG_STACK_GROWSUP
  78 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  79 {
  80         return addr - (n * sizeof(long));
  81 }
  82 #else
  83 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  84 {
  85         return addr + (n * sizeof(long));
  86 }
  87 #endif
  88
  89 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  90 {
  91         unsigned long ret;
  92         unsigned long addr = user_stack_pointer(regs);
  93
  94         addr = adjust_stack_addr(addr, n);
  95
  96         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
  97                 return 0;
  98
  99         return ret;
 100 }
 101
 102 /*
 103  * Uprobes-specific fetch functions
 104  */
 105 #define DEFINE_FETCH_stack(type)                                        \
 106 static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
 107                                           void *offset, void *dest)     \
 108 {                                                                       \
 109         *(type *)dest = (type)get_user_stack_nth(regs,                  \
 110                                               ((unsigned long)offset)); \
 111 }
 112 DEFINE_BASIC_FETCH_FUNCS(stack)
 113 /* No string on the stack entry */
 114 #define fetch_stack_string      NULL
 115 #define fetch_stack_string_size NULL
 116
 117 #define DEFINE_FETCH_memory(type)                                       \
 118 static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
 119                                                 void *addr, void *dest) \
 120 {                                                                       \
 121         type retval;                                                    \
 122         void __user *vaddr = (void __force __user *) addr;              \
 123                                                                         \
 124         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
 125                 *(type *)dest = 0;                                      \
 126         else                                                            \
 127                 *(type *) dest = retval;                                \
 128 }
 129 DEFINE_BASIC_FETCH_FUNCS(memory)
 130 /*
 131  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 132  * length and relative data location.
 133  */
 134 static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 135                                                       void *addr, void *dest)
 136 {
 137         long ret;
 138         u32 rloc = *(u32 *)dest;
 139         int maxlen  = get_rloc_len(rloc);
 140         u8 *dst = get_rloc_data(dest);
 141         void __user *src = (void __force __user *) addr;
 142
 143         if (!maxlen)
 144                 return;
 145
 146         ret = strncpy_from_user(dst, src, maxlen);
 147
 148         if (ret < 0) {  /* Failed to fetch string */
 149                 ((u8 *)get_rloc_data(dest))[0] = '\0';
 150                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 151         } else {
 152                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 153         }
 154 }
 155
 156 static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 157                                                       void *addr, void *dest)
 158 {
 159         int len;
 160         void __user *vaddr = (void __force __user *) addr;
 161
 162         len = strnlen_user(vaddr, MAX_STRING_SIZE);
 163
 164         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 165                 *(u32 *)dest = 0;
 166         else
 167                 *(u32 *)dest = len;
 168 }
 169
 170 /* uprobes do not support symbol fetch methods */
 171 #define fetch_symbol_u8                 NULL
 172 #define fetch_symbol_u16                NULL
 173 #define fetch_symbol_u32                NULL
 174 #define fetch_symbol_u64                NULL
 175 #define fetch_symbol_string             NULL
 176 #define fetch_symbol_string_size        NULL
 177
 178 /* Fetch type information table */
 179 const struct fetch_type uprobes_fetch_type_table[] = {
 180         /* Special types */
 181         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 182                                         sizeof(u32), 1, "__data_loc char[]"),
 183         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 184                                         string_size, sizeof(u32), 0, "u32"),
 185         /* Basic types */
 186         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 187         ASSIGN_FETCH_TYPE(u16, u16, 0),
 188         ASSIGN_FETCH_TYPE(u32, u32, 0),
 189         ASSIGN_FETCH_TYPE(u64, u64, 0),
 190         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 191         ASSIGN_FETCH_TYPE(s16, u16, 1),
 192         ASSIGN_FETCH_TYPE(s32, u32, 1),
 193         ASSIGN_FETCH_TYPE(s64, u64, 1),
 194
 195         ASSIGN_FETCH_TYPE_END
 196 };
 197
 198 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 199 {
 200         rwlock_init(&filter->rwlock);
 201         filter->nr_systemwide = 0;
 202         INIT_LIST_HEAD(&filter->perf_events);
 203 }
 204
 205 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 206 {
 207         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 208 }
 209
 210 static inline bool is_ret_probe(struct trace_uprobe *tu)
 211 {
 212         return tu->consumer.ret_handler != NULL;
 213 }
 214
 215 /*
 216  * Allocate new trace_uprobe and initialize it (including uprobes).
 217  */
 218 static struct trace_uprobe *
 219 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 220 {
 221         struct trace_uprobe *tu;
 222
 223         if (!event || !is_good_name(event))
 224                 return ERR_PTR(-EINVAL);
 225
 226         if (!group || !is_good_name(group))
 227                 return ERR_PTR(-EINVAL);
 228
 229         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 230         if (!tu)
 231                 return ERR_PTR(-ENOMEM);
 232
 233         tu->tp.call.class = &tu->tp.class;
 234         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 235         if (!tu->tp.call.name)
 236                 goto error;
 237
 238         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 239         if (!tu->tp.class.system)
 240                 goto error;
 241
 242         INIT_LIST_HEAD(&tu->list);
 243         tu->consumer.handler = uprobe_dispatcher;
 244         if (is_ret)
 245                 tu->consumer.ret_handler = uretprobe_dispatcher;
 246         init_trace_uprobe_filter(&tu->filter);
 247         tu->tp.call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
 248         return tu;
 249
 250 error:
 251         kfree(tu->tp.call.name);
 252         kfree(tu);
 253
 254         return ERR_PTR(-ENOMEM);
 255 }
 256
 257 static void free_trace_uprobe(struct trace_uprobe *tu)
 258 {
 259         int i;
 260
 261         for (i = 0; i < tu->tp.nr_args; i++)
 262                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 263
 264         iput(tu->inode);
 265         kfree(tu->tp.call.class->system);
 266         kfree(tu->tp.call.name);
 267         kfree(tu->filename);
 268         kfree(tu);
 269 }
 270
 271 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 272 {
 273         struct trace_uprobe *tu;
 274
 275         list_for_each_entry(tu, &uprobe_list, list)
 276                 if (strcmp(tu->tp.call.name, event) == 0 &&
 277                     strcmp(tu->tp.call.class->system, group) == 0)
 278                         return tu;
 279
 280         return NULL;
 281 }
 282
 283 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 284 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 285 {
 286         int ret;
 287
 288         ret = unregister_uprobe_event(tu);
 289         if (ret)
 290                 return ret;
 291
 292         list_del(&tu->list);
 293         free_trace_uprobe(tu);
 294         return 0;
 295 }
 296
 297 /* Register a trace_uprobe and probe_event */
 298 static int register_trace_uprobe(struct trace_uprobe *tu)
 299 {
 300         struct trace_uprobe *old_tu;
 301         int ret;
 302
 303         mutex_lock(&uprobe_lock);
 304
 305         /* register as an event */
 306         old_tu = find_probe_event(tu->tp.call.name, tu->tp.call.class->system);
 307         if (old_tu) {
 308                 /* delete old event */
 309                 ret = unregister_trace_uprobe(old_tu);
 310                 if (ret)
 311                         goto end;
 312         }
 313
 314         ret = register_uprobe_event(tu);
 315         if (ret) {
 316                 pr_warning("Failed to register probe event(%d)\n", ret);
 317                 goto end;
 318         }
 319
 320         list_add_tail(&tu->list, &uprobe_list);
 321
 322 end:
 323         mutex_unlock(&uprobe_lock);
 324
 325         return ret;
 326 }
 327
 328 /*
 329  * Argument syntax:
 330  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 331  *
 332  *  - Remove uprobe: -:[GRP/]EVENT
 333  */
 334 static int create_trace_uprobe(int argc, char **argv)
 335 {
 336         struct trace_uprobe *tu;
 337         struct inode *inode;
 338         char *arg, *event, *group, *filename;
 339         char buf[MAX_EVENT_NAME_LEN];
 340         struct path path;
 341         unsigned long offset;
 342         bool is_delete, is_return;
 343         int i, ret;
 344
 345         inode = NULL;
 346         ret = 0;
 347         is_delete = false;
 348         is_return = false;
 349         event = NULL;
 350         group = NULL;
 351
 352         /* argc must be >= 1 */
 353         if (argv[0][0] == '-')
 354                 is_delete = true;
 355         else if (argv[0][0] == 'r')
 356                 is_return = true;
 357         else if (argv[0][0] != 'p') {
 358                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 359                 return -EINVAL;
 360         }
 361
 362         if (argv[0][1] == ':') {
 363                 event = &argv[0][2];
 364                 arg = strchr(event, '/');
 365
 366                 if (arg) {
 367                         group = event;
 368                         event = arg + 1;
 369                         event[-1] = '\0';
 370
 371                         if (strlen(group) == 0) {
 372                                 pr_info("Group name is not specified\n");
 373                                 return -EINVAL;
 374                         }
 375                 }
 376                 if (strlen(event) == 0) {
 377                         pr_info("Event name is not specified\n");
 378                         return -EINVAL;
 379                 }
 380         }
 381         if (!group)
 382                 group = UPROBE_EVENT_SYSTEM;
 383
 384         if (is_delete) {
 385                 int ret;
 386
 387                 if (!event) {
 388                         pr_info("Delete command needs an event name.\n");
 389                         return -EINVAL;
 390                 }
 391                 mutex_lock(&uprobe_lock);
 392                 tu = find_probe_event(event, group);
 393
 394                 if (!tu) {
 395                         mutex_unlock(&uprobe_lock);
 396                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 397                         return -ENOENT;
 398                 }
 399                 /* delete an event */
 400                 ret = unregister_trace_uprobe(tu);
 401                 mutex_unlock(&uprobe_lock);
 402                 return ret;
 403         }
 404
 405         if (argc < 2) {
 406                 pr_info("Probe point is not specified.\n");
 407                 return -EINVAL;
 408         }
 409         if (isdigit(argv[1][0])) {
 410                 pr_info("probe point must be have a filename.\n");
 411                 return -EINVAL;
 412         }
 413         arg = strchr(argv[1], ':');
 414         if (!arg) {
 415                 ret = -EINVAL;
 416                 goto fail_address_parse;
 417         }
 418
 419         *arg++ = '\0';
 420         filename = argv[1];
 421         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 422         if (ret)
 423                 goto fail_address_parse;
 424
 425         inode = igrab(path.dentry->d_inode);
 426         path_put(&path);
 427
 428         if (!inode || !S_ISREG(inode->i_mode)) {
 429                 ret = -EINVAL;
 430                 goto fail_address_parse;
 431         }
 432
 433         ret = kstrtoul(arg, 0, &offset);
 434         if (ret)
 435                 goto fail_address_parse;
 436
 437         argc -= 2;
 438         argv += 2;
 439
 440         /* setup a probe */
 441         if (!event) {
 442                 char *tail;
 443                 char *ptr;
 444
 445                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 446                 if (!tail) {
 447                         ret = -ENOMEM;
 448                         goto fail_address_parse;
 449                 }
 450
 451                 ptr = strpbrk(tail, ".-_");
 452                 if (ptr)
 453                         *ptr = '\0';
 454
 455                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 456                 event = buf;
 457                 kfree(tail);
 458         }
 459
 460         tu = alloc_trace_uprobe(group, event, argc, is_return);
 461         if (IS_ERR(tu)) {
 462                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 463                 ret = PTR_ERR(tu);
 464                 goto fail_address_parse;
 465         }
 466         tu->offset = offset;
 467         tu->inode = inode;
 468         tu->filename = kstrdup(filename, GFP_KERNEL);
 469
 470         if (!tu->filename) {
 471                 pr_info("Failed to allocate filename.\n");
 472                 ret = -ENOMEM;
 473                 goto error;
 474         }
 475
 476         /* parse arguments */
 477         ret = 0;
 478         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 479                 struct probe_arg *parg = &tu->tp.args[i];
 480
 481                 /* Increment count for freeing args in error case */
 482                 tu->tp.nr_args++;
 483
 484                 /* Parse argument name */
 485                 arg = strchr(argv[i], '=');
 486                 if (arg) {
 487                         *arg++ = '\0';
 488                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 489                 } else {
 490                         arg = argv[i];
 491                         /* If argument name is omitted, set "argN" */
 492                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 493                         parg->name = kstrdup(buf, GFP_KERNEL);
 494                 }
 495
 496                 if (!parg->name) {
 497                         pr_info("Failed to allocate argument[%d] name.\n", i);
 498                         ret = -ENOMEM;
 499                         goto error;
 500                 }
 501
 502                 if (!is_good_name(parg->name)) {
 503                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 504                         ret = -EINVAL;
 505                         goto error;
 506                 }
 507
 508                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 509                         pr_info("Argument[%d] name '%s' conflicts with "
 510                                 "another field.\n", i, argv[i]);
 511                         ret = -EINVAL;
 512                         goto error;
 513                 }
 514
 515                 /* Parse fetch argument */
 516                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 517                                                  is_return, false);
 518                 if (ret) {
 519                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 520                         goto error;
 521                 }
 522         }
 523
 524         ret = register_trace_uprobe(tu);
 525         if (ret)
 526                 goto error;
 527         return 0;
 528
 529 error:
 530         free_trace_uprobe(tu);
 531         return ret;
 532
 533 fail_address_parse:
 534         if (inode)
 535                 iput(inode);
 536
 537         pr_info("Failed to parse address or file.\n");
 538
 539         return ret;
 540 }
 541
 542 static int cleanup_all_probes(void)
 543 {
 544         struct trace_uprobe *tu;
 545         int ret = 0;
 546
 547         mutex_lock(&uprobe_lock);
 548         while (!list_empty(&uprobe_list)) {
 549                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 550                 ret = unregister_trace_uprobe(tu);
 551                 if (ret)
 552                         break;
 553         }
 554         mutex_unlock(&uprobe_lock);
 555         return ret;
 556 }
 557
 558 /* Probes listing interfaces */
 559 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 560 {
 561         mutex_lock(&uprobe_lock);
 562         return seq_list_start(&uprobe_list, *pos);
 563 }
 564
 565 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 566 {
 567         return seq_list_next(v, &uprobe_list, pos);
 568 }
 569
 570 static void probes_seq_stop(struct seq_file *m, void *v)
 571 {
 572         mutex_unlock(&uprobe_lock);
 573 }
 574
 575 static int probes_seq_show(struct seq_file *m, void *v)
 576 {
 577         struct trace_uprobe *tu = v;
 578         char c = is_ret_probe(tu) ? 'r' : 'p';
 579         int i;
 580
 581         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system, tu->tp.call.name);
 582         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 583
 584         for (i = 0; i < tu->tp.nr_args; i++)
 585                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 586
 587         seq_printf(m, "\n");
 588         return 0;
 589 }
 590
 591 static const struct seq_operations probes_seq_op = {
 592         .start  = probes_seq_start,
 593         .next   = probes_seq_next,
 594         .stop   = probes_seq_stop,
 595         .show   = probes_seq_show
 596 };
 597
 598 static int probes_open(struct inode *inode, struct file *file)
 599 {
 600         int ret;
 601
 602         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 603                 ret = cleanup_all_probes();
 604                 if (ret)
 605                         return ret;
 606         }
 607
 608         return seq_open(file, &probes_seq_op);
 609 }
 610
 611 static ssize_t probes_write(struct file *file, const char __user *buffer,
 612                             size_t count, loff_t *ppos)
 613 {
 614         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 615 }
 616
 617 static const struct file_operations uprobe_events_ops = {
 618         .owner          = THIS_MODULE,
 619         .open           = probes_open,
 620         .read           = seq_read,
 621         .llseek         = seq_lseek,
 622         .release        = seq_release,
 623         .write          = probes_write,
 624 };
 625
 626 /* Probes profiling interfaces */
 627 static int probes_profile_seq_show(struct seq_file *m, void *v)
 628 {
 629         struct trace_uprobe *tu = v;
 630
 631         seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->tp.call.name, tu->nhit);
 632         return 0;
 633 }
 634
 635 static const struct seq_operations profile_seq_op = {
 636         .start  = probes_seq_start,
 637         .next   = probes_seq_next,
 638         .stop   = probes_seq_stop,
 639         .show   = probes_profile_seq_show
 640 };
 641
 642 static int profile_open(struct inode *inode, struct file *file)
 643 {
 644         return seq_open(file, &profile_seq_op);
 645 }
 646
 647 static const struct file_operations uprobe_profile_ops = {
 648         .owner          = THIS_MODULE,
 649         .open           = profile_open,
 650         .read           = seq_read,
 651         .llseek         = seq_lseek,
 652         .release        = seq_release,
 653 };
 654
 655 static void uprobe_trace_print(struct trace_uprobe *tu,
 656                                 unsigned long func, struct pt_regs *regs)
 657 {
 658         struct uprobe_trace_entry_head *entry;
 659         struct ring_buffer_event *event;
 660         struct ring_buffer *buffer;
 661         void *data;
 662         int size, i;
 663         struct ftrace_event_call *call = &tu->tp.call;
 664
 665         size = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 666         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 667                                                   size + tu->tp.size, 0, 0);
 668         if (!event)
 669                 return;
 670
 671         entry = ring_buffer_event_data(event);
 672         if (is_ret_probe(tu)) {
 673                 entry->vaddr[0] = func;
 674                 entry->vaddr[1] = instruction_pointer(regs);
 675                 data = DATAOF_TRACE_ENTRY(entry, true);
 676         } else {
 677                 entry->vaddr[0] = instruction_pointer(regs);
 678                 data = DATAOF_TRACE_ENTRY(entry, false);
 679         }
 680
 681         for (i = 0; i < tu->tp.nr_args; i++) {
 682                 call_fetch(&tu->tp.args[i].fetch, regs,
 683                            data + tu->tp.args[i].offset);
 684         }
 685
 686         if (!call_filter_check_discard(call, entry, buffer, event))
 687                 trace_buffer_unlock_commit(buffer, event, 0, 0);
 688 }
 689
 690 /* uprobe handler */
 691 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
 692 {
 693         if (!is_ret_probe(tu))
 694                 uprobe_trace_print(tu, 0, regs);
 695         return 0;
 696 }
 697
 698 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 699                                 struct pt_regs *regs)
 700 {
 701         uprobe_trace_print(tu, func, regs);
 702 }
 703
 704 /* Event entry printers */
 705 static enum print_line_t
 706 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 707 {
 708         struct uprobe_trace_entry_head *entry;
 709         struct trace_seq *s = &iter->seq;
 710         struct trace_uprobe *tu;
 711         u8 *data;
 712         int i;
 713
 714         entry = (struct uprobe_trace_entry_head *)iter->ent;
 715         tu = container_of(event, struct trace_uprobe, tp.call.event);
 716
 717         if (is_ret_probe(tu)) {
 718                 if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)", tu->tp.call.name,
 719                                         entry->vaddr[1], entry->vaddr[0]))
 720                         goto partial;
 721                 data = DATAOF_TRACE_ENTRY(entry, true);
 722         } else {
 723                 if (!trace_seq_printf(s, "%s: (0x%lx)", tu->tp.call.name,
 724                                         entry->vaddr[0]))
 725                         goto partial;
 726                 data = DATAOF_TRACE_ENTRY(entry, false);
 727         }
 728
 729         for (i = 0; i < tu->tp.nr_args; i++) {
 730                 struct probe_arg *parg = &tu->tp.args[i];
 731
 732                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 733                         goto partial;
 734         }
 735
 736         if (trace_seq_puts(s, "\n"))
 737                 return TRACE_TYPE_HANDLED;
 738
 739 partial:
 740         return TRACE_TYPE_PARTIAL_LINE;
 741 }
 742
 743 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 744                                 enum uprobe_filter_ctx ctx,
 745                                 struct mm_struct *mm);
 746
 747 static int
 748 probe_event_enable(struct trace_uprobe *tu, int flag, filter_func_t filter)
 749 {
 750         int ret = 0;
 751
 752         if (trace_probe_is_enabled(&tu->tp))
 753                 return -EINTR;
 754
 755         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 756
 757         tu->tp.flags |= flag;
 758         tu->consumer.filter = filter;
 759         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 760         if (ret)
 761                 tu->tp.flags &= ~flag;
 762
 763         return ret;
 764 }
 765
 766 static void probe_event_disable(struct trace_uprobe *tu, int flag)
 767 {
 768         if (!trace_probe_is_enabled(&tu->tp))
 769                 return;
 770
 771         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 772
 773         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 774         tu->tp.flags &= ~flag;
 775 }
 776
 777 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 778 {
 779         int ret, i, size;
 780         struct uprobe_trace_entry_head field;
 781         struct trace_uprobe *tu = event_call->data;
 782
 783         if (is_ret_probe(tu)) {
 784                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 785                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 786                 size = SIZEOF_TRACE_ENTRY(true);
 787         } else {
 788                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 789                 size = SIZEOF_TRACE_ENTRY(false);
 790         }
 791         /* Set argument names as fields */
 792         for (i = 0; i < tu->tp.nr_args; i++) {
 793                 struct probe_arg *parg = &tu->tp.args[i];
 794
 795                 ret = trace_define_field(event_call, parg->type->fmttype,
 796                                          parg->name, size + parg->offset,
 797                                          parg->type->size, parg->type->is_signed,
 798                                          FILTER_OTHER);
 799
 800                 if (ret)
 801                         return ret;
 802         }
 803         return 0;
 804 }
 805
 806 #ifdef CONFIG_PERF_EVENTS
 807 static bool
 808 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 809 {
 810         struct perf_event *event;
 811
 812         if (filter->nr_systemwide)
 813                 return true;
 814
 815         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
 816                 if (event->hw.tp_target->mm == mm)
 817                         return true;
 818         }
 819
 820         return false;
 821 }
 822
 823 static inline bool
 824 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 825 {
 826         return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 827 }
 828
 829 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 830 {
 831         bool done;
 832
 833         write_lock(&tu->filter.rwlock);
 834         if (event->hw.tp_target) {
 835                 /*
 836                  * event->parent != NULL means copy_process(), we can avoid
 837                  * uprobe_apply(). current->mm must be probed and we can rely
 838                  * on dup_mmap() which preserves the already installed bp's.
 839                  *
 840                  * attr.enable_on_exec means that exec/mmap will install the
 841                  * breakpoints we need.
 842                  */
 843                 done = tu->filter.nr_systemwide ||
 844                         event->parent || event->attr.enable_on_exec ||
 845                         uprobe_filter_event(tu, event);
 846                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
 847         } else {
 848                 done = tu->filter.nr_systemwide;
 849                 tu->filter.nr_systemwide++;
 850         }
 851         write_unlock(&tu->filter.rwlock);
 852
 853         if (!done)
 854                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
 855
 856         return 0;
 857 }
 858
 859 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 860 {
 861         bool done;
 862
 863         write_lock(&tu->filter.rwlock);
 864         if (event->hw.tp_target) {
 865                 list_del(&event->hw.tp_list);
 866                 done = tu->filter.nr_systemwide ||
 867                         (event->hw.tp_target->flags & PF_EXITING) ||
 868                         uprobe_filter_event(tu, event);
 869         } else {
 870                 tu->filter.nr_systemwide--;
 871                 done = tu->filter.nr_systemwide;
 872         }
 873         write_unlock(&tu->filter.rwlock);
 874
 875         if (!done)
 876                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 877
 878         return 0;
 879 }
 880
 881 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 882                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
 883 {
 884         struct trace_uprobe *tu;
 885         int ret;
 886
 887         tu = container_of(uc, struct trace_uprobe, consumer);
 888         read_lock(&tu->filter.rwlock);
 889         ret = __uprobe_perf_filter(&tu->filter, mm);
 890         read_unlock(&tu->filter.rwlock);
 891
 892         return ret;
 893 }
 894
 895 static void uprobe_perf_print(struct trace_uprobe *tu,
 896                                 unsigned long func, struct pt_regs *regs)
 897 {
 898         struct ftrace_event_call *call = &tu->tp.call;
 899         struct uprobe_trace_entry_head *entry;
 900         struct hlist_head *head;
 901         void *data;
 902         int size, rctx, i;
 903
 904         size = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 905         size = ALIGN(size + tu->tp.size + sizeof(u32), sizeof(u64)) - sizeof(u32);
 906
 907         preempt_disable();
 908         head = this_cpu_ptr(call->perf_events);
 909         if (hlist_empty(head))
 910                 goto out;
 911
 912         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
 913         if (!entry)
 914                 goto out;
 915
 916         if (is_ret_probe(tu)) {
 917                 entry->vaddr[0] = func;
 918                 entry->vaddr[1] = instruction_pointer(regs);
 919                 data = DATAOF_TRACE_ENTRY(entry, true);
 920         } else {
 921                 entry->vaddr[0] = instruction_pointer(regs);
 922                 data = DATAOF_TRACE_ENTRY(entry, false);
 923         }
 924
 925         for (i = 0; i < tu->tp.nr_args; i++) {
 926                 struct probe_arg *parg = &tu->tp.args[i];
 927
 928                 call_fetch(&parg->fetch, regs, data + parg->offset);
 929         }
 930
 931         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
 932  out:
 933         preempt_enable();
 934 }
 935
 936 /* uprobe profile handler */
 937 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
 938 {
 939         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
 940                 return UPROBE_HANDLER_REMOVE;
 941
 942         if (!is_ret_probe(tu))
 943                 uprobe_perf_print(tu, 0, regs);
 944         return 0;
 945 }
 946
 947 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
 948                                 struct pt_regs *regs)
 949 {
 950         uprobe_perf_print(tu, func, regs);
 951 }
 952 #endif  /* CONFIG_PERF_EVENTS */
 953
 954 static
 955 int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
 956 {
 957         struct trace_uprobe *tu = event->data;
 958
 959         switch (type) {
 960         case TRACE_REG_REGISTER:
 961                 return probe_event_enable(tu, TP_FLAG_TRACE, NULL);
 962
 963         case TRACE_REG_UNREGISTER:
 964                 probe_event_disable(tu, TP_FLAG_TRACE);
 965                 return 0;
 966
 967 #ifdef CONFIG_PERF_EVENTS
 968         case TRACE_REG_PERF_REGISTER:
 969                 return probe_event_enable(tu, TP_FLAG_PROFILE, uprobe_perf_filter);
 970
 971         case TRACE_REG_PERF_UNREGISTER:
 972                 probe_event_disable(tu, TP_FLAG_PROFILE);
 973                 return 0;
 974
 975         case TRACE_REG_PERF_OPEN:
 976                 return uprobe_perf_open(tu, data);
 977
 978         case TRACE_REG_PERF_CLOSE:
 979                 return uprobe_perf_close(tu, data);
 980
 981 #endif
 982         default:
 983                 return 0;
 984         }
 985         return 0;
 986 }
 987
 988 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
 989 {
 990         struct trace_uprobe *tu;
 991         int ret = 0;
 992
 993         tu = container_of(con, struct trace_uprobe, consumer);
 994         tu->nhit++;
 995
 996         if (tu->tp.flags & TP_FLAG_TRACE)
 997                 ret |= uprobe_trace_func(tu, regs);
 998
 999 #ifdef CONFIG_PERF_EVENTS
1000         if (tu->tp.flags & TP_FLAG_PROFILE)
1001                 ret |= uprobe_perf_func(tu, regs);
1002 #endif
1003         return ret;
1004 }
1005
1006 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1007                                 unsigned long func, struct pt_regs *regs)
1008 {
1009         struct trace_uprobe *tu;
1010
1011         tu = container_of(con, struct trace_uprobe, consumer);
1012
1013         if (tu->tp.flags & TP_FLAG_TRACE)
1014                 uretprobe_trace_func(tu, func, regs);
1015
1016 #ifdef CONFIG_PERF_EVENTS
1017         if (tu->tp.flags & TP_FLAG_PROFILE)
1018                 uretprobe_perf_func(tu, func, regs);
1019 #endif
1020         return 0;
1021 }
1022
1023 static struct trace_event_functions uprobe_funcs = {
1024         .trace          = print_uprobe_event
1025 };
1026
1027 static int register_uprobe_event(struct trace_uprobe *tu)
1028 {
1029         struct ftrace_event_call *call = &tu->tp.call;
1030         int ret;
1031
1032         /* Initialize ftrace_event_call */
1033         INIT_LIST_HEAD(&call->class->fields);
1034         call->event.funcs = &uprobe_funcs;
1035         call->class->define_fields = uprobe_event_define_fields;
1036
1037         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1038                 return -ENOMEM;
1039
1040         ret = register_ftrace_event(&call->event);
1041         if (!ret) {
1042                 kfree(call->print_fmt);
1043                 return -ENODEV;
1044         }
1045         call->flags = 0;
1046         call->class->reg = trace_uprobe_register;
1047         call->data = tu;
1048         ret = trace_add_event_call(call);
1049
1050         if (ret) {
1051                 pr_info("Failed to register uprobe event: %s\n", call->name);
1052                 kfree(call->print_fmt);
1053                 unregister_ftrace_event(&call->event);
1054         }
1055
1056         return ret;
1057 }
1058
1059 static int unregister_uprobe_event(struct trace_uprobe *tu)
1060 {
1061         int ret;
1062
1063         /* tu->event is unregistered in trace_remove_event_call() */
1064         ret = trace_remove_event_call(&tu->tp.call);
1065         if (ret)
1066                 return ret;
1067         kfree(tu->tp.call.print_fmt);
1068         tu->tp.call.print_fmt = NULL;
1069         return 0;
1070 }
1071
1072 /* Make a trace interface for controling probe points */
1073 static __init int init_uprobe_trace(void)
1074 {
1075         struct dentry *d_tracer;
1076
1077         d_tracer = tracing_init_dentry();
1078         if (!d_tracer)
1079                 return 0;
1080
1081         trace_create_file("uprobe_events", 0644, d_tracer,
1082                                     NULL, &uprobe_events_ops);
1083         /* Profile interface */
1084         trace_create_file("uprobe_profile", 0444, d_tracer,
1085                                     NULL, &uprobe_profile_ops);
1086         return 0;
1087 }
1088
1089 fs_initcall(init_uprobe_trace);