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 <[email protected]>
   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, struct pt_regs *regs, void *dest,
 221                    void *base)
 222 {
 223         unsigned long val;
 224
 225         /* 1st stage: get value from context */
 226         switch (code->op) {
 227         case FETCH_OP_REG:
 228                 val = regs_get_register(regs, code->param);
 229                 break;
 230         case FETCH_OP_STACK:
 231                 val = get_user_stack_nth(regs, code->param);
 232                 break;
 233         case FETCH_OP_STACKP:
 234                 val = user_stack_pointer(regs);
 235                 break;
 236         case FETCH_OP_RETVAL:
 237                 val = regs_return_value(regs);
 238                 break;
 239         case FETCH_OP_IMM:
 240                 val = code->immediate;
 241                 break;
 242         case FETCH_OP_COMM:
 243                 val = FETCH_TOKEN_COMM;
 244                 break;
 245         case FETCH_OP_DATA:
 246                 val = (unsigned long)code->data;
 247                 break;
 248         case FETCH_OP_FOFFS:
 249                 val = translate_user_vaddr(code->immediate);
 250                 break;
 251         default:
 252                 return -EILSEQ;
 253         }
 254         code++;
 255
 256         return process_fetch_insn_bottom(code, val, dest, base);
 257 }
 258 NOKPROBE_SYMBOL(process_fetch_insn)
 259
 260 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 261 {
 262         rwlock_init(&filter->rwlock);
 263         filter->nr_systemwide = 0;
 264         INIT_LIST_HEAD(&filter->perf_events);
 265 }
 266
 267 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 268 {
 269         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 270 }
 271
 272 static inline bool is_ret_probe(struct trace_uprobe *tu)
 273 {
 274         return tu->consumer.ret_handler != NULL;
 275 }
 276
 277 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 278 {
 279         struct trace_uprobe *tu = to_trace_uprobe(ev);
 280
 281         return trace_probe_is_enabled(&tu->tp);
 282 }
 283
 284 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 285                                             int argc, const char **argv)
 286 {
 287         char buf[MAX_ARGSTR_LEN + 1];
 288         int len;
 289
 290         if (!argc)
 291                 return true;
 292
 293         len = strlen(tu->filename);
 294         if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 295                 return false;
 296
 297         if (tu->ref_ctr_offset == 0)
 298                 snprintf(buf, sizeof(buf), "0x%0*lx",
 299                                 (int)(sizeof(void *) * 2), tu->offset);
 300         else
 301                 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 302                                 (int)(sizeof(void *) * 2), tu->offset,
 303                                 tu->ref_ctr_offset);
 304         if (strcmp(buf, &argv[0][len + 1]))
 305                 return false;
 306
 307         argc--; argv++;
 308
 309         return trace_probe_match_command_args(&tu->tp, argc, argv);
 310 }
 311
 312 static bool trace_uprobe_match(const char *system, const char *event,
 313                         int argc, const char **argv, struct dyn_event *ev)
 314 {
 315         struct trace_uprobe *tu = to_trace_uprobe(ev);
 316
 317         return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 318            (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 319            trace_uprobe_match_command_head(tu, argc, argv);
 320 }
 321
 322 static nokprobe_inline struct trace_uprobe *
 323 trace_uprobe_primary_from_call(struct trace_event_call *call)
 324 {
 325         struct trace_probe *tp;
 326
 327         tp = trace_probe_primary_from_call(call);
 328         if (WARN_ON_ONCE(!tp))
 329                 return NULL;
 330
 331         return container_of(tp, struct trace_uprobe, tp);
 332 }
 333
 334 /*
 335  * Allocate new trace_uprobe and initialize it (including uprobes).
 336  */
 337 static struct trace_uprobe *
 338 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 339 {
 340         struct trace_uprobe *tu;
 341         int ret;
 342
 343         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 344         if (!tu)
 345                 return ERR_PTR(-ENOMEM);
 346
 347         ret = trace_probe_init(&tu->tp, event, group, true);
 348         if (ret < 0)
 349                 goto error;
 350
 351         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 352         tu->consumer.handler = uprobe_dispatcher;
 353         if (is_ret)
 354                 tu->consumer.ret_handler = uretprobe_dispatcher;
 355         init_trace_uprobe_filter(tu->tp.event->filter);
 356         return tu;
 357
 358 error:
 359         kfree(tu);
 360
 361         return ERR_PTR(ret);
 362 }
 363
 364 static void free_trace_uprobe(struct trace_uprobe *tu)
 365 {
 366         if (!tu)
 367                 return;
 368
 369         path_put(&tu->path);
 370         trace_probe_cleanup(&tu->tp);
 371         kfree(tu->filename);
 372         kfree(tu);
 373 }
 374
 375 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 376 {
 377         struct dyn_event *pos;
 378         struct trace_uprobe *tu;
 379
 380         for_each_trace_uprobe(tu, pos)
 381                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 382                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 383                         return tu;
 384
 385         return NULL;
 386 }
 387
 388 /* Unregister a trace_uprobe and probe_event */
 389 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 390 {
 391         int ret;
 392
 393         if (trace_probe_has_sibling(&tu->tp))
 394                 goto unreg;
 395
 396         ret = unregister_uprobe_event(tu);
 397         if (ret)
 398                 return ret;
 399
 400 unreg:
 401         dyn_event_remove(&tu->devent);
 402         trace_probe_unlink(&tu->tp);
 403         free_trace_uprobe(tu);
 404         return 0;
 405 }
 406
 407 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 408                                          struct trace_uprobe *comp)
 409 {
 410         struct trace_probe_event *tpe = orig->tp.event;
 411         struct trace_probe *pos;
 412         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 413         int i;
 414
 415         list_for_each_entry(pos, &tpe->probes, list) {
 416                 orig = container_of(pos, struct trace_uprobe, tp);
 417                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 418                     comp->offset != orig->offset)
 419                         continue;
 420
 421                 /*
 422                  * trace_probe_compare_arg_type() ensured that nr_args and
 423                  * each argument name and type are same. Let's compare comm.
 424                  */
 425                 for (i = 0; i < orig->tp.nr_args; i++) {
 426                         if (strcmp(orig->tp.args[i].comm,
 427                                    comp->tp.args[i].comm))
 428                                 break;
 429                 }
 430
 431                 if (i == orig->tp.nr_args)
 432                         return true;
 433         }
 434
 435         return false;
 436 }
 437
 438 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 439 {
 440         int ret;
 441
 442         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 443         if (ret) {
 444                 /* Note that argument starts index = 2 */
 445                 trace_probe_log_set_index(ret + 1);
 446                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 447                 return -EEXIST;
 448         }
 449         if (trace_uprobe_has_same_uprobe(to, tu)) {
 450                 trace_probe_log_set_index(0);
 451                 trace_probe_log_err(0, SAME_PROBE);
 452                 return -EEXIST;
 453         }
 454
 455         /* Append to existing event */
 456         ret = trace_probe_append(&tu->tp, &to->tp);
 457         if (!ret)
 458                 dyn_event_add(&tu->devent);
 459
 460         return ret;
 461 }
 462
 463 /*
 464  * Uprobe with multiple reference counter is not allowed. i.e.
 465  * If inode and offset matches, reference counter offset *must*
 466  * match as well. Though, there is one exception: If user is
 467  * replacing old trace_uprobe with new one(same group/event),
 468  * then we allow same uprobe with new reference counter as far
 469  * as the new one does not conflict with any other existing
 470  * ones.
 471  */
 472 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 473 {
 474         struct dyn_event *pos;
 475         struct trace_uprobe *tmp;
 476         struct inode *new_inode = d_real_inode(new->path.dentry);
 477
 478         for_each_trace_uprobe(tmp, pos) {
 479                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 480                     new->offset == tmp->offset &&
 481                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 482                         pr_warn("Reference counter offset mismatch.");
 483                         return -EINVAL;
 484                 }
 485         }
 486         return 0;
 487 }
 488
 489 /* Register a trace_uprobe and probe_event */
 490 static int register_trace_uprobe(struct trace_uprobe *tu)
 491 {
 492         struct trace_uprobe *old_tu;
 493         int ret;
 494
 495         mutex_lock(&event_mutex);
 496
 497         ret = validate_ref_ctr_offset(tu);
 498         if (ret)
 499                 goto end;
 500
 501         /* register as an event */
 502         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 503                                   trace_probe_group_name(&tu->tp));
 504         if (old_tu) {
 505                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 506                         trace_probe_log_set_index(0);
 507                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 508                         ret = -EEXIST;
 509                 } else {
 510                         ret = append_trace_uprobe(tu, old_tu);
 511                 }
 512                 goto end;
 513         }
 514
 515         ret = register_uprobe_event(tu);
 516         if (ret) {
 517                 pr_warn("Failed to register probe event(%d)\n", ret);
 518                 goto end;
 519         }
 520
 521         dyn_event_add(&tu->devent);
 522
 523 end:
 524         mutex_unlock(&event_mutex);
 525
 526         return ret;
 527 }
 528
 529 /*
 530  * Argument syntax:
 531  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 532  */
 533 static int trace_uprobe_create(int argc, const char **argv)
 534 {
 535         struct trace_uprobe *tu;
 536         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 537         char *arg, *filename, *rctr, *rctr_end, *tmp;
 538         char buf[MAX_EVENT_NAME_LEN];
 539         struct path path;
 540         unsigned long offset, ref_ctr_offset;
 541         bool is_return = false;
 542         int i, ret;
 543
 544         ret = 0;
 545         ref_ctr_offset = 0;
 546
 547         switch (argv[0][0]) {
 548         case 'r':
 549                 is_return = true;
 550                 break;
 551         case 'p':
 552                 break;
 553         default:
 554                 return -ECANCELED;
 555         }
 556
 557         if (argc < 2)
 558                 return -ECANCELED;
 559
 560         if (argv[0][1] == ':')
 561                 event = &argv[0][2];
 562
 563         if (!strchr(argv[1], '/'))
 564                 return -ECANCELED;
 565
 566         filename = kstrdup(argv[1], GFP_KERNEL);
 567         if (!filename)
 568                 return -ENOMEM;
 569
 570         /* Find the last occurrence, in case the path contains ':' too. */
 571         arg = strrchr(filename, ':');
 572         if (!arg || !isdigit(arg[1])) {
 573                 kfree(filename);
 574                 return -ECANCELED;
 575         }
 576
 577         trace_probe_log_init("trace_uprobe", argc, argv);
 578         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 579
 580         *arg++ = '\0';
 581         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 582         if (ret) {
 583                 trace_probe_log_err(0, FILE_NOT_FOUND);
 584                 kfree(filename);
 585                 trace_probe_log_clear();
 586                 return ret;
 587         }
 588         if (!d_is_reg(path.dentry)) {
 589                 trace_probe_log_err(0, NO_REGULAR_FILE);
 590                 ret = -EINVAL;
 591                 goto fail_address_parse;
 592         }
 593
 594         /* Parse reference counter offset if specified. */
 595         rctr = strchr(arg, '(');
 596         if (rctr) {
 597                 rctr_end = strchr(rctr, ')');
 598                 if (!rctr_end) {
 599                         ret = -EINVAL;
 600                         rctr_end = rctr + strlen(rctr);
 601                         trace_probe_log_err(rctr_end - filename,
 602                                             REFCNT_OPEN_BRACE);
 603                         goto fail_address_parse;
 604                 } else if (rctr_end[1] != '\0') {
 605                         ret = -EINVAL;
 606                         trace_probe_log_err(rctr_end + 1 - filename,
 607                                             BAD_REFCNT_SUFFIX);
 608                         goto fail_address_parse;
 609                 }
 610
 611                 *rctr++ = '\0';
 612                 *rctr_end = '\0';
 613                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 614                 if (ret) {
 615                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 616                         goto fail_address_parse;
 617                 }
 618         }
 619
 620         /* Check if there is %return suffix */
 621         tmp = strchr(arg, '%');
 622         if (tmp) {
 623                 if (!strcmp(tmp, "%return")) {
 624                         *tmp = '\0';
 625                         is_return = true;
 626                 } else {
 627                         trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 628                         ret = -EINVAL;
 629                         goto fail_address_parse;
 630                 }
 631         }
 632
 633         /* Parse uprobe offset. */
 634         ret = kstrtoul(arg, 0, &offset);
 635         if (ret) {
 636                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 637                 goto fail_address_parse;
 638         }
 639
 640         /* setup a probe */
 641         trace_probe_log_set_index(0);
 642         if (event) {
 643                 ret = traceprobe_parse_event_name(&event, &group, buf,
 644                                                   event - argv[0]);
 645                 if (ret)
 646                         goto fail_address_parse;
 647         } else {
 648                 char *tail;
 649                 char *ptr;
 650
 651                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 652                 if (!tail) {
 653                         ret = -ENOMEM;
 654                         goto fail_address_parse;
 655                 }
 656
 657                 ptr = strpbrk(tail, ".-_");
 658                 if (ptr)
 659                         *ptr = '\0';
 660
 661                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 662                 event = buf;
 663                 kfree(tail);
 664         }
 665
 666         argc -= 2;
 667         argv += 2;
 668
 669         tu = alloc_trace_uprobe(group, event, argc, is_return);
 670         if (IS_ERR(tu)) {
 671                 ret = PTR_ERR(tu);
 672                 /* This must return -ENOMEM otherwise there is a bug */
 673                 WARN_ON_ONCE(ret != -ENOMEM);
 674                 goto fail_address_parse;
 675         }
 676         tu->offset = offset;
 677         tu->ref_ctr_offset = ref_ctr_offset;
 678         tu->path = path;
 679         tu->filename = filename;
 680
 681         /* parse arguments */
 682         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 683                 tmp = kstrdup(argv[i], GFP_KERNEL);
 684                 if (!tmp) {
 685                         ret = -ENOMEM;
 686                         goto error;
 687                 }
 688
 689                 trace_probe_log_set_index(i + 2);
 690                 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 691                                         is_return ? TPARG_FL_RETURN : 0);
 692                 kfree(tmp);
 693                 if (ret)
 694                         goto error;
 695         }
 696
 697         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 698         if (ret < 0)
 699                 goto error;
 700
 701         ret = register_trace_uprobe(tu);
 702         if (!ret)
 703                 goto out;
 704
 705 error:
 706         free_trace_uprobe(tu);
 707 out:
 708         trace_probe_log_clear();
 709         return ret;
 710
 711 fail_address_parse:
 712         trace_probe_log_clear();
 713         path_put(&path);
 714         kfree(filename);
 715
 716         return ret;
 717 }
 718
 719 static int create_or_delete_trace_uprobe(int argc, char **argv)
 720 {
 721         int ret;
 722
 723         if (argv[0][0] == '-')
 724                 return dyn_event_release(argc, argv, &trace_uprobe_ops);
 725
 726         ret = trace_uprobe_create(argc, (const char **)argv);
 727         return ret == -ECANCELED ? -EINVAL : ret;
 728 }
 729
 730 static int trace_uprobe_release(struct dyn_event *ev)
 731 {
 732         struct trace_uprobe *tu = to_trace_uprobe(ev);
 733
 734         return unregister_trace_uprobe(tu);
 735 }
 736
 737 /* Probes listing interfaces */
 738 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 739 {
 740         struct trace_uprobe *tu = to_trace_uprobe(ev);
 741         char c = is_ret_probe(tu) ? 'r' : 'p';
 742         int i;
 743
 744         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 745                         trace_probe_name(&tu->tp), tu->filename,
 746                         (int)(sizeof(void *) * 2), tu->offset);
 747
 748         if (tu->ref_ctr_offset)
 749                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 750
 751         for (i = 0; i < tu->tp.nr_args; i++)
 752                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 753
 754         seq_putc(m, '\n');
 755         return 0;
 756 }
 757
 758 static int probes_seq_show(struct seq_file *m, void *v)
 759 {
 760         struct dyn_event *ev = v;
 761
 762         if (!is_trace_uprobe(ev))
 763                 return 0;
 764
 765         return trace_uprobe_show(m, ev);
 766 }
 767
 768 static const struct seq_operations probes_seq_op = {
 769         .start  = dyn_event_seq_start,
 770         .next   = dyn_event_seq_next,
 771         .stop   = dyn_event_seq_stop,
 772         .show   = probes_seq_show
 773 };
 774
 775 static int probes_open(struct inode *inode, struct file *file)
 776 {
 777         int ret;
 778
 779         ret = security_locked_down(LOCKDOWN_TRACEFS);
 780         if (ret)
 781                 return ret;
 782
 783         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 784                 ret = dyn_events_release_all(&trace_uprobe_ops);
 785                 if (ret)
 786                         return ret;
 787         }
 788
 789         return seq_open(file, &probes_seq_op);
 790 }
 791
 792 static ssize_t probes_write(struct file *file, const char __user *buffer,
 793                             size_t count, loff_t *ppos)
 794 {
 795         return trace_parse_run_command(file, buffer, count, ppos,
 796                                         create_or_delete_trace_uprobe);
 797 }
 798
 799 static const struct file_operations uprobe_events_ops = {
 800         .owner          = THIS_MODULE,
 801         .open           = probes_open,
 802         .read           = seq_read,
 803         .llseek         = seq_lseek,
 804         .release        = seq_release,
 805         .write          = probes_write,
 806 };
 807
 808 /* Probes profiling interfaces */
 809 static int probes_profile_seq_show(struct seq_file *m, void *v)
 810 {
 811         struct dyn_event *ev = v;
 812         struct trace_uprobe *tu;
 813
 814         if (!is_trace_uprobe(ev))
 815                 return 0;
 816
 817         tu = to_trace_uprobe(ev);
 818         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 819                         trace_probe_name(&tu->tp), tu->nhit);
 820         return 0;
 821 }
 822
 823 static const struct seq_operations profile_seq_op = {
 824         .start  = dyn_event_seq_start,
 825         .next   = dyn_event_seq_next,
 826         .stop   = dyn_event_seq_stop,
 827         .show   = probes_profile_seq_show
 828 };
 829
 830 static int profile_open(struct inode *inode, struct file *file)
 831 {
 832         int ret;
 833
 834         ret = security_locked_down(LOCKDOWN_TRACEFS);
 835         if (ret)
 836                 return ret;
 837
 838         return seq_open(file, &profile_seq_op);
 839 }
 840
 841 static const struct file_operations uprobe_profile_ops = {
 842         .owner          = THIS_MODULE,
 843         .open           = profile_open,
 844         .read           = seq_read,
 845         .llseek         = seq_lseek,
 846         .release        = seq_release,
 847 };
 848
 849 struct uprobe_cpu_buffer {
 850         struct mutex mutex;
 851         void *buf;
 852 };
 853 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 854 static int uprobe_buffer_refcnt;
 855
 856 static int uprobe_buffer_init(void)
 857 {
 858         int cpu, err_cpu;
 859
 860         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 861         if (uprobe_cpu_buffer == NULL)
 862                 return -ENOMEM;
 863
 864         for_each_possible_cpu(cpu) {
 865                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 866                                                   GFP_KERNEL, 0);
 867                 if (p == NULL) {
 868                         err_cpu = cpu;
 869                         goto err;
 870                 }
 871                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 872                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 873         }
 874
 875         return 0;
 876
 877 err:
 878         for_each_possible_cpu(cpu) {
 879                 if (cpu == err_cpu)
 880                         break;
 881                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 882         }
 883
 884         free_percpu(uprobe_cpu_buffer);
 885         return -ENOMEM;
 886 }
 887
 888 static int uprobe_buffer_enable(void)
 889 {
 890         int ret = 0;
 891
 892         BUG_ON(!mutex_is_locked(&event_mutex));
 893
 894         if (uprobe_buffer_refcnt++ == 0) {
 895                 ret = uprobe_buffer_init();
 896                 if (ret < 0)
 897                         uprobe_buffer_refcnt--;
 898         }
 899
 900         return ret;
 901 }
 902
 903 static void uprobe_buffer_disable(void)
 904 {
 905         int cpu;
 906
 907         BUG_ON(!mutex_is_locked(&event_mutex));
 908
 909         if (--uprobe_buffer_refcnt == 0) {
 910                 for_each_possible_cpu(cpu)
 911                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 912                                                              cpu)->buf);
 913
 914                 free_percpu(uprobe_cpu_buffer);
 915                 uprobe_cpu_buffer = NULL;
 916         }
 917 }
 918
 919 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 920 {
 921         struct uprobe_cpu_buffer *ucb;
 922         int cpu;
 923
 924         cpu = raw_smp_processor_id();
 925         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 926
 927         /*
 928          * Use per-cpu buffers for fastest access, but we might migrate
 929          * so the mutex makes sure we have sole access to it.
 930          */
 931         mutex_lock(&ucb->mutex);
 932
 933         return ucb;
 934 }
 935
 936 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 937 {
 938         mutex_unlock(&ucb->mutex);
 939 }
 940
 941 static void __uprobe_trace_func(struct trace_uprobe *tu,
 942                                 unsigned long func, struct pt_regs *regs,
 943                                 struct uprobe_cpu_buffer *ucb, int dsize,
 944                                 struct trace_event_file *trace_file)
 945 {
 946         struct uprobe_trace_entry_head *entry;
 947         struct trace_buffer *buffer;
 948         struct ring_buffer_event *event;
 949         void *data;
 950         int size, esize;
 951         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 952
 953         WARN_ON(call != trace_file->event_call);
 954
 955         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 956                 return;
 957
 958         if (trace_trigger_soft_disabled(trace_file))
 959                 return;
 960
 961         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 962         size = esize + tu->tp.size + dsize;
 963         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 964                                                 call->event.type, size, 0, 0);
 965         if (!event)
 966                 return;
 967
 968         entry = ring_buffer_event_data(event);
 969         if (is_ret_probe(tu)) {
 970                 entry->vaddr[0] = func;
 971                 entry->vaddr[1] = instruction_pointer(regs);
 972                 data = DATAOF_TRACE_ENTRY(entry, true);
 973         } else {
 974                 entry->vaddr[0] = instruction_pointer(regs);
 975                 data = DATAOF_TRACE_ENTRY(entry, false);
 976         }
 977
 978         memcpy(data, ucb->buf, tu->tp.size + dsize);
 979
 980         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 981 }
 982
 983 /* uprobe handler */
 984 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 985                              struct uprobe_cpu_buffer *ucb, int dsize)
 986 {
 987         struct event_file_link *link;
 988
 989         if (is_ret_probe(tu))
 990                 return 0;
 991
 992         rcu_read_lock();
 993         trace_probe_for_each_link_rcu(link, &tu->tp)
 994                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 995         rcu_read_unlock();
 996
 997         return 0;
 998 }
 999
1000 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1001                                  struct pt_regs *regs,
1002                                  struct uprobe_cpu_buffer *ucb, int dsize)
1003 {
1004         struct event_file_link *link;
1005
1006         rcu_read_lock();
1007         trace_probe_for_each_link_rcu(link, &tu->tp)
1008                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1009         rcu_read_unlock();
1010 }
1011
1012 /* Event entry printers */
1013 static enum print_line_t
1014 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1015 {
1016         struct uprobe_trace_entry_head *entry;
1017         struct trace_seq *s = &iter->seq;
1018         struct trace_uprobe *tu;
1019         u8 *data;
1020
1021         entry = (struct uprobe_trace_entry_head *)iter->ent;
1022         tu = trace_uprobe_primary_from_call(
1023                 container_of(event, struct trace_event_call, event));
1024         if (unlikely(!tu))
1025                 goto out;
1026
1027         if (is_ret_probe(tu)) {
1028                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1029                                  trace_probe_name(&tu->tp),
1030                                  entry->vaddr[1], entry->vaddr[0]);
1031                 data = DATAOF_TRACE_ENTRY(entry, true);
1032         } else {
1033                 trace_seq_printf(s, "%s: (0x%lx)",
1034                                  trace_probe_name(&tu->tp),
1035                                  entry->vaddr[0]);
1036                 data = DATAOF_TRACE_ENTRY(entry, false);
1037         }
1038
1039         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1040                 goto out;
1041
1042         trace_seq_putc(s, '\n');
1043
1044  out:
1045         return trace_handle_return(s);
1046 }
1047
1048 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1049                                 enum uprobe_filter_ctx ctx,
1050                                 struct mm_struct *mm);
1051
1052 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1053 {
1054         int ret;
1055
1056         tu->consumer.filter = filter;
1057         tu->inode = d_real_inode(tu->path.dentry);
1058
1059         if (tu->ref_ctr_offset)
1060                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1061                                 tu->ref_ctr_offset, &tu->consumer);
1062         else
1063                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1064
1065         if (ret)
1066                 tu->inode = NULL;
1067
1068         return ret;
1069 }
1070
1071 static void __probe_event_disable(struct trace_probe *tp)
1072 {
1073         struct trace_probe *pos;
1074         struct trace_uprobe *tu;
1075
1076         tu = container_of(tp, struct trace_uprobe, tp);
1077         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1078
1079         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1080                 tu = container_of(pos, struct trace_uprobe, tp);
1081                 if (!tu->inode)
1082                         continue;
1083
1084                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1085                 tu->inode = NULL;
1086         }
1087 }
1088
1089 static int probe_event_enable(struct trace_event_call *call,
1090                         struct trace_event_file *file, filter_func_t filter)
1091 {
1092         struct trace_probe *pos, *tp;
1093         struct trace_uprobe *tu;
1094         bool enabled;
1095         int ret;
1096
1097         tp = trace_probe_primary_from_call(call);
1098         if (WARN_ON_ONCE(!tp))
1099                 return -ENODEV;
1100         enabled = trace_probe_is_enabled(tp);
1101
1102         /* This may also change "enabled" state */
1103         if (file) {
1104                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1105                         return -EINTR;
1106
1107                 ret = trace_probe_add_file(tp, file);
1108                 if (ret < 0)
1109                         return ret;
1110         } else {
1111                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1112                         return -EINTR;
1113
1114                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1115         }
1116
1117         tu = container_of(tp, struct trace_uprobe, tp);
1118         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1119
1120         if (enabled)
1121                 return 0;
1122
1123         ret = uprobe_buffer_enable();
1124         if (ret)
1125                 goto err_flags;
1126
1127         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1128                 tu = container_of(pos, struct trace_uprobe, tp);
1129                 ret = trace_uprobe_enable(tu, filter);
1130                 if (ret) {
1131                         __probe_event_disable(tp);
1132                         goto err_buffer;
1133                 }
1134         }
1135
1136         return 0;
1137
1138  err_buffer:
1139         uprobe_buffer_disable();
1140
1141  err_flags:
1142         if (file)
1143                 trace_probe_remove_file(tp, file);
1144         else
1145                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1146
1147         return ret;
1148 }
1149
1150 static void probe_event_disable(struct trace_event_call *call,
1151                                 struct trace_event_file *file)
1152 {
1153         struct trace_probe *tp;
1154
1155         tp = trace_probe_primary_from_call(call);
1156         if (WARN_ON_ONCE(!tp))
1157                 return;
1158
1159         if (!trace_probe_is_enabled(tp))
1160                 return;
1161
1162         if (file) {
1163                 if (trace_probe_remove_file(tp, file) < 0)
1164                         return;
1165
1166                 if (trace_probe_is_enabled(tp))
1167                         return;
1168         } else
1169                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1170
1171         __probe_event_disable(tp);
1172         uprobe_buffer_disable();
1173 }
1174
1175 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1176 {
1177         int ret, size;
1178         struct uprobe_trace_entry_head field;
1179         struct trace_uprobe *tu;
1180
1181         tu = trace_uprobe_primary_from_call(event_call);
1182         if (unlikely(!tu))
1183                 return -ENODEV;
1184
1185         if (is_ret_probe(tu)) {
1186                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1187                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1188                 size = SIZEOF_TRACE_ENTRY(true);
1189         } else {
1190                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1191                 size = SIZEOF_TRACE_ENTRY(false);
1192         }
1193
1194         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1195 }
1196
1197 #ifdef CONFIG_PERF_EVENTS
1198 static bool
1199 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1200 {
1201         struct perf_event *event;
1202
1203         if (filter->nr_systemwide)
1204                 return true;
1205
1206         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1207                 if (event->hw.target->mm == mm)
1208                         return true;
1209         }
1210
1211         return false;
1212 }
1213
1214 static inline bool
1215 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1216                           struct perf_event *event)
1217 {
1218         return __uprobe_perf_filter(filter, event->hw.target->mm);
1219 }
1220
1221 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1222                                        struct perf_event *event)
1223 {
1224         bool done;
1225
1226         write_lock(&filter->rwlock);
1227         if (event->hw.target) {
1228                 list_del(&event->hw.tp_list);
1229                 done = filter->nr_systemwide ||
1230                         (event->hw.target->flags & PF_EXITING) ||
1231                         trace_uprobe_filter_event(filter, event);
1232         } else {
1233                 filter->nr_systemwide--;
1234                 done = filter->nr_systemwide;
1235         }
1236         write_unlock(&filter->rwlock);
1237
1238         return done;
1239 }
1240
1241 /* This returns true if the filter always covers target mm */
1242 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1243                                     struct perf_event *event)
1244 {
1245         bool done;
1246
1247         write_lock(&filter->rwlock);
1248         if (event->hw.target) {
1249                 /*
1250                  * event->parent != NULL means copy_process(), we can avoid
1251                  * uprobe_apply(). current->mm must be probed and we can rely
1252                  * on dup_mmap() which preserves the already installed bp's.
1253                  *
1254                  * attr.enable_on_exec means that exec/mmap will install the
1255                  * breakpoints we need.
1256                  */
1257                 done = filter->nr_systemwide ||
1258                         event->parent || event->attr.enable_on_exec ||
1259                         trace_uprobe_filter_event(filter, event);
1260                 list_add(&event->hw.tp_list, &filter->perf_events);
1261         } else {
1262                 done = filter->nr_systemwide;
1263                 filter->nr_systemwide++;
1264         }
1265         write_unlock(&filter->rwlock);
1266
1267         return done;
1268 }
1269
1270 static int uprobe_perf_close(struct trace_event_call *call,
1271                              struct perf_event *event)
1272 {
1273         struct trace_probe *pos, *tp;
1274         struct trace_uprobe *tu;
1275         int ret = 0;
1276
1277         tp = trace_probe_primary_from_call(call);
1278         if (WARN_ON_ONCE(!tp))
1279                 return -ENODEV;
1280
1281         tu = container_of(tp, struct trace_uprobe, tp);
1282         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1283                 return 0;
1284
1285         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1286                 tu = container_of(pos, struct trace_uprobe, tp);
1287                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1288                 if (ret)
1289                         break;
1290         }
1291
1292         return ret;
1293 }
1294
1295 static int uprobe_perf_open(struct trace_event_call *call,
1296                             struct perf_event *event)
1297 {
1298         struct trace_probe *pos, *tp;
1299         struct trace_uprobe *tu;
1300         int err = 0;
1301
1302         tp = trace_probe_primary_from_call(call);
1303         if (WARN_ON_ONCE(!tp))
1304                 return -ENODEV;
1305
1306         tu = container_of(tp, struct trace_uprobe, tp);
1307         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1308                 return 0;
1309
1310         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1311                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1312                 if (err) {
1313                         uprobe_perf_close(call, event);
1314                         break;
1315                 }
1316         }
1317
1318         return err;
1319 }
1320
1321 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1322                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1323 {
1324         struct trace_uprobe_filter *filter;
1325         struct trace_uprobe *tu;
1326         int ret;
1327
1328         tu = container_of(uc, struct trace_uprobe, consumer);
1329         filter = tu->tp.event->filter;
1330
1331         read_lock(&filter->rwlock);
1332         ret = __uprobe_perf_filter(filter, mm);
1333         read_unlock(&filter->rwlock);
1334
1335         return ret;
1336 }
1337
1338 static void __uprobe_perf_func(struct trace_uprobe *tu,
1339                                unsigned long func, struct pt_regs *regs,
1340                                struct uprobe_cpu_buffer *ucb, int dsize)
1341 {
1342         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1343         struct uprobe_trace_entry_head *entry;
1344         struct hlist_head *head;
1345         void *data;
1346         int size, esize;
1347         int rctx;
1348
1349         if (bpf_prog_array_valid(call)) {
1350                 u32 ret;
1351
1352                 preempt_disable();
1353                 ret = trace_call_bpf(call, regs);
1354                 preempt_enable();
1355                 if (!ret)
1356                         return;
1357         }
1358
1359         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1360
1361         size = esize + tu->tp.size + dsize;
1362         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1363         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1364                 return;
1365
1366         preempt_disable();
1367         head = this_cpu_ptr(call->perf_events);
1368         if (hlist_empty(head))
1369                 goto out;
1370
1371         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1372         if (!entry)
1373                 goto out;
1374
1375         if (is_ret_probe(tu)) {
1376                 entry->vaddr[0] = func;
1377                 entry->vaddr[1] = instruction_pointer(regs);
1378                 data = DATAOF_TRACE_ENTRY(entry, true);
1379         } else {
1380                 entry->vaddr[0] = instruction_pointer(regs);
1381                 data = DATAOF_TRACE_ENTRY(entry, false);
1382         }
1383
1384         memcpy(data, ucb->buf, tu->tp.size + dsize);
1385
1386         if (size - esize > tu->tp.size + dsize) {
1387                 int len = tu->tp.size + dsize;
1388
1389                 memset(data + len, 0, size - esize - len);
1390         }
1391
1392         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1393                               head, NULL);
1394  out:
1395         preempt_enable();
1396 }
1397
1398 /* uprobe profile handler */
1399 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1400                             struct uprobe_cpu_buffer *ucb, int dsize)
1401 {
1402         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1403                 return UPROBE_HANDLER_REMOVE;
1404
1405         if (!is_ret_probe(tu))
1406                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1407         return 0;
1408 }
1409
1410 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1411                                 struct pt_regs *regs,
1412                                 struct uprobe_cpu_buffer *ucb, int dsize)
1413 {
1414         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1415 }
1416
1417 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1418                         const char **filename, u64 *probe_offset,
1419                         bool perf_type_tracepoint)
1420 {
1421         const char *pevent = trace_event_name(event->tp_event);
1422         const char *group = event->tp_event->class->system;
1423         struct trace_uprobe *tu;
1424
1425         if (perf_type_tracepoint)
1426                 tu = find_probe_event(pevent, group);
1427         else
1428                 tu = trace_uprobe_primary_from_call(event->tp_event);
1429         if (!tu)
1430                 return -EINVAL;
1431
1432         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1433                                     : BPF_FD_TYPE_UPROBE;
1434         *filename = tu->filename;
1435         *probe_offset = tu->offset;
1436         return 0;
1437 }
1438 #endif  /* CONFIG_PERF_EVENTS */
1439
1440 static int
1441 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1442                       void *data)
1443 {
1444         struct trace_event_file *file = data;
1445
1446         switch (type) {
1447         case TRACE_REG_REGISTER:
1448                 return probe_event_enable(event, file, NULL);
1449
1450         case TRACE_REG_UNREGISTER:
1451                 probe_event_disable(event, file);
1452                 return 0;
1453
1454 #ifdef CONFIG_PERF_EVENTS
1455         case TRACE_REG_PERF_REGISTER:
1456                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1457
1458         case TRACE_REG_PERF_UNREGISTER:
1459                 probe_event_disable(event, NULL);
1460                 return 0;
1461
1462         case TRACE_REG_PERF_OPEN:
1463                 return uprobe_perf_open(event, data);
1464
1465         case TRACE_REG_PERF_CLOSE:
1466                 return uprobe_perf_close(event, data);
1467
1468 #endif
1469         default:
1470                 return 0;
1471         }
1472 }
1473
1474 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1475 {
1476         struct trace_uprobe *tu;
1477         struct uprobe_dispatch_data udd;
1478         struct uprobe_cpu_buffer *ucb;
1479         int dsize, esize;
1480         int ret = 0;
1481
1482
1483         tu = container_of(con, struct trace_uprobe, consumer);
1484         tu->nhit++;
1485
1486         udd.tu = tu;
1487         udd.bp_addr = instruction_pointer(regs);
1488
1489         current->utask->vaddr = (unsigned long) &udd;
1490
1491         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1492                 return 0;
1493
1494         dsize = __get_data_size(&tu->tp, regs);
1495         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1496
1497         ucb = uprobe_buffer_get();
1498         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1499
1500         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1501                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1502
1503 #ifdef CONFIG_PERF_EVENTS
1504         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1505                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1506 #endif
1507         uprobe_buffer_put(ucb);
1508         return ret;
1509 }
1510
1511 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1512                                 unsigned long func, struct pt_regs *regs)
1513 {
1514         struct trace_uprobe *tu;
1515         struct uprobe_dispatch_data udd;
1516         struct uprobe_cpu_buffer *ucb;
1517         int dsize, esize;
1518
1519         tu = container_of(con, struct trace_uprobe, consumer);
1520
1521         udd.tu = tu;
1522         udd.bp_addr = func;
1523
1524         current->utask->vaddr = (unsigned long) &udd;
1525
1526         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1527                 return 0;
1528
1529         dsize = __get_data_size(&tu->tp, regs);
1530         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1531
1532         ucb = uprobe_buffer_get();
1533         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1534
1535         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1536                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1537
1538 #ifdef CONFIG_PERF_EVENTS
1539         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1540                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1541 #endif
1542         uprobe_buffer_put(ucb);
1543         return 0;
1544 }
1545
1546 static struct trace_event_functions uprobe_funcs = {
1547         .trace          = print_uprobe_event
1548 };
1549
1550 static struct trace_event_fields uprobe_fields_array[] = {
1551         { .type = TRACE_FUNCTION_TYPE,
1552           .define_fields = uprobe_event_define_fields },
1553         {}
1554 };
1555
1556 static inline void init_trace_event_call(struct trace_uprobe *tu)
1557 {
1558         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1559         call->event.funcs = &uprobe_funcs;
1560         call->class->fields_array = uprobe_fields_array;
1561
1562         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1563         call->class->reg = trace_uprobe_register;
1564 }
1565
1566 static int register_uprobe_event(struct trace_uprobe *tu)
1567 {
1568         init_trace_event_call(tu);
1569
1570         return trace_probe_register_event_call(&tu->tp);
1571 }
1572
1573 static int unregister_uprobe_event(struct trace_uprobe *tu)
1574 {
1575         return trace_probe_unregister_event_call(&tu->tp);
1576 }
1577
1578 #ifdef CONFIG_PERF_EVENTS
1579 struct trace_event_call *
1580 create_local_trace_uprobe(char *name, unsigned long offs,
1581                           unsigned long ref_ctr_offset, bool is_return)
1582 {
1583         struct trace_uprobe *tu;
1584         struct path path;
1585         int ret;
1586
1587         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1588         if (ret)
1589                 return ERR_PTR(ret);
1590
1591         if (!d_is_reg(path.dentry)) {
1592                 path_put(&path);
1593                 return ERR_PTR(-EINVAL);
1594         }
1595
1596         /*
1597          * local trace_kprobes are not added to dyn_event, so they are never
1598          * searched in find_trace_kprobe(). Therefore, there is no concern of
1599          * duplicated name "DUMMY_EVENT" here.
1600          */
1601         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1602                                 is_return);
1603
1604         if (IS_ERR(tu)) {
1605                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1606                         (int)PTR_ERR(tu));
1607                 path_put(&path);
1608                 return ERR_CAST(tu);
1609         }
1610
1611         tu->offset = offs;
1612         tu->path = path;
1613         tu->ref_ctr_offset = ref_ctr_offset;
1614         tu->filename = kstrdup(name, GFP_KERNEL);
1615         init_trace_event_call(tu);
1616
1617         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1618                 ret = -ENOMEM;
1619                 goto error;
1620         }
1621
1622         return trace_probe_event_call(&tu->tp);
1623 error:
1624         free_trace_uprobe(tu);
1625         return ERR_PTR(ret);
1626 }
1627
1628 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1629 {
1630         struct trace_uprobe *tu;
1631
1632         tu = trace_uprobe_primary_from_call(event_call);
1633
1634         free_trace_uprobe(tu);
1635 }
1636 #endif /* CONFIG_PERF_EVENTS */
1637
1638 /* Make a trace interface for controling probe points */
1639 static __init int init_uprobe_trace(void)
1640 {
1641         int ret;
1642
1643         ret = dyn_event_register(&trace_uprobe_ops);
1644         if (ret)
1645                 return ret;
1646
1647         ret = tracing_init_dentry();
1648         if (ret)
1649                 return 0;
1650
1651         trace_create_file("uprobe_events", 0644, NULL,
1652                                     NULL, &uprobe_events_ops);
1653         /* Profile interface */
1654         trace_create_file("uprobe_profile", 0444, NULL,
1655                                     NULL, &uprobe_profile_ops);
1656         return 0;
1657 }
1658
1659 fs_initcall(init_uprobe_trace);