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/security.h>
  11 #include <linux/ctype.h>
  12 #include <linux/module.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/uprobes.h>
  15 #include <linux/namei.h>
  16 #include <linux/string.h>
  17 #include <linux/rculist.h>
  18
  19 #include "trace_dynevent.h"
  20 #include "trace_probe.h"
  21 #include "trace_probe_tmpl.h"
  22
  23 #define UPROBE_EVENT_SYSTEM     "uprobes"
  24
  25 struct uprobe_trace_entry_head {
  26         struct trace_entry      ent;
  27         unsigned long           vaddr[];
  28 };
  29
  30 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  31         (sizeof(struct uprobe_trace_entry_head) +       \
  32          sizeof(unsigned long) * (is_return ? 2 : 1))
  33
  34 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  35         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  36
  37 static int trace_uprobe_create(int argc, const char **argv);
  38 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  39 static int trace_uprobe_release(struct dyn_event *ev);
  40 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  41 static bool trace_uprobe_match(const char *system, const char *event,
  42                         int argc, const char **argv, struct dyn_event *ev);
  43
  44 static struct dyn_event_operations trace_uprobe_ops = {
  45         .create = trace_uprobe_create,
  46         .show = trace_uprobe_show,
  47         .is_busy = trace_uprobe_is_busy,
  48         .free = trace_uprobe_release,
  49         .match = trace_uprobe_match,
  50 };
  51
  52 /*
  53  * uprobe event core functions
  54  */
  55 struct trace_uprobe {
  56         struct dyn_event                devent;
  57         struct uprobe_consumer          consumer;
  58         struct path                     path;
  59         struct inode                    *inode;
  60         char                            *filename;
  61         unsigned long                   offset;
  62         unsigned long                   ref_ctr_offset;
  63         unsigned long                   nhit;
  64         struct trace_probe              tp;
  65 };
  66
  67 static bool is_trace_uprobe(struct dyn_event *ev)
  68 {
  69         return ev->ops == &trace_uprobe_ops;
  70 }
  71
  72 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  73 {
  74         return container_of(ev, struct trace_uprobe, devent);
  75 }
  76
  77 /**
  78  * for_each_trace_uprobe - iterate over the trace_uprobe list
  79  * @pos:        the struct trace_uprobe * for each entry
  80  * @dpos:       the struct dyn_event * to use as a loop cursor
  81  */
  82 #define for_each_trace_uprobe(pos, dpos)        \
  83         for_each_dyn_event(dpos)                \
  84                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  85
  86 #define SIZEOF_TRACE_UPROBE(n)                          \
  87         (offsetof(struct trace_uprobe, tp.args) +       \
  88         (sizeof(struct probe_arg) * (n)))
  89
  90 static int register_uprobe_event(struct trace_uprobe *tu);
  91 static int unregister_uprobe_event(struct trace_uprobe *tu);
  92
  93 struct uprobe_dispatch_data {
  94         struct trace_uprobe     *tu;
  95         unsigned long           bp_addr;
  96 };
  97
  98 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  99 static int uretprobe_dispatcher(struct uprobe_consumer *con,
 100                                 unsigned long func, struct pt_regs *regs);
 101
 102 #ifdef CONFIG_STACK_GROWSUP
 103 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 104 {
 105         return addr - (n * sizeof(long));
 106 }
 107 #else
 108 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 109 {
 110         return addr + (n * sizeof(long));
 111 }
 112 #endif
 113
 114 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 115 {
 116         unsigned long ret;
 117         unsigned long addr = user_stack_pointer(regs);
 118
 119         addr = adjust_stack_addr(addr, n);
 120
 121         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 122                 return 0;
 123
 124         return ret;
 125 }
 126
 127 /*
 128  * Uprobes-specific fetch functions
 129  */
 130 static nokprobe_inline int
 131 probe_mem_read(void *dest, void *src, size_t size)
 132 {
 133         void __user *vaddr = (void __force __user *)src;
 134
 135         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 136 }
 137
 138 static nokprobe_inline int
 139 probe_mem_read_user(void *dest, void *src, size_t size)
 140 {
 141         return probe_mem_read(dest, src, size);
 142 }
 143
 144 /*
 145  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 146  * length and relative data location.
 147  */
 148 static nokprobe_inline int
 149 fetch_store_string(unsigned long addr, void *dest, void *base)
 150 {
 151         long ret;
 152         u32 loc = *(u32 *)dest;
 153         int maxlen  = get_loc_len(loc);
 154         u8 *dst = get_loc_data(dest, base);
 155         void __user *src = (void __force __user *) addr;
 156
 157         if (unlikely(!maxlen))
 158                 return -ENOMEM;
 159
 160         if (addr == FETCH_TOKEN_COMM)
 161                 ret = strlcpy(dst, current->comm, maxlen);
 162         else
 163                 ret = strncpy_from_user(dst, src, maxlen);
 164         if (ret >= 0) {
 165                 if (ret == maxlen)
 166                         dst[ret - 1] = '\0';
 167                 else
 168                         /*
 169                          * Include the terminating null byte. In this case it
 170                          * was copied by strncpy_from_user but not accounted
 171                          * for in ret.
 172                          */
 173                         ret++;
 174                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 175         }
 176
 177         return ret;
 178 }
 179
 180 static nokprobe_inline int
 181 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 182 {
 183         return fetch_store_string(addr, dest, base);
 184 }
 185
 186 /* Return the length of string -- including null terminal byte */
 187 static nokprobe_inline int
 188 fetch_store_strlen(unsigned long addr)
 189 {
 190         int len;
 191         void __user *vaddr = (void __force __user *) addr;
 192
 193         if (addr == FETCH_TOKEN_COMM)
 194                 len = strlen(current->comm) + 1;
 195         else
 196                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 197
 198         return (len > MAX_STRING_SIZE) ? 0 : len;
 199 }
 200
 201 static nokprobe_inline int
 202 fetch_store_strlen_user(unsigned long addr)
 203 {
 204         return fetch_store_strlen(addr);
 205 }
 206
 207 static unsigned long translate_user_vaddr(unsigned long file_offset)
 208 {
 209         unsigned long base_addr;
 210         struct uprobe_dispatch_data *udd;
 211
 212         udd = (void *) current->utask->vaddr;
 213
 214         base_addr = udd->bp_addr - udd->tu->offset;
 215         return base_addr + file_offset;
 216 }
 217
 218 /* Note that we don't verify it, since the code does not come from user space */
 219 static int
 220 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 221                    void *base)
 222 {
 223         struct pt_regs *regs = rec;
 224         unsigned long val;
 225
 226         /* 1st stage: get value from context */
 227         switch (code->op) {
 228         case FETCH_OP_REG:
 229                 val = regs_get_register(regs, code->param);
 230                 break;
 231         case FETCH_OP_STACK:
 232                 val = get_user_stack_nth(regs, code->param);
 233                 break;
 234         case FETCH_OP_STACKP:
 235                 val = user_stack_pointer(regs);
 236                 break;
 237         case FETCH_OP_RETVAL:
 238                 val = regs_return_value(regs);
 239                 break;
 240         case FETCH_OP_IMM:
 241                 val = code->immediate;
 242                 break;
 243         case FETCH_OP_COMM:
 244                 val = FETCH_TOKEN_COMM;
 245                 break;
 246         case FETCH_OP_DATA:
 247                 val = (unsigned long)code->data;
 248                 break;
 249         case FETCH_OP_FOFFS:
 250                 val = translate_user_vaddr(code->immediate);
 251                 break;
 252         default:
 253                 return -EILSEQ;
 254         }
 255         code++;
 256
 257         return process_fetch_insn_bottom(code, val, dest, base);
 258 }
 259 NOKPROBE_SYMBOL(process_fetch_insn)
 260
 261 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 262 {
 263         rwlock_init(&filter->rwlock);
 264         filter->nr_systemwide = 0;
 265         INIT_LIST_HEAD(&filter->perf_events);
 266 }
 267
 268 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 269 {
 270         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 271 }
 272
 273 static inline bool is_ret_probe(struct trace_uprobe *tu)
 274 {
 275         return tu->consumer.ret_handler != NULL;
 276 }
 277
 278 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 279 {
 280         struct trace_uprobe *tu = to_trace_uprobe(ev);
 281
 282         return trace_probe_is_enabled(&tu->tp);
 283 }
 284
 285 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 286                                             int argc, const char **argv)
 287 {
 288         char buf[MAX_ARGSTR_LEN + 1];
 289         int len;
 290
 291         if (!argc)
 292                 return true;
 293
 294         len = strlen(tu->filename);
 295         if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 296                 return false;
 297
 298         if (tu->ref_ctr_offset == 0)
 299                 snprintf(buf, sizeof(buf), "0x%0*lx",
 300                                 (int)(sizeof(void *) * 2), tu->offset);
 301         else
 302                 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 303                                 (int)(sizeof(void *) * 2), tu->offset,
 304                                 tu->ref_ctr_offset);
 305         if (strcmp(buf, &argv[0][len + 1]))
 306                 return false;
 307
 308         argc--; argv++;
 309
 310         return trace_probe_match_command_args(&tu->tp, argc, argv);
 311 }
 312
 313 static bool trace_uprobe_match(const char *system, const char *event,
 314                         int argc, const char **argv, struct dyn_event *ev)
 315 {
 316         struct trace_uprobe *tu = to_trace_uprobe(ev);
 317
 318         return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 319            (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 320            trace_uprobe_match_command_head(tu, argc, argv);
 321 }
 322
 323 static nokprobe_inline struct trace_uprobe *
 324 trace_uprobe_primary_from_call(struct trace_event_call *call)
 325 {
 326         struct trace_probe *tp;
 327
 328         tp = trace_probe_primary_from_call(call);
 329         if (WARN_ON_ONCE(!tp))
 330                 return NULL;
 331
 332         return container_of(tp, struct trace_uprobe, tp);
 333 }
 334
 335 /*
 336  * Allocate new trace_uprobe and initialize it (including uprobes).
 337  */
 338 static struct trace_uprobe *
 339 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 340 {
 341         struct trace_uprobe *tu;
 342         int ret;
 343
 344         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 345         if (!tu)
 346                 return ERR_PTR(-ENOMEM);
 347
 348         ret = trace_probe_init(&tu->tp, event, group, true);
 349         if (ret < 0)
 350                 goto error;
 351
 352         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 353         tu->consumer.handler = uprobe_dispatcher;
 354         if (is_ret)
 355                 tu->consumer.ret_handler = uretprobe_dispatcher;
 356         init_trace_uprobe_filter(tu->tp.event->filter);
 357         return tu;
 358
 359 error:
 360         kfree(tu);
 361
 362         return ERR_PTR(ret);
 363 }
 364
 365 static void free_trace_uprobe(struct trace_uprobe *tu)
 366 {
 367         if (!tu)
 368                 return;
 369
 370         path_put(&tu->path);
 371         trace_probe_cleanup(&tu->tp);
 372         kfree(tu->filename);
 373         kfree(tu);
 374 }
 375
 376 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 377 {
 378         struct dyn_event *pos;
 379         struct trace_uprobe *tu;
 380
 381         for_each_trace_uprobe(tu, pos)
 382                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 383                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 384                         return tu;
 385
 386         return NULL;
 387 }
 388
 389 /* Unregister a trace_uprobe and probe_event */
 390 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 391 {
 392         int ret;
 393
 394         if (trace_probe_has_sibling(&tu->tp))
 395                 goto unreg;
 396
 397         ret = unregister_uprobe_event(tu);
 398         if (ret)
 399                 return ret;
 400
 401 unreg:
 402         dyn_event_remove(&tu->devent);
 403         trace_probe_unlink(&tu->tp);
 404         free_trace_uprobe(tu);
 405         return 0;
 406 }
 407
 408 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 409                                          struct trace_uprobe *comp)
 410 {
 411         struct trace_probe_event *tpe = orig->tp.event;
 412         struct trace_probe *pos;
 413         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 414         int i;
 415
 416         list_for_each_entry(pos, &tpe->probes, list) {
 417                 orig = container_of(pos, struct trace_uprobe, tp);
 418                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 419                     comp->offset != orig->offset)
 420                         continue;
 421
 422                 /*
 423                  * trace_probe_compare_arg_type() ensured that nr_args and
 424                  * each argument name and type are same. Let's compare comm.
 425                  */
 426                 for (i = 0; i < orig->tp.nr_args; i++) {
 427                         if (strcmp(orig->tp.args[i].comm,
 428                                    comp->tp.args[i].comm))
 429                                 break;
 430                 }
 431
 432                 if (i == orig->tp.nr_args)
 433                         return true;
 434         }
 435
 436         return false;
 437 }
 438
 439 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 440 {
 441         int ret;
 442
 443         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 444         if (ret) {
 445                 /* Note that argument starts index = 2 */
 446                 trace_probe_log_set_index(ret + 1);
 447                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 448                 return -EEXIST;
 449         }
 450         if (trace_uprobe_has_same_uprobe(to, tu)) {
 451                 trace_probe_log_set_index(0);
 452                 trace_probe_log_err(0, SAME_PROBE);
 453                 return -EEXIST;
 454         }
 455
 456         /* Append to existing event */
 457         ret = trace_probe_append(&tu->tp, &to->tp);
 458         if (!ret)
 459                 dyn_event_add(&tu->devent);
 460
 461         return ret;
 462 }
 463
 464 /*
 465  * Uprobe with multiple reference counter is not allowed. i.e.
 466  * If inode and offset matches, reference counter offset *must*
 467  * match as well. Though, there is one exception: If user is
 468  * replacing old trace_uprobe with new one(same group/event),
 469  * then we allow same uprobe with new reference counter as far
 470  * as the new one does not conflict with any other existing
 471  * ones.
 472  */
 473 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 474 {
 475         struct dyn_event *pos;
 476         struct trace_uprobe *tmp;
 477         struct inode *new_inode = d_real_inode(new->path.dentry);
 478
 479         for_each_trace_uprobe(tmp, pos) {
 480                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 481                     new->offset == tmp->offset &&
 482                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 483                         pr_warn("Reference counter offset mismatch.");
 484                         return -EINVAL;
 485                 }
 486         }
 487         return 0;
 488 }
 489
 490 /* Register a trace_uprobe and probe_event */
 491 static int register_trace_uprobe(struct trace_uprobe *tu)
 492 {
 493         struct trace_uprobe *old_tu;
 494         int ret;
 495
 496         mutex_lock(&event_mutex);
 497
 498         ret = validate_ref_ctr_offset(tu);
 499         if (ret)
 500                 goto end;
 501
 502         /* register as an event */
 503         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 504                                   trace_probe_group_name(&tu->tp));
 505         if (old_tu) {
 506                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 507                         trace_probe_log_set_index(0);
 508                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 509                         ret = -EEXIST;
 510                 } else {
 511                         ret = append_trace_uprobe(tu, old_tu);
 512                 }
 513                 goto end;
 514         }
 515
 516         ret = register_uprobe_event(tu);
 517         if (ret) {
 518                 if (ret == -EEXIST) {
 519                         trace_probe_log_set_index(0);
 520                         trace_probe_log_err(0, EVENT_EXIST);
 521                 } else
 522                         pr_warn("Failed to register probe event(%d)\n", ret);
 523                 goto end;
 524         }
 525
 526         dyn_event_add(&tu->devent);
 527
 528 end:
 529         mutex_unlock(&event_mutex);
 530
 531         return ret;
 532 }
 533
 534 /*
 535  * Argument syntax:
 536  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 537  */
 538 static int trace_uprobe_create(int argc, const char **argv)
 539 {
 540         struct trace_uprobe *tu;
 541         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 542         char *arg, *filename, *rctr, *rctr_end, *tmp;
 543         char buf[MAX_EVENT_NAME_LEN];
 544         struct path path;
 545         unsigned long offset, ref_ctr_offset;
 546         bool is_return = false;
 547         int i, ret;
 548
 549         ret = 0;
 550         ref_ctr_offset = 0;
 551
 552         switch (argv[0][0]) {
 553         case 'r':
 554                 is_return = true;
 555                 break;
 556         case 'p':
 557                 break;
 558         default:
 559                 return -ECANCELED;
 560         }
 561
 562         if (argc < 2)
 563                 return -ECANCELED;
 564
 565         if (argv[0][1] == ':')
 566                 event = &argv[0][2];
 567
 568         if (!strchr(argv[1], '/'))
 569                 return -ECANCELED;
 570
 571         filename = kstrdup(argv[1], GFP_KERNEL);
 572         if (!filename)
 573                 return -ENOMEM;
 574
 575         /* Find the last occurrence, in case the path contains ':' too. */
 576         arg = strrchr(filename, ':');
 577         if (!arg || !isdigit(arg[1])) {
 578                 kfree(filename);
 579                 return -ECANCELED;
 580         }
 581
 582         trace_probe_log_init("trace_uprobe", argc, argv);
 583         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 584
 585         *arg++ = '\0';
 586         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 587         if (ret) {
 588                 trace_probe_log_err(0, FILE_NOT_FOUND);
 589                 kfree(filename);
 590                 trace_probe_log_clear();
 591                 return ret;
 592         }
 593         if (!d_is_reg(path.dentry)) {
 594                 trace_probe_log_err(0, NO_REGULAR_FILE);
 595                 ret = -EINVAL;
 596                 goto fail_address_parse;
 597         }
 598
 599         /* Parse reference counter offset if specified. */
 600         rctr = strchr(arg, '(');
 601         if (rctr) {
 602                 rctr_end = strchr(rctr, ')');
 603                 if (!rctr_end) {
 604                         ret = -EINVAL;
 605                         rctr_end = rctr + strlen(rctr);
 606                         trace_probe_log_err(rctr_end - filename,
 607                                             REFCNT_OPEN_BRACE);
 608                         goto fail_address_parse;
 609                 } else if (rctr_end[1] != '\0') {
 610                         ret = -EINVAL;
 611                         trace_probe_log_err(rctr_end + 1 - filename,
 612                                             BAD_REFCNT_SUFFIX);
 613                         goto fail_address_parse;
 614                 }
 615
 616                 *rctr++ = '\0';
 617                 *rctr_end = '\0';
 618                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 619                 if (ret) {
 620                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 621                         goto fail_address_parse;
 622                 }
 623         }
 624
 625         /* Parse uprobe offset. */
 626         ret = kstrtoul(arg, 0, &offset);
 627         if (ret) {
 628                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 629                 goto fail_address_parse;
 630         }
 631
 632         /* setup a probe */
 633         trace_probe_log_set_index(0);
 634         if (event) {
 635                 ret = traceprobe_parse_event_name(&event, &group, buf,
 636                                                   event - argv[0]);
 637                 if (ret)
 638                         goto fail_address_parse;
 639         } else {
 640                 char *tail;
 641                 char *ptr;
 642
 643                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 644                 if (!tail) {
 645                         ret = -ENOMEM;
 646                         goto fail_address_parse;
 647                 }
 648
 649                 ptr = strpbrk(tail, ".-_");
 650                 if (ptr)
 651                         *ptr = '\0';
 652
 653                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 654                 event = buf;
 655                 kfree(tail);
 656         }
 657
 658         argc -= 2;
 659         argv += 2;
 660
 661         tu = alloc_trace_uprobe(group, event, argc, is_return);
 662         if (IS_ERR(tu)) {
 663                 ret = PTR_ERR(tu);
 664                 /* This must return -ENOMEM otherwise there is a bug */
 665                 WARN_ON_ONCE(ret != -ENOMEM);
 666                 goto fail_address_parse;
 667         }
 668         tu->offset = offset;
 669         tu->ref_ctr_offset = ref_ctr_offset;
 670         tu->path = path;
 671         tu->filename = filename;
 672
 673         /* parse arguments */
 674         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 675                 tmp = kstrdup(argv[i], GFP_KERNEL);
 676                 if (!tmp) {
 677                         ret = -ENOMEM;
 678                         goto error;
 679                 }
 680
 681                 trace_probe_log_set_index(i + 2);
 682                 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 683                                         is_return ? TPARG_FL_RETURN : 0);
 684                 kfree(tmp);
 685                 if (ret)
 686                         goto error;
 687         }
 688
 689         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 690         if (ret < 0)
 691                 goto error;
 692
 693         ret = register_trace_uprobe(tu);
 694         if (!ret)
 695                 goto out;
 696
 697 error:
 698         free_trace_uprobe(tu);
 699 out:
 700         trace_probe_log_clear();
 701         return ret;
 702
 703 fail_address_parse:
 704         trace_probe_log_clear();
 705         path_put(&path);
 706         kfree(filename);
 707
 708         return ret;
 709 }
 710
 711 static int create_or_delete_trace_uprobe(int argc, char **argv)
 712 {
 713         int ret;
 714
 715         if (argv[0][0] == '-')
 716                 return dyn_event_release(argc, argv, &trace_uprobe_ops);
 717
 718         ret = trace_uprobe_create(argc, (const char **)argv);
 719         return ret == -ECANCELED ? -EINVAL : ret;
 720 }
 721
 722 static int trace_uprobe_release(struct dyn_event *ev)
 723 {
 724         struct trace_uprobe *tu = to_trace_uprobe(ev);
 725
 726         return unregister_trace_uprobe(tu);
 727 }
 728
 729 /* Probes listing interfaces */
 730 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 731 {
 732         struct trace_uprobe *tu = to_trace_uprobe(ev);
 733         char c = is_ret_probe(tu) ? 'r' : 'p';
 734         int i;
 735
 736         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 737                         trace_probe_name(&tu->tp), tu->filename,
 738                         (int)(sizeof(void *) * 2), tu->offset);
 739
 740         if (tu->ref_ctr_offset)
 741                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 742
 743         for (i = 0; i < tu->tp.nr_args; i++)
 744                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 745
 746         seq_putc(m, '\n');
 747         return 0;
 748 }
 749
 750 static int probes_seq_show(struct seq_file *m, void *v)
 751 {
 752         struct dyn_event *ev = v;
 753
 754         if (!is_trace_uprobe(ev))
 755                 return 0;
 756
 757         return trace_uprobe_show(m, ev);
 758 }
 759
 760 static const struct seq_operations probes_seq_op = {
 761         .start  = dyn_event_seq_start,
 762         .next   = dyn_event_seq_next,
 763         .stop   = dyn_event_seq_stop,
 764         .show   = probes_seq_show
 765 };
 766
 767 static int probes_open(struct inode *inode, struct file *file)
 768 {
 769         int ret;
 770
 771         ret = security_locked_down(LOCKDOWN_TRACEFS);
 772         if (ret)
 773                 return ret;
 774
 775         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 776                 ret = dyn_events_release_all(&trace_uprobe_ops);
 777                 if (ret)
 778                         return ret;
 779         }
 780
 781         return seq_open(file, &probes_seq_op);
 782 }
 783
 784 static ssize_t probes_write(struct file *file, const char __user *buffer,
 785                             size_t count, loff_t *ppos)
 786 {
 787         return trace_parse_run_command(file, buffer, count, ppos,
 788                                         create_or_delete_trace_uprobe);
 789 }
 790
 791 static const struct file_operations uprobe_events_ops = {
 792         .owner          = THIS_MODULE,
 793         .open           = probes_open,
 794         .read           = seq_read,
 795         .llseek         = seq_lseek,
 796         .release        = seq_release,
 797         .write          = probes_write,
 798 };
 799
 800 /* Probes profiling interfaces */
 801 static int probes_profile_seq_show(struct seq_file *m, void *v)
 802 {
 803         struct dyn_event *ev = v;
 804         struct trace_uprobe *tu;
 805
 806         if (!is_trace_uprobe(ev))
 807                 return 0;
 808
 809         tu = to_trace_uprobe(ev);
 810         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 811                         trace_probe_name(&tu->tp), tu->nhit);
 812         return 0;
 813 }
 814
 815 static const struct seq_operations profile_seq_op = {
 816         .start  = dyn_event_seq_start,
 817         .next   = dyn_event_seq_next,
 818         .stop   = dyn_event_seq_stop,
 819         .show   = probes_profile_seq_show
 820 };
 821
 822 static int profile_open(struct inode *inode, struct file *file)
 823 {
 824         int ret;
 825
 826         ret = security_locked_down(LOCKDOWN_TRACEFS);
 827         if (ret)
 828                 return ret;
 829
 830         return seq_open(file, &profile_seq_op);
 831 }
 832
 833 static const struct file_operations uprobe_profile_ops = {
 834         .owner          = THIS_MODULE,
 835         .open           = profile_open,
 836         .read           = seq_read,
 837         .llseek         = seq_lseek,
 838         .release        = seq_release,
 839 };
 840
 841 struct uprobe_cpu_buffer {
 842         struct mutex mutex;
 843         void *buf;
 844 };
 845 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 846 static int uprobe_buffer_refcnt;
 847
 848 static int uprobe_buffer_init(void)
 849 {
 850         int cpu, err_cpu;
 851
 852         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 853         if (uprobe_cpu_buffer == NULL)
 854                 return -ENOMEM;
 855
 856         for_each_possible_cpu(cpu) {
 857                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 858                                                   GFP_KERNEL, 0);
 859                 if (p == NULL) {
 860                         err_cpu = cpu;
 861                         goto err;
 862                 }
 863                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 864                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 865         }
 866
 867         return 0;
 868
 869 err:
 870         for_each_possible_cpu(cpu) {
 871                 if (cpu == err_cpu)
 872                         break;
 873                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 874         }
 875
 876         free_percpu(uprobe_cpu_buffer);
 877         return -ENOMEM;
 878 }
 879
 880 static int uprobe_buffer_enable(void)
 881 {
 882         int ret = 0;
 883
 884         BUG_ON(!mutex_is_locked(&event_mutex));
 885
 886         if (uprobe_buffer_refcnt++ == 0) {
 887                 ret = uprobe_buffer_init();
 888                 if (ret < 0)
 889                         uprobe_buffer_refcnt--;
 890         }
 891
 892         return ret;
 893 }
 894
 895 static void uprobe_buffer_disable(void)
 896 {
 897         int cpu;
 898
 899         BUG_ON(!mutex_is_locked(&event_mutex));
 900
 901         if (--uprobe_buffer_refcnt == 0) {
 902                 for_each_possible_cpu(cpu)
 903                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 904                                                              cpu)->buf);
 905
 906                 free_percpu(uprobe_cpu_buffer);
 907                 uprobe_cpu_buffer = NULL;
 908         }
 909 }
 910
 911 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 912 {
 913         struct uprobe_cpu_buffer *ucb;
 914         int cpu;
 915
 916         cpu = raw_smp_processor_id();
 917         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 918
 919         /*
 920          * Use per-cpu buffers for fastest access, but we might migrate
 921          * so the mutex makes sure we have sole access to it.
 922          */
 923         mutex_lock(&ucb->mutex);
 924
 925         return ucb;
 926 }
 927
 928 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 929 {
 930         mutex_unlock(&ucb->mutex);
 931 }
 932
 933 static void __uprobe_trace_func(struct trace_uprobe *tu,
 934                                 unsigned long func, struct pt_regs *regs,
 935                                 struct uprobe_cpu_buffer *ucb, int dsize,
 936                                 struct trace_event_file *trace_file)
 937 {
 938         struct uprobe_trace_entry_head *entry;
 939         struct ring_buffer_event *event;
 940         struct ring_buffer *buffer;
 941         void *data;
 942         int size, esize;
 943         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 944
 945         WARN_ON(call != trace_file->event_call);
 946
 947         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 948                 return;
 949
 950         if (trace_trigger_soft_disabled(trace_file))
 951                 return;
 952
 953         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 954         size = esize + tu->tp.size + dsize;
 955         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 956                                                 call->event.type, size, 0, 0);
 957         if (!event)
 958                 return;
 959
 960         entry = ring_buffer_event_data(event);
 961         if (is_ret_probe(tu)) {
 962                 entry->vaddr[0] = func;
 963                 entry->vaddr[1] = instruction_pointer(regs);
 964                 data = DATAOF_TRACE_ENTRY(entry, true);
 965         } else {
 966                 entry->vaddr[0] = instruction_pointer(regs);
 967                 data = DATAOF_TRACE_ENTRY(entry, false);
 968         }
 969
 970         memcpy(data, ucb->buf, tu->tp.size + dsize);
 971
 972         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 973 }
 974
 975 /* uprobe handler */
 976 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 977                              struct uprobe_cpu_buffer *ucb, int dsize)
 978 {
 979         struct event_file_link *link;
 980
 981         if (is_ret_probe(tu))
 982                 return 0;
 983
 984         rcu_read_lock();
 985         trace_probe_for_each_link_rcu(link, &tu->tp)
 986                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 987         rcu_read_unlock();
 988
 989         return 0;
 990 }
 991
 992 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 993                                  struct pt_regs *regs,
 994                                  struct uprobe_cpu_buffer *ucb, int dsize)
 995 {
 996         struct event_file_link *link;
 997
 998         rcu_read_lock();
 999         trace_probe_for_each_link_rcu(link, &tu->tp)
1000                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1001         rcu_read_unlock();
1002 }
1003
1004 /* Event entry printers */
1005 static enum print_line_t
1006 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1007 {
1008         struct uprobe_trace_entry_head *entry;
1009         struct trace_seq *s = &iter->seq;
1010         struct trace_uprobe *tu;
1011         u8 *data;
1012
1013         entry = (struct uprobe_trace_entry_head *)iter->ent;
1014         tu = trace_uprobe_primary_from_call(
1015                 container_of(event, struct trace_event_call, event));
1016         if (unlikely(!tu))
1017                 goto out;
1018
1019         if (is_ret_probe(tu)) {
1020                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1021                                  trace_probe_name(&tu->tp),
1022                                  entry->vaddr[1], entry->vaddr[0]);
1023                 data = DATAOF_TRACE_ENTRY(entry, true);
1024         } else {
1025                 trace_seq_printf(s, "%s: (0x%lx)",
1026                                  trace_probe_name(&tu->tp),
1027                                  entry->vaddr[0]);
1028                 data = DATAOF_TRACE_ENTRY(entry, false);
1029         }
1030
1031         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1032                 goto out;
1033
1034         trace_seq_putc(s, '\n');
1035
1036  out:
1037         return trace_handle_return(s);
1038 }
1039
1040 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1041                                 enum uprobe_filter_ctx ctx,
1042                                 struct mm_struct *mm);
1043
1044 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1045 {
1046         int ret;
1047
1048         tu->consumer.filter = filter;
1049         tu->inode = d_real_inode(tu->path.dentry);
1050
1051         if (tu->ref_ctr_offset)
1052                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1053                                 tu->ref_ctr_offset, &tu->consumer);
1054         else
1055                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1056
1057         if (ret)
1058                 tu->inode = NULL;
1059
1060         return ret;
1061 }
1062
1063 static void __probe_event_disable(struct trace_probe *tp)
1064 {
1065         struct trace_probe *pos;
1066         struct trace_uprobe *tu;
1067
1068         tu = container_of(tp, struct trace_uprobe, tp);
1069         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1070
1071         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1072                 tu = container_of(pos, struct trace_uprobe, tp);
1073                 if (!tu->inode)
1074                         continue;
1075
1076                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1077                 tu->inode = NULL;
1078         }
1079 }
1080
1081 static int probe_event_enable(struct trace_event_call *call,
1082                         struct trace_event_file *file, filter_func_t filter)
1083 {
1084         struct trace_probe *pos, *tp;
1085         struct trace_uprobe *tu;
1086         bool enabled;
1087         int ret;
1088
1089         tp = trace_probe_primary_from_call(call);
1090         if (WARN_ON_ONCE(!tp))
1091                 return -ENODEV;
1092         enabled = trace_probe_is_enabled(tp);
1093
1094         /* This may also change "enabled" state */
1095         if (file) {
1096                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1097                         return -EINTR;
1098
1099                 ret = trace_probe_add_file(tp, file);
1100                 if (ret < 0)
1101                         return ret;
1102         } else {
1103                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1104                         return -EINTR;
1105
1106                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1107         }
1108
1109         tu = container_of(tp, struct trace_uprobe, tp);
1110         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1111
1112         if (enabled)
1113                 return 0;
1114
1115         ret = uprobe_buffer_enable();
1116         if (ret)
1117                 goto err_flags;
1118
1119         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1120                 tu = container_of(pos, struct trace_uprobe, tp);
1121                 ret = trace_uprobe_enable(tu, filter);
1122                 if (ret) {
1123                         __probe_event_disable(tp);
1124                         goto err_buffer;
1125                 }
1126         }
1127
1128         return 0;
1129
1130  err_buffer:
1131         uprobe_buffer_disable();
1132
1133  err_flags:
1134         if (file)
1135                 trace_probe_remove_file(tp, file);
1136         else
1137                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1138
1139         return ret;
1140 }
1141
1142 static void probe_event_disable(struct trace_event_call *call,
1143                                 struct trace_event_file *file)
1144 {
1145         struct trace_probe *tp;
1146
1147         tp = trace_probe_primary_from_call(call);
1148         if (WARN_ON_ONCE(!tp))
1149                 return;
1150
1151         if (!trace_probe_is_enabled(tp))
1152                 return;
1153
1154         if (file) {
1155                 if (trace_probe_remove_file(tp, file) < 0)
1156                         return;
1157
1158                 if (trace_probe_is_enabled(tp))
1159                         return;
1160         } else
1161                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1162
1163         __probe_event_disable(tp);
1164         uprobe_buffer_disable();
1165 }
1166
1167 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1168 {
1169         int ret, size;
1170         struct uprobe_trace_entry_head field;
1171         struct trace_uprobe *tu;
1172
1173         tu = trace_uprobe_primary_from_call(event_call);
1174         if (unlikely(!tu))
1175                 return -ENODEV;
1176
1177         if (is_ret_probe(tu)) {
1178                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1179                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1180                 size = SIZEOF_TRACE_ENTRY(true);
1181         } else {
1182                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1183                 size = SIZEOF_TRACE_ENTRY(false);
1184         }
1185
1186         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1187 }
1188
1189 #ifdef CONFIG_PERF_EVENTS
1190 static bool
1191 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1192 {
1193         struct perf_event *event;
1194
1195         if (filter->nr_systemwide)
1196                 return true;
1197
1198         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1199                 if (event->hw.target->mm == mm)
1200                         return true;
1201         }
1202
1203         return false;
1204 }
1205
1206 static inline bool
1207 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1208                           struct perf_event *event)
1209 {
1210         return __uprobe_perf_filter(filter, event->hw.target->mm);
1211 }
1212
1213 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1214                                        struct perf_event *event)
1215 {
1216         bool done;
1217
1218         write_lock(&filter->rwlock);
1219         if (event->hw.target) {
1220                 list_del(&event->hw.tp_list);
1221                 done = filter->nr_systemwide ||
1222                         (event->hw.target->flags & PF_EXITING) ||
1223                         trace_uprobe_filter_event(filter, event);
1224         } else {
1225                 filter->nr_systemwide--;
1226                 done = filter->nr_systemwide;
1227         }
1228         write_unlock(&filter->rwlock);
1229
1230         return done;
1231 }
1232
1233 /* This returns true if the filter always covers target mm */
1234 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1235                                     struct perf_event *event)
1236 {
1237         bool done;
1238
1239         write_lock(&filter->rwlock);
1240         if (event->hw.target) {
1241                 /*
1242                  * event->parent != NULL means copy_process(), we can avoid
1243                  * uprobe_apply(). current->mm must be probed and we can rely
1244                  * on dup_mmap() which preserves the already installed bp's.
1245                  *
1246                  * attr.enable_on_exec means that exec/mmap will install the
1247                  * breakpoints we need.
1248                  */
1249                 done = filter->nr_systemwide ||
1250                         event->parent || event->attr.enable_on_exec ||
1251                         trace_uprobe_filter_event(filter, event);
1252                 list_add(&event->hw.tp_list, &filter->perf_events);
1253         } else {
1254                 done = filter->nr_systemwide;
1255                 filter->nr_systemwide++;
1256         }
1257         write_unlock(&filter->rwlock);
1258
1259         return done;
1260 }
1261
1262 static int uprobe_perf_close(struct trace_event_call *call,
1263                              struct perf_event *event)
1264 {
1265         struct trace_probe *pos, *tp;
1266         struct trace_uprobe *tu;
1267         int ret = 0;
1268
1269         tp = trace_probe_primary_from_call(call);
1270         if (WARN_ON_ONCE(!tp))
1271                 return -ENODEV;
1272
1273         tu = container_of(tp, struct trace_uprobe, tp);
1274         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1275                 return 0;
1276
1277         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1278                 tu = container_of(pos, struct trace_uprobe, tp);
1279                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1280                 if (ret)
1281                         break;
1282         }
1283
1284         return ret;
1285 }
1286
1287 static int uprobe_perf_open(struct trace_event_call *call,
1288                             struct perf_event *event)
1289 {
1290         struct trace_probe *pos, *tp;
1291         struct trace_uprobe *tu;
1292         int err = 0;
1293
1294         tp = trace_probe_primary_from_call(call);
1295         if (WARN_ON_ONCE(!tp))
1296                 return -ENODEV;
1297
1298         tu = container_of(tp, struct trace_uprobe, tp);
1299         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1300                 return 0;
1301
1302         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1303                 tu = container_of(pos, struct trace_uprobe, tp);
1304                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1305                 if (err) {
1306                         uprobe_perf_close(call, event);
1307                         break;
1308                 }
1309         }
1310
1311         return err;
1312 }
1313
1314 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1315                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1316 {
1317         struct trace_uprobe_filter *filter;
1318         struct trace_uprobe *tu;
1319         int ret;
1320
1321         tu = container_of(uc, struct trace_uprobe, consumer);
1322         filter = tu->tp.event->filter;
1323
1324         read_lock(&filter->rwlock);
1325         ret = __uprobe_perf_filter(filter, mm);
1326         read_unlock(&filter->rwlock);
1327
1328         return ret;
1329 }
1330
1331 static void __uprobe_perf_func(struct trace_uprobe *tu,
1332                                unsigned long func, struct pt_regs *regs,
1333                                struct uprobe_cpu_buffer *ucb, int dsize)
1334 {
1335         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1336         struct uprobe_trace_entry_head *entry;
1337         struct hlist_head *head;
1338         void *data;
1339         int size, esize;
1340         int rctx;
1341
1342         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1343                 return;
1344
1345         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1346
1347         size = esize + tu->tp.size + dsize;
1348         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1349         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1350                 return;
1351
1352         preempt_disable();
1353         head = this_cpu_ptr(call->perf_events);
1354         if (hlist_empty(head))
1355                 goto out;
1356
1357         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1358         if (!entry)
1359                 goto out;
1360
1361         if (is_ret_probe(tu)) {
1362                 entry->vaddr[0] = func;
1363                 entry->vaddr[1] = instruction_pointer(regs);
1364                 data = DATAOF_TRACE_ENTRY(entry, true);
1365         } else {
1366                 entry->vaddr[0] = instruction_pointer(regs);
1367                 data = DATAOF_TRACE_ENTRY(entry, false);
1368         }
1369
1370         memcpy(data, ucb->buf, tu->tp.size + dsize);
1371
1372         if (size - esize > tu->tp.size + dsize) {
1373                 int len = tu->tp.size + dsize;
1374
1375                 memset(data + len, 0, size - esize - len);
1376         }
1377
1378         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1379                               head, NULL);
1380  out:
1381         preempt_enable();
1382 }
1383
1384 /* uprobe profile handler */
1385 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1386                             struct uprobe_cpu_buffer *ucb, int dsize)
1387 {
1388         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1389                 return UPROBE_HANDLER_REMOVE;
1390
1391         if (!is_ret_probe(tu))
1392                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1393         return 0;
1394 }
1395
1396 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1397                                 struct pt_regs *regs,
1398                                 struct uprobe_cpu_buffer *ucb, int dsize)
1399 {
1400         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1401 }
1402
1403 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1404                         const char **filename, u64 *probe_offset,
1405                         u64 *probe_addr, bool perf_type_tracepoint)
1406 {
1407         const char *pevent = trace_event_name(event->tp_event);
1408         const char *group = event->tp_event->class->system;
1409         struct trace_uprobe *tu;
1410
1411         if (perf_type_tracepoint)
1412                 tu = find_probe_event(pevent, group);
1413         else
1414                 tu = trace_uprobe_primary_from_call(event->tp_event);
1415         if (!tu)
1416                 return -EINVAL;
1417
1418         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1419                                     : BPF_FD_TYPE_UPROBE;
1420         *filename = tu->filename;
1421         *probe_offset = tu->offset;
1422         *probe_addr = 0;
1423         return 0;
1424 }
1425 #endif  /* CONFIG_PERF_EVENTS */
1426
1427 static int
1428 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1429                       void *data)
1430 {
1431         struct trace_event_file *file = data;
1432
1433         switch (type) {
1434         case TRACE_REG_REGISTER:
1435                 return probe_event_enable(event, file, NULL);
1436
1437         case TRACE_REG_UNREGISTER:
1438                 probe_event_disable(event, file);
1439                 return 0;
1440
1441 #ifdef CONFIG_PERF_EVENTS
1442         case TRACE_REG_PERF_REGISTER:
1443                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1444
1445         case TRACE_REG_PERF_UNREGISTER:
1446                 probe_event_disable(event, NULL);
1447                 return 0;
1448
1449         case TRACE_REG_PERF_OPEN:
1450                 return uprobe_perf_open(event, data);
1451
1452         case TRACE_REG_PERF_CLOSE:
1453                 return uprobe_perf_close(event, data);
1454
1455 #endif
1456         default:
1457                 return 0;
1458         }
1459         return 0;
1460 }
1461
1462 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1463 {
1464         struct trace_uprobe *tu;
1465         struct uprobe_dispatch_data udd;
1466         struct uprobe_cpu_buffer *ucb;
1467         int dsize, esize;
1468         int ret = 0;
1469
1470
1471         tu = container_of(con, struct trace_uprobe, consumer);
1472         tu->nhit++;
1473
1474         udd.tu = tu;
1475         udd.bp_addr = instruction_pointer(regs);
1476
1477         current->utask->vaddr = (unsigned long) &udd;
1478
1479         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1480                 return 0;
1481
1482         dsize = __get_data_size(&tu->tp, regs);
1483         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1484
1485         ucb = uprobe_buffer_get();
1486         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1487
1488         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1489                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1490
1491 #ifdef CONFIG_PERF_EVENTS
1492         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1493                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1494 #endif
1495         uprobe_buffer_put(ucb);
1496         return ret;
1497 }
1498
1499 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1500                                 unsigned long func, struct pt_regs *regs)
1501 {
1502         struct trace_uprobe *tu;
1503         struct uprobe_dispatch_data udd;
1504         struct uprobe_cpu_buffer *ucb;
1505         int dsize, esize;
1506
1507         tu = container_of(con, struct trace_uprobe, consumer);
1508
1509         udd.tu = tu;
1510         udd.bp_addr = func;
1511
1512         current->utask->vaddr = (unsigned long) &udd;
1513
1514         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1515                 return 0;
1516
1517         dsize = __get_data_size(&tu->tp, regs);
1518         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1519
1520         ucb = uprobe_buffer_get();
1521         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1522
1523         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1524                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1525
1526 #ifdef CONFIG_PERF_EVENTS
1527         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1528                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1529 #endif
1530         uprobe_buffer_put(ucb);
1531         return 0;
1532 }
1533
1534 static struct trace_event_functions uprobe_funcs = {
1535         .trace          = print_uprobe_event
1536 };
1537
1538 static inline void init_trace_event_call(struct trace_uprobe *tu)
1539 {
1540         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1541
1542         call->event.funcs = &uprobe_funcs;
1543         call->class->define_fields = uprobe_event_define_fields;
1544
1545         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1546         call->class->reg = trace_uprobe_register;
1547 }
1548
1549 static int register_uprobe_event(struct trace_uprobe *tu)
1550 {
1551         init_trace_event_call(tu);
1552
1553         return trace_probe_register_event_call(&tu->tp);
1554 }
1555
1556 static int unregister_uprobe_event(struct trace_uprobe *tu)
1557 {
1558         return trace_probe_unregister_event_call(&tu->tp);
1559 }
1560
1561 #ifdef CONFIG_PERF_EVENTS
1562 struct trace_event_call *
1563 create_local_trace_uprobe(char *name, unsigned long offs,
1564                           unsigned long ref_ctr_offset, bool is_return)
1565 {
1566         struct trace_uprobe *tu;
1567         struct path path;
1568         int ret;
1569
1570         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1571         if (ret)
1572                 return ERR_PTR(ret);
1573
1574         if (!d_is_reg(path.dentry)) {
1575                 path_put(&path);
1576                 return ERR_PTR(-EINVAL);
1577         }
1578
1579         /*
1580          * local trace_kprobes are not added to dyn_event, so they are never
1581          * searched in find_trace_kprobe(). Therefore, there is no concern of
1582          * duplicated name "DUMMY_EVENT" here.
1583          */
1584         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1585                                 is_return);
1586
1587         if (IS_ERR(tu)) {
1588                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1589                         (int)PTR_ERR(tu));
1590                 path_put(&path);
1591                 return ERR_CAST(tu);
1592         }
1593
1594         tu->offset = offs;
1595         tu->path = path;
1596         tu->ref_ctr_offset = ref_ctr_offset;
1597         tu->filename = kstrdup(name, GFP_KERNEL);
1598         init_trace_event_call(tu);
1599
1600         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1601                 ret = -ENOMEM;
1602                 goto error;
1603         }
1604
1605         return trace_probe_event_call(&tu->tp);
1606 error:
1607         free_trace_uprobe(tu);
1608         return ERR_PTR(ret);
1609 }
1610
1611 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1612 {
1613         struct trace_uprobe *tu;
1614
1615         tu = trace_uprobe_primary_from_call(event_call);
1616
1617         free_trace_uprobe(tu);
1618 }
1619 #endif /* CONFIG_PERF_EVENTS */
1620
1621 /* Make a trace interface for controling probe points */
1622 static __init int init_uprobe_trace(void)
1623 {
1624         struct dentry *d_tracer;
1625         int ret;
1626
1627         ret = dyn_event_register(&trace_uprobe_ops);
1628         if (ret)
1629                 return ret;
1630
1631         d_tracer = tracing_init_dentry();
1632         if (IS_ERR(d_tracer))
1633                 return 0;
1634
1635         trace_create_file("uprobe_events", 0644, d_tracer,
1636                                     NULL, &uprobe_events_ops);
1637         /* Profile interface */
1638         trace_create_file("uprobe_profile", 0444, d_tracer,
1639                                     NULL, &uprobe_profile_ops);
1640         return 0;
1641 }
1642
1643 fs_initcall(init_uprobe_trace);