kernel/trace/trace_uprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * uprobes-based tracing events
   4  *
   5  * Copyright (C) IBM Corporation, 2010-2012
   6  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
   7  */
   8 #define pr_fmt(fmt)     "trace_uprobe: " fmt
   9
  10 #include <linux/ctype.h>
  11 #include <linux/module.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/uprobes.h>
  14 #include <linux/namei.h>
  15 #include <linux/string.h>
  16 #include <linux/rculist.h>
  17
  18 #include "trace_dynevent.h"
  19 #include "trace_probe.h"
  20 #include "trace_probe_tmpl.h"
  21
  22 #define UPROBE_EVENT_SYSTEM     "uprobes"
  23
  24 struct uprobe_trace_entry_head {
  25         struct trace_entry      ent;
  26         unsigned long           vaddr[];
  27 };
  28
  29 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  30         (sizeof(struct uprobe_trace_entry_head) +       \
  31          sizeof(unsigned long) * (is_return ? 2 : 1))
  32
  33 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  34         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  35
  36 struct trace_uprobe_filter {
  37         rwlock_t                rwlock;
  38         int                     nr_systemwide;
  39         struct list_head        perf_events;
  40 };
  41
  42 static int trace_uprobe_create(int argc, const char **argv);
  43 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  44 static int trace_uprobe_release(struct dyn_event *ev);
  45 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  46 static bool trace_uprobe_match(const char *system, const char *event,
  47                                struct dyn_event *ev);
  48
  49 static struct dyn_event_operations trace_uprobe_ops = {
  50         .create = trace_uprobe_create,
  51         .show = trace_uprobe_show,
  52         .is_busy = trace_uprobe_is_busy,
  53         .free = trace_uprobe_release,
  54         .match = trace_uprobe_match,
  55 };
  56
  57 /*
  58  * uprobe event core functions
  59  */
  60 struct trace_uprobe {
  61         struct dyn_event                devent;
  62         struct trace_uprobe_filter      filter;
  63         struct uprobe_consumer          consumer;
  64         struct path                     path;
  65         struct inode                    *inode;
  66         char                            *filename;
  67         unsigned long                   offset;
  68         unsigned long                   ref_ctr_offset;
  69         unsigned long                   nhit;
  70         struct trace_probe              tp;
  71 };
  72
  73 static bool is_trace_uprobe(struct dyn_event *ev)
  74 {
  75         return ev->ops == &trace_uprobe_ops;
  76 }
  77
  78 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  79 {
  80         return container_of(ev, struct trace_uprobe, devent);
  81 }
  82
  83 /**
  84  * for_each_trace_uprobe - iterate over the trace_uprobe list
  85  * @pos:        the struct trace_uprobe * for each entry
  86  * @dpos:       the struct dyn_event * to use as a loop cursor
  87  */
  88 #define for_each_trace_uprobe(pos, dpos)        \
  89         for_each_dyn_event(dpos)                \
  90                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  91
  92 #define SIZEOF_TRACE_UPROBE(n)                          \
  93         (offsetof(struct trace_uprobe, tp.args) +       \
  94         (sizeof(struct probe_arg) * (n)))
  95
  96 static int register_uprobe_event(struct trace_uprobe *tu);
  97 static int unregister_uprobe_event(struct trace_uprobe *tu);
  98
  99 struct uprobe_dispatch_data {
 100         struct trace_uprobe     *tu;
 101         unsigned long           bp_addr;
 102 };
 103
 104 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
 105 static int uretprobe_dispatcher(struct uprobe_consumer *con,
 106                                 unsigned long func, struct pt_regs *regs);
 107
 108 #ifdef CONFIG_STACK_GROWSUP
 109 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 110 {
 111         return addr - (n * sizeof(long));
 112 }
 113 #else
 114 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 115 {
 116         return addr + (n * sizeof(long));
 117 }
 118 #endif
 119
 120 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 121 {
 122         unsigned long ret;
 123         unsigned long addr = user_stack_pointer(regs);
 124
 125         addr = adjust_stack_addr(addr, n);
 126
 127         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 128                 return 0;
 129
 130         return ret;
 131 }
 132
 133 /*
 134  * Uprobes-specific fetch functions
 135  */
 136 static nokprobe_inline int
 137 probe_mem_read(void *dest, void *src, size_t size)
 138 {
 139         void __user *vaddr = (void __force __user *)src;
 140
 141         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 142 }
 143
 144 static nokprobe_inline int
 145 probe_mem_read_user(void *dest, void *src, size_t size)
 146 {
 147         return probe_mem_read(dest, src, size);
 148 }
 149
 150 /*
 151  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 152  * length and relative data location.
 153  */
 154 static nokprobe_inline int
 155 fetch_store_string(unsigned long addr, void *dest, void *base)
 156 {
 157         long ret;
 158         u32 loc = *(u32 *)dest;
 159         int maxlen  = get_loc_len(loc);
 160         u8 *dst = get_loc_data(dest, base);
 161         void __user *src = (void __force __user *) addr;
 162
 163         if (unlikely(!maxlen))
 164                 return -ENOMEM;
 165
 166         if (addr == FETCH_TOKEN_COMM)
 167                 ret = strlcpy(dst, current->comm, maxlen);
 168         else
 169                 ret = strncpy_from_user(dst, src, maxlen);
 170         if (ret >= 0) {
 171                 if (ret == maxlen)
 172                         dst[ret - 1] = '\0';
 173                 else
 174                         /*
 175                          * Include the terminating null byte. In this case it
 176                          * was copied by strncpy_from_user but not accounted
 177                          * for in ret.
 178                          */
 179                         ret++;
 180                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 181         }
 182
 183         return ret;
 184 }
 185
 186 static nokprobe_inline int
 187 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 188 {
 189         return fetch_store_string(addr, dest, base);
 190 }
 191
 192 /* Return the length of string -- including null terminal byte */
 193 static nokprobe_inline int
 194 fetch_store_strlen(unsigned long addr)
 195 {
 196         int len;
 197         void __user *vaddr = (void __force __user *) addr;
 198
 199         if (addr == FETCH_TOKEN_COMM)
 200                 len = strlen(current->comm) + 1;
 201         else
 202                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 203
 204         return (len > MAX_STRING_SIZE) ? 0 : len;
 205 }
 206
 207 static nokprobe_inline int
 208 fetch_store_strlen_user(unsigned long addr)
 209 {
 210         return fetch_store_strlen(addr);
 211 }
 212
 213 static unsigned long translate_user_vaddr(unsigned long file_offset)
 214 {
 215         unsigned long base_addr;
 216         struct uprobe_dispatch_data *udd;
 217
 218         udd = (void *) current->utask->vaddr;
 219
 220         base_addr = udd->bp_addr - udd->tu->offset;
 221         return base_addr + file_offset;
 222 }
 223
 224 /* Note that we don't verify it, since the code does not come from user space */
 225 static int
 226 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 227                    void *base)
 228 {
 229         unsigned long val;
 230
 231         /* 1st stage: get value from context */
 232         switch (code->op) {
 233         case FETCH_OP_REG:
 234                 val = regs_get_register(regs, code->param);
 235                 break;
 236         case FETCH_OP_STACK:
 237                 val = get_user_stack_nth(regs, code->param);
 238                 break;
 239         case FETCH_OP_STACKP:
 240                 val = user_stack_pointer(regs);
 241                 break;
 242         case FETCH_OP_RETVAL:
 243                 val = regs_return_value(regs);
 244                 break;
 245         case FETCH_OP_IMM:
 246                 val = code->immediate;
 247                 break;
 248         case FETCH_OP_COMM:
 249                 val = FETCH_TOKEN_COMM;
 250                 break;
 251         case FETCH_OP_FOFFS:
 252                 val = translate_user_vaddr(code->immediate);
 253                 break;
 254         default:
 255                 return -EILSEQ;
 256         }
 257         code++;
 258
 259         return process_fetch_insn_bottom(code, val, dest, base);
 260 }
 261 NOKPROBE_SYMBOL(process_fetch_insn)
 262
 263 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 264 {
 265         rwlock_init(&filter->rwlock);
 266         filter->nr_systemwide = 0;
 267         INIT_LIST_HEAD(&filter->perf_events);
 268 }
 269
 270 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 271 {
 272         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 273 }
 274
 275 static inline bool is_ret_probe(struct trace_uprobe *tu)
 276 {
 277         return tu->consumer.ret_handler != NULL;
 278 }
 279
 280 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 281 {
 282         struct trace_uprobe *tu = to_trace_uprobe(ev);
 283
 284         return trace_probe_is_enabled(&tu->tp);
 285 }
 286
 287 static bool trace_uprobe_match(const char *system, const char *event,
 288                                struct dyn_event *ev)
 289 {
 290         struct trace_uprobe *tu = to_trace_uprobe(ev);
 291
 292         return strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 293                 (!system || strcmp(tu->tp.call.class->system, system) == 0);
 294 }
 295
 296 /*
 297  * Allocate new trace_uprobe and initialize it (including uprobes).
 298  */
 299 static struct trace_uprobe *
 300 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 301 {
 302         struct trace_uprobe *tu;
 303         int ret;
 304
 305         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 306         if (!tu)
 307                 return ERR_PTR(-ENOMEM);
 308
 309         ret = trace_probe_init(&tu->tp, event, group);
 310         if (ret < 0)
 311                 goto error;
 312
 313         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 314         tu->consumer.handler = uprobe_dispatcher;
 315         if (is_ret)
 316                 tu->consumer.ret_handler = uretprobe_dispatcher;
 317         init_trace_uprobe_filter(&tu->filter);
 318         return tu;
 319
 320 error:
 321         kfree(tu);
 322
 323         return ERR_PTR(ret);
 324 }
 325
 326 static void free_trace_uprobe(struct trace_uprobe *tu)
 327 {
 328         if (!tu)
 329                 return;
 330
 331         path_put(&tu->path);
 332         trace_probe_cleanup(&tu->tp);
 333         kfree(tu->filename);
 334         kfree(tu);
 335 }
 336
 337 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 338 {
 339         struct dyn_event *pos;
 340         struct trace_uprobe *tu;
 341
 342         for_each_trace_uprobe(tu, pos)
 343                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 344                     strcmp(tu->tp.call.class->system, group) == 0)
 345                         return tu;
 346
 347         return NULL;
 348 }
 349
 350 /* Unregister a trace_uprobe and probe_event */
 351 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 352 {
 353         int ret;
 354
 355         ret = unregister_uprobe_event(tu);
 356         if (ret)
 357                 return ret;
 358
 359         dyn_event_remove(&tu->devent);
 360         free_trace_uprobe(tu);
 361         return 0;
 362 }
 363
 364 /*
 365  * Uprobe with multiple reference counter is not allowed. i.e.
 366  * If inode and offset matches, reference counter offset *must*
 367  * match as well. Though, there is one exception: If user is
 368  * replacing old trace_uprobe with new one(same group/event),
 369  * then we allow same uprobe with new reference counter as far
 370  * as the new one does not conflict with any other existing
 371  * ones.
 372  */
 373 static struct trace_uprobe *find_old_trace_uprobe(struct trace_uprobe *new)
 374 {
 375         struct dyn_event *pos;
 376         struct trace_uprobe *tmp, *old = NULL;
 377         struct inode *new_inode = d_real_inode(new->path.dentry);
 378
 379         old = find_probe_event(trace_event_name(&new->tp.call),
 380                                 new->tp.call.class->system);
 381
 382         for_each_trace_uprobe(tmp, pos) {
 383                 if ((old ? old != tmp : true) &&
 384                     new_inode == d_real_inode(tmp->path.dentry) &&
 385                     new->offset == tmp->offset &&
 386                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 387                         pr_warn("Reference counter offset mismatch.");
 388                         return ERR_PTR(-EINVAL);
 389                 }
 390         }
 391         return old;
 392 }
 393
 394 /* Register a trace_uprobe and probe_event */
 395 static int register_trace_uprobe(struct trace_uprobe *tu)
 396 {
 397         struct trace_uprobe *old_tu;
 398         int ret;
 399
 400         mutex_lock(&event_mutex);
 401
 402         /* register as an event */
 403         old_tu = find_old_trace_uprobe(tu);
 404         if (IS_ERR(old_tu)) {
 405                 ret = PTR_ERR(old_tu);
 406                 goto end;
 407         }
 408
 409         if (old_tu) {
 410                 /* delete old event */
 411                 ret = unregister_trace_uprobe(old_tu);
 412                 if (ret)
 413                         goto end;
 414         }
 415
 416         ret = register_uprobe_event(tu);
 417         if (ret) {
 418                 pr_warn("Failed to register probe event(%d)\n", ret);
 419                 goto end;
 420         }
 421
 422         dyn_event_add(&tu->devent);
 423
 424 end:
 425         mutex_unlock(&event_mutex);
 426
 427         return ret;
 428 }
 429
 430 /*
 431  * Argument syntax:
 432  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 433  *
 434  *  - Remove uprobe: -:[GRP/]EVENT
 435  */
 436 static int trace_uprobe_create(int argc, const char **argv)
 437 {
 438         struct trace_uprobe *tu;
 439         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 440         char *arg, *filename, *rctr, *rctr_end, *tmp;
 441         char buf[MAX_EVENT_NAME_LEN];
 442         struct path path;
 443         unsigned long offset, ref_ctr_offset;
 444         bool is_return = false;
 445         int i, ret;
 446
 447         ret = 0;
 448         ref_ctr_offset = 0;
 449
 450         /* argc must be >= 1 */
 451         if (argv[0][0] == 'r')
 452                 is_return = true;
 453         else if (argv[0][0] != 'p' || argc < 2)
 454                 return -ECANCELED;
 455
 456         if (argv[0][1] == ':')
 457                 event = &argv[0][2];
 458
 459         if (!strchr(argv[1], '/'))
 460                 return -ECANCELED;
 461
 462         filename = kstrdup(argv[1], GFP_KERNEL);
 463         if (!filename)
 464                 return -ENOMEM;
 465
 466         /* Find the last occurrence, in case the path contains ':' too. */
 467         arg = strrchr(filename, ':');
 468         if (!arg || !isdigit(arg[1])) {
 469                 kfree(filename);
 470                 return -ECANCELED;
 471         }
 472
 473         trace_probe_log_init("trace_uprobe", argc, argv);
 474         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 475
 476         *arg++ = '\0';
 477         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 478         if (ret) {
 479                 trace_probe_log_err(0, FILE_NOT_FOUND);
 480                 kfree(filename);
 481                 trace_probe_log_clear();
 482                 return ret;
 483         }
 484         if (!d_is_reg(path.dentry)) {
 485                 trace_probe_log_err(0, NO_REGULAR_FILE);
 486                 ret = -EINVAL;
 487                 goto fail_address_parse;
 488         }
 489
 490         /* Parse reference counter offset if specified. */
 491         rctr = strchr(arg, '(');
 492         if (rctr) {
 493                 rctr_end = strchr(rctr, ')');
 494                 if (!rctr_end) {
 495                         ret = -EINVAL;
 496                         rctr_end = rctr + strlen(rctr);
 497                         trace_probe_log_err(rctr_end - filename,
 498                                             REFCNT_OPEN_BRACE);
 499                         goto fail_address_parse;
 500                 } else if (rctr_end[1] != '\0') {
 501                         ret = -EINVAL;
 502                         trace_probe_log_err(rctr_end + 1 - filename,
 503                                             BAD_REFCNT_SUFFIX);
 504                         goto fail_address_parse;
 505                 }
 506
 507                 *rctr++ = '\0';
 508                 *rctr_end = '\0';
 509                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 510                 if (ret) {
 511                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 512                         goto fail_address_parse;
 513                 }
 514         }
 515
 516         /* Parse uprobe offset. */
 517         ret = kstrtoul(arg, 0, &offset);
 518         if (ret) {
 519                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 520                 goto fail_address_parse;
 521         }
 522
 523         /* setup a probe */
 524         trace_probe_log_set_index(0);
 525         if (event) {
 526                 ret = traceprobe_parse_event_name(&event, &group, buf,
 527                                                   event - argv[0]);
 528                 if (ret)
 529                         goto fail_address_parse;
 530         } else {
 531                 char *tail;
 532                 char *ptr;
 533
 534                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 535                 if (!tail) {
 536                         ret = -ENOMEM;
 537                         goto fail_address_parse;
 538                 }
 539
 540                 ptr = strpbrk(tail, ".-_");
 541                 if (ptr)
 542                         *ptr = '\0';
 543
 544                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 545                 event = buf;
 546                 kfree(tail);
 547         }
 548
 549         argc -= 2;
 550         argv += 2;
 551
 552         tu = alloc_trace_uprobe(group, event, argc, is_return);
 553         if (IS_ERR(tu)) {
 554                 ret = PTR_ERR(tu);
 555                 /* This must return -ENOMEM otherwise there is a bug */
 556                 WARN_ON_ONCE(ret != -ENOMEM);
 557                 goto fail_address_parse;
 558         }
 559         tu->offset = offset;
 560         tu->ref_ctr_offset = ref_ctr_offset;
 561         tu->path = path;
 562         tu->filename = filename;
 563
 564         /* parse arguments */
 565         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 566                 tmp = kstrdup(argv[i], GFP_KERNEL);
 567                 if (!tmp) {
 568                         ret = -ENOMEM;
 569                         goto error;
 570                 }
 571
 572                 trace_probe_log_set_index(i + 2);
 573                 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 574                                         is_return ? TPARG_FL_RETURN : 0);
 575                 kfree(tmp);
 576                 if (ret)
 577                         goto error;
 578         }
 579
 580         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 581         if (ret < 0)
 582                 goto error;
 583
 584         ret = register_trace_uprobe(tu);
 585         if (!ret)
 586                 goto out;
 587
 588 error:
 589         free_trace_uprobe(tu);
 590 out:
 591         trace_probe_log_clear();
 592         return ret;
 593
 594 fail_address_parse:
 595         trace_probe_log_clear();
 596         path_put(&path);
 597         kfree(filename);
 598
 599         return ret;
 600 }
 601
 602 static int create_or_delete_trace_uprobe(int argc, char **argv)
 603 {
 604         int ret;
 605
 606         if (argv[0][0] == '-')
 607                 return dyn_event_release(argc, argv, &trace_uprobe_ops);
 608
 609         ret = trace_uprobe_create(argc, (const char **)argv);
 610         return ret == -ECANCELED ? -EINVAL : ret;
 611 }
 612
 613 static int trace_uprobe_release(struct dyn_event *ev)
 614 {
 615         struct trace_uprobe *tu = to_trace_uprobe(ev);
 616
 617         return unregister_trace_uprobe(tu);
 618 }
 619
 620 /* Probes listing interfaces */
 621 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 622 {
 623         struct trace_uprobe *tu = to_trace_uprobe(ev);
 624         char c = is_ret_probe(tu) ? 'r' : 'p';
 625         int i;
 626
 627         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, tu->tp.call.class->system,
 628                         trace_event_name(&tu->tp.call), tu->filename,
 629                         (int)(sizeof(void *) * 2), tu->offset);
 630
 631         if (tu->ref_ctr_offset)
 632                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 633
 634         for (i = 0; i < tu->tp.nr_args; i++)
 635                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 636
 637         seq_putc(m, '\n');
 638         return 0;
 639 }
 640
 641 static int probes_seq_show(struct seq_file *m, void *v)
 642 {
 643         struct dyn_event *ev = v;
 644
 645         if (!is_trace_uprobe(ev))
 646                 return 0;
 647
 648         return trace_uprobe_show(m, ev);
 649 }
 650
 651 static const struct seq_operations probes_seq_op = {
 652         .start  = dyn_event_seq_start,
 653         .next   = dyn_event_seq_next,
 654         .stop   = dyn_event_seq_stop,
 655         .show   = probes_seq_show
 656 };
 657
 658 static int probes_open(struct inode *inode, struct file *file)
 659 {
 660         int ret;
 661
 662         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 663                 ret = dyn_events_release_all(&trace_uprobe_ops);
 664                 if (ret)
 665                         return ret;
 666         }
 667
 668         return seq_open(file, &probes_seq_op);
 669 }
 670
 671 static ssize_t probes_write(struct file *file, const char __user *buffer,
 672                             size_t count, loff_t *ppos)
 673 {
 674         return trace_parse_run_command(file, buffer, count, ppos,
 675                                         create_or_delete_trace_uprobe);
 676 }
 677
 678 static const struct file_operations uprobe_events_ops = {
 679         .owner          = THIS_MODULE,
 680         .open           = probes_open,
 681         .read           = seq_read,
 682         .llseek         = seq_lseek,
 683         .release        = seq_release,
 684         .write          = probes_write,
 685 };
 686
 687 /* Probes profiling interfaces */
 688 static int probes_profile_seq_show(struct seq_file *m, void *v)
 689 {
 690         struct dyn_event *ev = v;
 691         struct trace_uprobe *tu;
 692
 693         if (!is_trace_uprobe(ev))
 694                 return 0;
 695
 696         tu = to_trace_uprobe(ev);
 697         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 698                         trace_event_name(&tu->tp.call), tu->nhit);
 699         return 0;
 700 }
 701
 702 static const struct seq_operations profile_seq_op = {
 703         .start  = dyn_event_seq_start,
 704         .next   = dyn_event_seq_next,
 705         .stop   = dyn_event_seq_stop,
 706         .show   = probes_profile_seq_show
 707 };
 708
 709 static int profile_open(struct inode *inode, struct file *file)
 710 {
 711         return seq_open(file, &profile_seq_op);
 712 }
 713
 714 static const struct file_operations uprobe_profile_ops = {
 715         .owner          = THIS_MODULE,
 716         .open           = profile_open,
 717         .read           = seq_read,
 718         .llseek         = seq_lseek,
 719         .release        = seq_release,
 720 };
 721
 722 struct uprobe_cpu_buffer {
 723         struct mutex mutex;
 724         void *buf;
 725 };
 726 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 727 static int uprobe_buffer_refcnt;
 728
 729 static int uprobe_buffer_init(void)
 730 {
 731         int cpu, err_cpu;
 732
 733         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 734         if (uprobe_cpu_buffer == NULL)
 735                 return -ENOMEM;
 736
 737         for_each_possible_cpu(cpu) {
 738                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 739                                                   GFP_KERNEL, 0);
 740                 if (p == NULL) {
 741                         err_cpu = cpu;
 742                         goto err;
 743                 }
 744                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 745                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 746         }
 747
 748         return 0;
 749
 750 err:
 751         for_each_possible_cpu(cpu) {
 752                 if (cpu == err_cpu)
 753                         break;
 754                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 755         }
 756
 757         free_percpu(uprobe_cpu_buffer);
 758         return -ENOMEM;
 759 }
 760
 761 static int uprobe_buffer_enable(void)
 762 {
 763         int ret = 0;
 764
 765         BUG_ON(!mutex_is_locked(&event_mutex));
 766
 767         if (uprobe_buffer_refcnt++ == 0) {
 768                 ret = uprobe_buffer_init();
 769                 if (ret < 0)
 770                         uprobe_buffer_refcnt--;
 771         }
 772
 773         return ret;
 774 }
 775
 776 static void uprobe_buffer_disable(void)
 777 {
 778         int cpu;
 779
 780         BUG_ON(!mutex_is_locked(&event_mutex));
 781
 782         if (--uprobe_buffer_refcnt == 0) {
 783                 for_each_possible_cpu(cpu)
 784                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 785                                                              cpu)->buf);
 786
 787                 free_percpu(uprobe_cpu_buffer);
 788                 uprobe_cpu_buffer = NULL;
 789         }
 790 }
 791
 792 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 793 {
 794         struct uprobe_cpu_buffer *ucb;
 795         int cpu;
 796
 797         cpu = raw_smp_processor_id();
 798         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 799
 800         /*
 801          * Use per-cpu buffers for fastest access, but we might migrate
 802          * so the mutex makes sure we have sole access to it.
 803          */
 804         mutex_lock(&ucb->mutex);
 805
 806         return ucb;
 807 }
 808
 809 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 810 {
 811         mutex_unlock(&ucb->mutex);
 812 }
 813
 814 static void __uprobe_trace_func(struct trace_uprobe *tu,
 815                                 unsigned long func, struct pt_regs *regs,
 816                                 struct uprobe_cpu_buffer *ucb, int dsize,
 817                                 struct trace_event_file *trace_file)
 818 {
 819         struct uprobe_trace_entry_head *entry;
 820         struct ring_buffer_event *event;
 821         struct ring_buffer *buffer;
 822         void *data;
 823         int size, esize;
 824         struct trace_event_call *call = &tu->tp.call;
 825
 826         WARN_ON(call != trace_file->event_call);
 827
 828         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 829                 return;
 830
 831         if (trace_trigger_soft_disabled(trace_file))
 832                 return;
 833
 834         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 835         size = esize + tu->tp.size + dsize;
 836         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 837                                                 call->event.type, size, 0, 0);
 838         if (!event)
 839                 return;
 840
 841         entry = ring_buffer_event_data(event);
 842         if (is_ret_probe(tu)) {
 843                 entry->vaddr[0] = func;
 844                 entry->vaddr[1] = instruction_pointer(regs);
 845                 data = DATAOF_TRACE_ENTRY(entry, true);
 846         } else {
 847                 entry->vaddr[0] = instruction_pointer(regs);
 848                 data = DATAOF_TRACE_ENTRY(entry, false);
 849         }
 850
 851         memcpy(data, ucb->buf, tu->tp.size + dsize);
 852
 853         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 854 }
 855
 856 /* uprobe handler */
 857 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 858                              struct uprobe_cpu_buffer *ucb, int dsize)
 859 {
 860         struct event_file_link *link;
 861
 862         if (is_ret_probe(tu))
 863                 return 0;
 864
 865         rcu_read_lock();
 866         list_for_each_entry_rcu(link, &tu->tp.files, list)
 867                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 868         rcu_read_unlock();
 869
 870         return 0;
 871 }
 872
 873 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 874                                  struct pt_regs *regs,
 875                                  struct uprobe_cpu_buffer *ucb, int dsize)
 876 {
 877         struct event_file_link *link;
 878
 879         rcu_read_lock();
 880         list_for_each_entry_rcu(link, &tu->tp.files, list)
 881                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 882         rcu_read_unlock();
 883 }
 884
 885 /* Event entry printers */
 886 static enum print_line_t
 887 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 888 {
 889         struct uprobe_trace_entry_head *entry;
 890         struct trace_seq *s = &iter->seq;
 891         struct trace_uprobe *tu;
 892         u8 *data;
 893
 894         entry = (struct uprobe_trace_entry_head *)iter->ent;
 895         tu = container_of(event, struct trace_uprobe, tp.call.event);
 896
 897         if (is_ret_probe(tu)) {
 898                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 899                                  trace_event_name(&tu->tp.call),
 900                                  entry->vaddr[1], entry->vaddr[0]);
 901                 data = DATAOF_TRACE_ENTRY(entry, true);
 902         } else {
 903                 trace_seq_printf(s, "%s: (0x%lx)",
 904                                  trace_event_name(&tu->tp.call),
 905                                  entry->vaddr[0]);
 906                 data = DATAOF_TRACE_ENTRY(entry, false);
 907         }
 908
 909         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
 910                 goto out;
 911
 912         trace_seq_putc(s, '\n');
 913
 914  out:
 915         return trace_handle_return(s);
 916 }
 917
 918 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 919                                 enum uprobe_filter_ctx ctx,
 920                                 struct mm_struct *mm);
 921
 922 static int
 923 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
 924                    filter_func_t filter)
 925 {
 926         bool enabled = trace_probe_is_enabled(&tu->tp);
 927         struct event_file_link *link = NULL;
 928         int ret;
 929
 930         if (file) {
 931                 if (tu->tp.flags & TP_FLAG_PROFILE)
 932                         return -EINTR;
 933
 934                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 935                 if (!link)
 936                         return -ENOMEM;
 937
 938                 link->file = file;
 939                 list_add_tail_rcu(&link->list, &tu->tp.files);
 940
 941                 tu->tp.flags |= TP_FLAG_TRACE;
 942         } else {
 943                 if (tu->tp.flags & TP_FLAG_TRACE)
 944                         return -EINTR;
 945
 946                 tu->tp.flags |= TP_FLAG_PROFILE;
 947         }
 948
 949         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 950
 951         if (enabled)
 952                 return 0;
 953
 954         ret = uprobe_buffer_enable();
 955         if (ret)
 956                 goto err_flags;
 957
 958         tu->consumer.filter = filter;
 959         tu->inode = d_real_inode(tu->path.dentry);
 960         if (tu->ref_ctr_offset) {
 961                 ret = uprobe_register_refctr(tu->inode, tu->offset,
 962                                 tu->ref_ctr_offset, &tu->consumer);
 963         } else {
 964                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 965         }
 966
 967         if (ret)
 968                 goto err_buffer;
 969
 970         return 0;
 971
 972  err_buffer:
 973         uprobe_buffer_disable();
 974
 975  err_flags:
 976         if (file) {
 977                 list_del(&link->list);
 978                 kfree(link);
 979                 tu->tp.flags &= ~TP_FLAG_TRACE;
 980         } else {
 981                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 982         }
 983         return ret;
 984 }
 985
 986 static void
 987 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 988 {
 989         if (!trace_probe_is_enabled(&tu->tp))
 990                 return;
 991
 992         if (file) {
 993                 struct event_file_link *link;
 994
 995                 link = find_event_file_link(&tu->tp, file);
 996                 if (!link)
 997                         return;
 998
 999                 list_del_rcu(&link->list);
1000                 /* synchronize with u{,ret}probe_trace_func */
1001                 synchronize_rcu();
1002                 kfree(link);
1003
1004                 if (!list_empty(&tu->tp.files))
1005                         return;
1006         }
1007
1008         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
1009
1010         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1011         tu->inode = NULL;
1012         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
1013
1014         uprobe_buffer_disable();
1015 }
1016
1017 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1018 {
1019         int ret, size;
1020         struct uprobe_trace_entry_head field;
1021         struct trace_uprobe *tu = event_call->data;
1022
1023         if (is_ret_probe(tu)) {
1024                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1025                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1026                 size = SIZEOF_TRACE_ENTRY(true);
1027         } else {
1028                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1029                 size = SIZEOF_TRACE_ENTRY(false);
1030         }
1031
1032         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1033 }
1034
1035 #ifdef CONFIG_PERF_EVENTS
1036 static bool
1037 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1038 {
1039         struct perf_event *event;
1040
1041         if (filter->nr_systemwide)
1042                 return true;
1043
1044         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1045                 if (event->hw.target->mm == mm)
1046                         return true;
1047         }
1048
1049         return false;
1050 }
1051
1052 static inline bool
1053 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1054 {
1055         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1056 }
1057
1058 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1059 {
1060         bool done;
1061
1062         write_lock(&tu->filter.rwlock);
1063         if (event->hw.target) {
1064                 list_del(&event->hw.tp_list);
1065                 done = tu->filter.nr_systemwide ||
1066                         (event->hw.target->flags & PF_EXITING) ||
1067                         uprobe_filter_event(tu, event);
1068         } else {
1069                 tu->filter.nr_systemwide--;
1070                 done = tu->filter.nr_systemwide;
1071         }
1072         write_unlock(&tu->filter.rwlock);
1073
1074         if (!done)
1075                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1076
1077         return 0;
1078 }
1079
1080 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1081 {
1082         bool done;
1083         int err;
1084
1085         write_lock(&tu->filter.rwlock);
1086         if (event->hw.target) {
1087                 /*
1088                  * event->parent != NULL means copy_process(), we can avoid
1089                  * uprobe_apply(). current->mm must be probed and we can rely
1090                  * on dup_mmap() which preserves the already installed bp's.
1091                  *
1092                  * attr.enable_on_exec means that exec/mmap will install the
1093                  * breakpoints we need.
1094                  */
1095                 done = tu->filter.nr_systemwide ||
1096                         event->parent || event->attr.enable_on_exec ||
1097                         uprobe_filter_event(tu, event);
1098                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1099         } else {
1100                 done = tu->filter.nr_systemwide;
1101                 tu->filter.nr_systemwide++;
1102         }
1103         write_unlock(&tu->filter.rwlock);
1104
1105         err = 0;
1106         if (!done) {
1107                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1108                 if (err)
1109                         uprobe_perf_close(tu, event);
1110         }
1111         return err;
1112 }
1113
1114 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1115                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1116 {
1117         struct trace_uprobe *tu;
1118         int ret;
1119
1120         tu = container_of(uc, struct trace_uprobe, consumer);
1121         read_lock(&tu->filter.rwlock);
1122         ret = __uprobe_perf_filter(&tu->filter, mm);
1123         read_unlock(&tu->filter.rwlock);
1124
1125         return ret;
1126 }
1127
1128 static void __uprobe_perf_func(struct trace_uprobe *tu,
1129                                unsigned long func, struct pt_regs *regs,
1130                                struct uprobe_cpu_buffer *ucb, int dsize)
1131 {
1132         struct trace_event_call *call = &tu->tp.call;
1133         struct uprobe_trace_entry_head *entry;
1134         struct hlist_head *head;
1135         void *data;
1136         int size, esize;
1137         int rctx;
1138
1139         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1140                 return;
1141
1142         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1143
1144         size = esize + tu->tp.size + dsize;
1145         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1146         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1147                 return;
1148
1149         preempt_disable();
1150         head = this_cpu_ptr(call->perf_events);
1151         if (hlist_empty(head))
1152                 goto out;
1153
1154         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1155         if (!entry)
1156                 goto out;
1157
1158         if (is_ret_probe(tu)) {
1159                 entry->vaddr[0] = func;
1160                 entry->vaddr[1] = instruction_pointer(regs);
1161                 data = DATAOF_TRACE_ENTRY(entry, true);
1162         } else {
1163                 entry->vaddr[0] = instruction_pointer(regs);
1164                 data = DATAOF_TRACE_ENTRY(entry, false);
1165         }
1166
1167         memcpy(data, ucb->buf, tu->tp.size + dsize);
1168
1169         if (size - esize > tu->tp.size + dsize) {
1170                 int len = tu->tp.size + dsize;
1171
1172                 memset(data + len, 0, size - esize - len);
1173         }
1174
1175         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1176                               head, NULL);
1177  out:
1178         preempt_enable();
1179 }
1180
1181 /* uprobe profile handler */
1182 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1183                             struct uprobe_cpu_buffer *ucb, int dsize)
1184 {
1185         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1186                 return UPROBE_HANDLER_REMOVE;
1187
1188         if (!is_ret_probe(tu))
1189                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1190         return 0;
1191 }
1192
1193 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1194                                 struct pt_regs *regs,
1195                                 struct uprobe_cpu_buffer *ucb, int dsize)
1196 {
1197         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1198 }
1199
1200 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1201                         const char **filename, u64 *probe_offset,
1202                         bool perf_type_tracepoint)
1203 {
1204         const char *pevent = trace_event_name(event->tp_event);
1205         const char *group = event->tp_event->class->system;
1206         struct trace_uprobe *tu;
1207
1208         if (perf_type_tracepoint)
1209                 tu = find_probe_event(pevent, group);
1210         else
1211                 tu = event->tp_event->data;
1212         if (!tu)
1213                 return -EINVAL;
1214
1215         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1216                                     : BPF_FD_TYPE_UPROBE;
1217         *filename = tu->filename;
1218         *probe_offset = tu->offset;
1219         return 0;
1220 }
1221 #endif  /* CONFIG_PERF_EVENTS */
1222
1223 static int
1224 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1225                       void *data)
1226 {
1227         struct trace_uprobe *tu = event->data;
1228         struct trace_event_file *file = data;
1229
1230         switch (type) {
1231         case TRACE_REG_REGISTER:
1232                 return probe_event_enable(tu, file, NULL);
1233
1234         case TRACE_REG_UNREGISTER:
1235                 probe_event_disable(tu, file);
1236                 return 0;
1237
1238 #ifdef CONFIG_PERF_EVENTS
1239         case TRACE_REG_PERF_REGISTER:
1240                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1241
1242         case TRACE_REG_PERF_UNREGISTER:
1243                 probe_event_disable(tu, NULL);
1244                 return 0;
1245
1246         case TRACE_REG_PERF_OPEN:
1247                 return uprobe_perf_open(tu, data);
1248
1249         case TRACE_REG_PERF_CLOSE:
1250                 return uprobe_perf_close(tu, data);
1251
1252 #endif
1253         default:
1254                 return 0;
1255         }
1256         return 0;
1257 }
1258
1259 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1260 {
1261         struct trace_uprobe *tu;
1262         struct uprobe_dispatch_data udd;
1263         struct uprobe_cpu_buffer *ucb;
1264         int dsize, esize;
1265         int ret = 0;
1266
1267
1268         tu = container_of(con, struct trace_uprobe, consumer);
1269         tu->nhit++;
1270
1271         udd.tu = tu;
1272         udd.bp_addr = instruction_pointer(regs);
1273
1274         current->utask->vaddr = (unsigned long) &udd;
1275
1276         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1277                 return 0;
1278
1279         dsize = __get_data_size(&tu->tp, regs);
1280         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1281
1282         ucb = uprobe_buffer_get();
1283         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1284
1285         if (tu->tp.flags & TP_FLAG_TRACE)
1286                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1287
1288 #ifdef CONFIG_PERF_EVENTS
1289         if (tu->tp.flags & TP_FLAG_PROFILE)
1290                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1291 #endif
1292         uprobe_buffer_put(ucb);
1293         return ret;
1294 }
1295
1296 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1297                                 unsigned long func, struct pt_regs *regs)
1298 {
1299         struct trace_uprobe *tu;
1300         struct uprobe_dispatch_data udd;
1301         struct uprobe_cpu_buffer *ucb;
1302         int dsize, esize;
1303
1304         tu = container_of(con, struct trace_uprobe, consumer);
1305
1306         udd.tu = tu;
1307         udd.bp_addr = func;
1308
1309         current->utask->vaddr = (unsigned long) &udd;
1310
1311         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1312                 return 0;
1313
1314         dsize = __get_data_size(&tu->tp, regs);
1315         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1316
1317         ucb = uprobe_buffer_get();
1318         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1319
1320         if (tu->tp.flags & TP_FLAG_TRACE)
1321                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1322
1323 #ifdef CONFIG_PERF_EVENTS
1324         if (tu->tp.flags & TP_FLAG_PROFILE)
1325                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1326 #endif
1327         uprobe_buffer_put(ucb);
1328         return 0;
1329 }
1330
1331 static struct trace_event_functions uprobe_funcs = {
1332         .trace          = print_uprobe_event
1333 };
1334
1335 static inline void init_trace_event_call(struct trace_uprobe *tu,
1336                                          struct trace_event_call *call)
1337 {
1338         call->event.funcs = &uprobe_funcs;
1339         call->class->define_fields = uprobe_event_define_fields;
1340
1341         call->flags = TRACE_EVENT_FL_UPROBE;
1342         call->class->reg = trace_uprobe_register;
1343         call->data = tu;
1344 }
1345
1346 static int register_uprobe_event(struct trace_uprobe *tu)
1347 {
1348         struct trace_event_call *call = &tu->tp.call;
1349         int ret = 0;
1350
1351         init_trace_event_call(tu, call);
1352
1353         ret = register_trace_event(&call->event);
1354         if (!ret)
1355                 return -ENODEV;
1356
1357         ret = trace_add_event_call(call);
1358
1359         if (ret) {
1360                 pr_info("Failed to register uprobe event: %s\n",
1361                         trace_event_name(call));
1362                 unregister_trace_event(&call->event);
1363         }
1364
1365         return ret;
1366 }
1367
1368 static int unregister_uprobe_event(struct trace_uprobe *tu)
1369 {
1370         /* tu->event is unregistered in trace_remove_event_call() */
1371         return trace_remove_event_call(&tu->tp.call);
1372 }
1373
1374 #ifdef CONFIG_PERF_EVENTS
1375 struct trace_event_call *
1376 create_local_trace_uprobe(char *name, unsigned long offs,
1377                           unsigned long ref_ctr_offset, bool is_return)
1378 {
1379         struct trace_uprobe *tu;
1380         struct path path;
1381         int ret;
1382
1383         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1384         if (ret)
1385                 return ERR_PTR(ret);
1386
1387         if (!d_is_reg(path.dentry)) {
1388                 path_put(&path);
1389                 return ERR_PTR(-EINVAL);
1390         }
1391
1392         /*
1393          * local trace_kprobes are not added to dyn_event, so they are never
1394          * searched in find_trace_kprobe(). Therefore, there is no concern of
1395          * duplicated name "DUMMY_EVENT" here.
1396          */
1397         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1398                                 is_return);
1399
1400         if (IS_ERR(tu)) {
1401                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1402                         (int)PTR_ERR(tu));
1403                 path_put(&path);
1404                 return ERR_CAST(tu);
1405         }
1406
1407         tu->offset = offs;
1408         tu->path = path;
1409         tu->ref_ctr_offset = ref_ctr_offset;
1410         tu->filename = kstrdup(name, GFP_KERNEL);
1411         init_trace_event_call(tu, &tu->tp.call);
1412
1413         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1414                 ret = -ENOMEM;
1415                 goto error;
1416         }
1417
1418         return &tu->tp.call;
1419 error:
1420         free_trace_uprobe(tu);
1421         return ERR_PTR(ret);
1422 }
1423
1424 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1425 {
1426         struct trace_uprobe *tu;
1427
1428         tu = container_of(event_call, struct trace_uprobe, tp.call);
1429
1430         free_trace_uprobe(tu);
1431 }
1432 #endif /* CONFIG_PERF_EVENTS */
1433
1434 /* Make a trace interface for controling probe points */
1435 static __init int init_uprobe_trace(void)
1436 {
1437         struct dentry *d_tracer;
1438         int ret;
1439
1440         ret = dyn_event_register(&trace_uprobe_ops);
1441         if (ret)
1442                 return ret;
1443
1444         d_tracer = tracing_init_dentry();
1445         if (IS_ERR(d_tracer))
1446                 return 0;
1447
1448         trace_create_file("uprobe_events", 0644, d_tracer,
1449                                     NULL, &uprobe_events_ops);
1450         /* Profile interface */
1451         trace_create_file("uprobe_profile", 0444, d_tracer,
1452                                     NULL, &uprobe_profile_ops);
1453         return 0;
1454 }
1455
1456 fs_initcall(init_uprobe_trace);