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