kernel/trace/trace_uprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * uprobes-based tracing events
   4  *
   5  * Copyright (C) IBM Corporation, 2010-2012
   6  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
   7  */
   8 #define pr_fmt(fmt)     "trace_uprobe: " fmt
   9
  10 #include <linux/security.h>
  11 #include <linux/ctype.h>
  12 #include <linux/module.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/uprobes.h>
  15 #include <linux/namei.h>
  16 #include <linux/string.h>
  17 #include <linux/rculist.h>
  18
  19 #include "trace_dynevent.h"
  20 #include "trace_probe.h"
  21 #include "trace_probe_tmpl.h"
  22
  23 #define UPROBE_EVENT_SYSTEM     "uprobes"
  24
  25 struct uprobe_trace_entry_head {
  26         struct trace_entry      ent;
  27         unsigned long           vaddr[];
  28 };
  29
  30 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  31         (sizeof(struct uprobe_trace_entry_head) +       \
  32          sizeof(unsigned long) * (is_return ? 2 : 1))
  33
  34 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  35         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  36
  37 static int trace_uprobe_create(const char *raw_command);
  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 int trace_uprobe_create(const char *raw_command)
 720 {
 721         return trace_probe_create(raw_command, __trace_uprobe_create);
 722 }
 723
 724 static int create_or_delete_trace_uprobe(const char *raw_command)
 725 {
 726         int ret;
 727
 728         if (raw_command[0] == '-')
 729                 return dyn_event_release(raw_command, &trace_uprobe_ops);
 730
 731         ret = trace_uprobe_create(raw_command);
 732         return ret == -ECANCELED ? -EINVAL : ret;
 733 }
 734
 735 static int trace_uprobe_release(struct dyn_event *ev)
 736 {
 737         struct trace_uprobe *tu = to_trace_uprobe(ev);
 738
 739         return unregister_trace_uprobe(tu);
 740 }
 741
 742 /* Probes listing interfaces */
 743 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 744 {
 745         struct trace_uprobe *tu = to_trace_uprobe(ev);
 746         char c = is_ret_probe(tu) ? 'r' : 'p';
 747         int i;
 748
 749         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 750                         trace_probe_name(&tu->tp), tu->filename,
 751                         (int)(sizeof(void *) * 2), tu->offset);
 752
 753         if (tu->ref_ctr_offset)
 754                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 755
 756         for (i = 0; i < tu->tp.nr_args; i++)
 757                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 758
 759         seq_putc(m, '\n');
 760         return 0;
 761 }
 762
 763 static int probes_seq_show(struct seq_file *m, void *v)
 764 {
 765         struct dyn_event *ev = v;
 766
 767         if (!is_trace_uprobe(ev))
 768                 return 0;
 769
 770         return trace_uprobe_show(m, ev);
 771 }
 772
 773 static const struct seq_operations probes_seq_op = {
 774         .start  = dyn_event_seq_start,
 775         .next   = dyn_event_seq_next,
 776         .stop   = dyn_event_seq_stop,
 777         .show   = probes_seq_show
 778 };
 779
 780 static int probes_open(struct inode *inode, struct file *file)
 781 {
 782         int ret;
 783
 784         ret = security_locked_down(LOCKDOWN_TRACEFS);
 785         if (ret)
 786                 return ret;
 787
 788         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 789                 ret = dyn_events_release_all(&trace_uprobe_ops);
 790                 if (ret)
 791                         return ret;
 792         }
 793
 794         return seq_open(file, &probes_seq_op);
 795 }
 796
 797 static ssize_t probes_write(struct file *file, const char __user *buffer,
 798                             size_t count, loff_t *ppos)
 799 {
 800         return trace_parse_run_command(file, buffer, count, ppos,
 801                                         create_or_delete_trace_uprobe);
 802 }
 803
 804 static const struct file_operations uprobe_events_ops = {
 805         .owner          = THIS_MODULE,
 806         .open           = probes_open,
 807         .read           = seq_read,
 808         .llseek         = seq_lseek,
 809         .release        = seq_release,
 810         .write          = probes_write,
 811 };
 812
 813 /* Probes profiling interfaces */
 814 static int probes_profile_seq_show(struct seq_file *m, void *v)
 815 {
 816         struct dyn_event *ev = v;
 817         struct trace_uprobe *tu;
 818
 819         if (!is_trace_uprobe(ev))
 820                 return 0;
 821
 822         tu = to_trace_uprobe(ev);
 823         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 824                         trace_probe_name(&tu->tp), tu->nhit);
 825         return 0;
 826 }
 827
 828 static const struct seq_operations profile_seq_op = {
 829         .start  = dyn_event_seq_start,
 830         .next   = dyn_event_seq_next,
 831         .stop   = dyn_event_seq_stop,
 832         .show   = probes_profile_seq_show
 833 };
 834
 835 static int profile_open(struct inode *inode, struct file *file)
 836 {
 837         int ret;
 838
 839         ret = security_locked_down(LOCKDOWN_TRACEFS);
 840         if (ret)
 841                 return ret;
 842
 843         return seq_open(file, &profile_seq_op);
 844 }
 845
 846 static const struct file_operations uprobe_profile_ops = {
 847         .owner          = THIS_MODULE,
 848         .open           = profile_open,
 849         .read           = seq_read,
 850         .llseek         = seq_lseek,
 851         .release        = seq_release,
 852 };
 853
 854 struct uprobe_cpu_buffer {
 855         struct mutex mutex;
 856         void *buf;
 857 };
 858 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 859 static int uprobe_buffer_refcnt;
 860
 861 static int uprobe_buffer_init(void)
 862 {
 863         int cpu, err_cpu;
 864
 865         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 866         if (uprobe_cpu_buffer == NULL)
 867                 return -ENOMEM;
 868
 869         for_each_possible_cpu(cpu) {
 870                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 871                                                   GFP_KERNEL, 0);
 872                 if (p == NULL) {
 873                         err_cpu = cpu;
 874                         goto err;
 875                 }
 876                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 877                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 878         }
 879
 880         return 0;
 881
 882 err:
 883         for_each_possible_cpu(cpu) {
 884                 if (cpu == err_cpu)
 885                         break;
 886                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 887         }
 888
 889         free_percpu(uprobe_cpu_buffer);
 890         return -ENOMEM;
 891 }
 892
 893 static int uprobe_buffer_enable(void)
 894 {
 895         int ret = 0;
 896
 897         BUG_ON(!mutex_is_locked(&event_mutex));
 898
 899         if (uprobe_buffer_refcnt++ == 0) {
 900                 ret = uprobe_buffer_init();
 901                 if (ret < 0)
 902                         uprobe_buffer_refcnt--;
 903         }
 904
 905         return ret;
 906 }
 907
 908 static void uprobe_buffer_disable(void)
 909 {
 910         int cpu;
 911
 912         BUG_ON(!mutex_is_locked(&event_mutex));
 913
 914         if (--uprobe_buffer_refcnt == 0) {
 915                 for_each_possible_cpu(cpu)
 916                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 917                                                              cpu)->buf);
 918
 919                 free_percpu(uprobe_cpu_buffer);
 920                 uprobe_cpu_buffer = NULL;
 921         }
 922 }
 923
 924 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 925 {
 926         struct uprobe_cpu_buffer *ucb;
 927         int cpu;
 928
 929         cpu = raw_smp_processor_id();
 930         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 931
 932         /*
 933          * Use per-cpu buffers for fastest access, but we might migrate
 934          * so the mutex makes sure we have sole access to it.
 935          */
 936         mutex_lock(&ucb->mutex);
 937
 938         return ucb;
 939 }
 940
 941 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 942 {
 943         mutex_unlock(&ucb->mutex);
 944 }
 945
 946 static void __uprobe_trace_func(struct trace_uprobe *tu,
 947                                 unsigned long func, struct pt_regs *regs,
 948                                 struct uprobe_cpu_buffer *ucb, int dsize,
 949                                 struct trace_event_file *trace_file)
 950 {
 951         struct uprobe_trace_entry_head *entry;
 952         struct trace_buffer *buffer;
 953         struct ring_buffer_event *event;
 954         void *data;
 955         int size, esize;
 956         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 957
 958         WARN_ON(call != trace_file->event_call);
 959
 960         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 961                 return;
 962
 963         if (trace_trigger_soft_disabled(trace_file))
 964                 return;
 965
 966         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 967         size = esize + tu->tp.size + dsize;
 968         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 969                                                 call->event.type, size, 0);
 970         if (!event)
 971                 return;
 972
 973         entry = ring_buffer_event_data(event);
 974         if (is_ret_probe(tu)) {
 975                 entry->vaddr[0] = func;
 976                 entry->vaddr[1] = instruction_pointer(regs);
 977                 data = DATAOF_TRACE_ENTRY(entry, true);
 978         } else {
 979                 entry->vaddr[0] = instruction_pointer(regs);
 980                 data = DATAOF_TRACE_ENTRY(entry, false);
 981         }
 982
 983         memcpy(data, ucb->buf, tu->tp.size + dsize);
 984
 985         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
 986 }
 987
 988 /* uprobe handler */
 989 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 990                              struct uprobe_cpu_buffer *ucb, int dsize)
 991 {
 992         struct event_file_link *link;
 993
 994         if (is_ret_probe(tu))
 995                 return 0;
 996
 997         rcu_read_lock();
 998         trace_probe_for_each_link_rcu(link, &tu->tp)
 999                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1000         rcu_read_unlock();
1001
1002         return 0;
1003 }
1004
1005 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1006                                  struct pt_regs *regs,
1007                                  struct uprobe_cpu_buffer *ucb, int dsize)
1008 {
1009         struct event_file_link *link;
1010
1011         rcu_read_lock();
1012         trace_probe_for_each_link_rcu(link, &tu->tp)
1013                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1014         rcu_read_unlock();
1015 }
1016
1017 /* Event entry printers */
1018 static enum print_line_t
1019 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1020 {
1021         struct uprobe_trace_entry_head *entry;
1022         struct trace_seq *s = &iter->seq;
1023         struct trace_uprobe *tu;
1024         u8 *data;
1025
1026         entry = (struct uprobe_trace_entry_head *)iter->ent;
1027         tu = trace_uprobe_primary_from_call(
1028                 container_of(event, struct trace_event_call, event));
1029         if (unlikely(!tu))
1030                 goto out;
1031
1032         if (is_ret_probe(tu)) {
1033                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1034                                  trace_probe_name(&tu->tp),
1035                                  entry->vaddr[1], entry->vaddr[0]);
1036                 data = DATAOF_TRACE_ENTRY(entry, true);
1037         } else {
1038                 trace_seq_printf(s, "%s: (0x%lx)",
1039                                  trace_probe_name(&tu->tp),
1040                                  entry->vaddr[0]);
1041                 data = DATAOF_TRACE_ENTRY(entry, false);
1042         }
1043
1044         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1045                 goto out;
1046
1047         trace_seq_putc(s, '\n');
1048
1049  out:
1050         return trace_handle_return(s);
1051 }
1052
1053 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1054                                 enum uprobe_filter_ctx ctx,
1055                                 struct mm_struct *mm);
1056
1057 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1058 {
1059         int ret;
1060
1061         tu->consumer.filter = filter;
1062         tu->inode = d_real_inode(tu->path.dentry);
1063
1064         if (tu->ref_ctr_offset)
1065                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1066                                 tu->ref_ctr_offset, &tu->consumer);
1067         else
1068                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1069
1070         if (ret)
1071                 tu->inode = NULL;
1072
1073         return ret;
1074 }
1075
1076 static void __probe_event_disable(struct trace_probe *tp)
1077 {
1078         struct trace_probe *pos;
1079         struct trace_uprobe *tu;
1080
1081         tu = container_of(tp, struct trace_uprobe, tp);
1082         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1083
1084         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1085                 tu = container_of(pos, struct trace_uprobe, tp);
1086                 if (!tu->inode)
1087                         continue;
1088
1089                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1090                 tu->inode = NULL;
1091         }
1092 }
1093
1094 static int probe_event_enable(struct trace_event_call *call,
1095                         struct trace_event_file *file, filter_func_t filter)
1096 {
1097         struct trace_probe *pos, *tp;
1098         struct trace_uprobe *tu;
1099         bool enabled;
1100         int ret;
1101
1102         tp = trace_probe_primary_from_call(call);
1103         if (WARN_ON_ONCE(!tp))
1104                 return -ENODEV;
1105         enabled = trace_probe_is_enabled(tp);
1106
1107         /* This may also change "enabled" state */
1108         if (file) {
1109                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1110                         return -EINTR;
1111
1112                 ret = trace_probe_add_file(tp, file);
1113                 if (ret < 0)
1114                         return ret;
1115         } else {
1116                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1117                         return -EINTR;
1118
1119                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1120         }
1121
1122         tu = container_of(tp, struct trace_uprobe, tp);
1123         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1124
1125         if (enabled)
1126                 return 0;
1127
1128         ret = uprobe_buffer_enable();
1129         if (ret)
1130                 goto err_flags;
1131
1132         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1133                 tu = container_of(pos, struct trace_uprobe, tp);
1134                 ret = trace_uprobe_enable(tu, filter);
1135                 if (ret) {
1136                         __probe_event_disable(tp);
1137                         goto err_buffer;
1138                 }
1139         }
1140
1141         return 0;
1142
1143  err_buffer:
1144         uprobe_buffer_disable();
1145
1146  err_flags:
1147         if (file)
1148                 trace_probe_remove_file(tp, file);
1149         else
1150                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1151
1152         return ret;
1153 }
1154
1155 static void probe_event_disable(struct trace_event_call *call,
1156                                 struct trace_event_file *file)
1157 {
1158         struct trace_probe *tp;
1159
1160         tp = trace_probe_primary_from_call(call);
1161         if (WARN_ON_ONCE(!tp))
1162                 return;
1163
1164         if (!trace_probe_is_enabled(tp))
1165                 return;
1166
1167         if (file) {
1168                 if (trace_probe_remove_file(tp, file) < 0)
1169                         return;
1170
1171                 if (trace_probe_is_enabled(tp))
1172                         return;
1173         } else
1174                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1175
1176         __probe_event_disable(tp);
1177         uprobe_buffer_disable();
1178 }
1179
1180 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1181 {
1182         int ret, size;
1183         struct uprobe_trace_entry_head field;
1184         struct trace_uprobe *tu;
1185
1186         tu = trace_uprobe_primary_from_call(event_call);
1187         if (unlikely(!tu))
1188                 return -ENODEV;
1189
1190         if (is_ret_probe(tu)) {
1191                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1192                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1193                 size = SIZEOF_TRACE_ENTRY(true);
1194         } else {
1195                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1196                 size = SIZEOF_TRACE_ENTRY(false);
1197         }
1198
1199         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1200 }
1201
1202 #ifdef CONFIG_PERF_EVENTS
1203 static bool
1204 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1205 {
1206         struct perf_event *event;
1207
1208         if (filter->nr_systemwide)
1209                 return true;
1210
1211         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1212                 if (event->hw.target->mm == mm)
1213                         return true;
1214         }
1215
1216         return false;
1217 }
1218
1219 static inline bool
1220 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1221                           struct perf_event *event)
1222 {
1223         return __uprobe_perf_filter(filter, event->hw.target->mm);
1224 }
1225
1226 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1227                                        struct perf_event *event)
1228 {
1229         bool done;
1230
1231         write_lock(&filter->rwlock);
1232         if (event->hw.target) {
1233                 list_del(&event->hw.tp_list);
1234                 done = filter->nr_systemwide ||
1235                         (event->hw.target->flags & PF_EXITING) ||
1236                         trace_uprobe_filter_event(filter, event);
1237         } else {
1238                 filter->nr_systemwide--;
1239                 done = filter->nr_systemwide;
1240         }
1241         write_unlock(&filter->rwlock);
1242
1243         return done;
1244 }
1245
1246 /* This returns true if the filter always covers target mm */
1247 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1248                                     struct perf_event *event)
1249 {
1250         bool done;
1251
1252         write_lock(&filter->rwlock);
1253         if (event->hw.target) {
1254                 /*
1255                  * event->parent != NULL means copy_process(), we can avoid
1256                  * uprobe_apply(). current->mm must be probed and we can rely
1257                  * on dup_mmap() which preserves the already installed bp's.
1258                  *
1259                  * attr.enable_on_exec means that exec/mmap will install the
1260                  * breakpoints we need.
1261                  */
1262                 done = filter->nr_systemwide ||
1263                         event->parent || event->attr.enable_on_exec ||
1264                         trace_uprobe_filter_event(filter, event);
1265                 list_add(&event->hw.tp_list, &filter->perf_events);
1266         } else {
1267                 done = filter->nr_systemwide;
1268                 filter->nr_systemwide++;
1269         }
1270         write_unlock(&filter->rwlock);
1271
1272         return done;
1273 }
1274
1275 static int uprobe_perf_close(struct trace_event_call *call,
1276                              struct perf_event *event)
1277 {
1278         struct trace_probe *pos, *tp;
1279         struct trace_uprobe *tu;
1280         int ret = 0;
1281
1282         tp = trace_probe_primary_from_call(call);
1283         if (WARN_ON_ONCE(!tp))
1284                 return -ENODEV;
1285
1286         tu = container_of(tp, struct trace_uprobe, tp);
1287         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1288                 return 0;
1289
1290         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1291                 tu = container_of(pos, struct trace_uprobe, tp);
1292                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1293                 if (ret)
1294                         break;
1295         }
1296
1297         return ret;
1298 }
1299
1300 static int uprobe_perf_open(struct trace_event_call *call,
1301                             struct perf_event *event)
1302 {
1303         struct trace_probe *pos, *tp;
1304         struct trace_uprobe *tu;
1305         int err = 0;
1306
1307         tp = trace_probe_primary_from_call(call);
1308         if (WARN_ON_ONCE(!tp))
1309                 return -ENODEV;
1310
1311         tu = container_of(tp, struct trace_uprobe, tp);
1312         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1313                 return 0;
1314
1315         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1316                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1317                 if (err) {
1318                         uprobe_perf_close(call, event);
1319                         break;
1320                 }
1321         }
1322
1323         return err;
1324 }
1325
1326 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1327                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1328 {
1329         struct trace_uprobe_filter *filter;
1330         struct trace_uprobe *tu;
1331         int ret;
1332
1333         tu = container_of(uc, struct trace_uprobe, consumer);
1334         filter = tu->tp.event->filter;
1335
1336         read_lock(&filter->rwlock);
1337         ret = __uprobe_perf_filter(filter, mm);
1338         read_unlock(&filter->rwlock);
1339
1340         return ret;
1341 }
1342
1343 static void __uprobe_perf_func(struct trace_uprobe *tu,
1344                                unsigned long func, struct pt_regs *regs,
1345                                struct uprobe_cpu_buffer *ucb, int dsize)
1346 {
1347         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1348         struct uprobe_trace_entry_head *entry;
1349         struct hlist_head *head;
1350         void *data;
1351         int size, esize;
1352         int rctx;
1353
1354         if (bpf_prog_array_valid(call)) {
1355                 u32 ret;
1356
1357                 preempt_disable();
1358                 ret = trace_call_bpf(call, regs);
1359                 preempt_enable();
1360                 if (!ret)
1361                         return;
1362         }
1363
1364         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1365
1366         size = esize + tu->tp.size + dsize;
1367         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1368         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1369                 return;
1370
1371         preempt_disable();
1372         head = this_cpu_ptr(call->perf_events);
1373         if (hlist_empty(head))
1374                 goto out;
1375
1376         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1377         if (!entry)
1378                 goto out;
1379
1380         if (is_ret_probe(tu)) {
1381                 entry->vaddr[0] = func;
1382                 entry->vaddr[1] = instruction_pointer(regs);
1383                 data = DATAOF_TRACE_ENTRY(entry, true);
1384         } else {
1385                 entry->vaddr[0] = instruction_pointer(regs);
1386                 data = DATAOF_TRACE_ENTRY(entry, false);
1387         }
1388
1389         memcpy(data, ucb->buf, tu->tp.size + dsize);
1390
1391         if (size - esize > tu->tp.size + dsize) {
1392                 int len = tu->tp.size + dsize;
1393
1394                 memset(data + len, 0, size - esize - len);
1395         }
1396
1397         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1398                               head, NULL);
1399  out:
1400         preempt_enable();
1401 }
1402
1403 /* uprobe profile handler */
1404 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1405                             struct uprobe_cpu_buffer *ucb, int dsize)
1406 {
1407         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1408                 return UPROBE_HANDLER_REMOVE;
1409
1410         if (!is_ret_probe(tu))
1411                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1412         return 0;
1413 }
1414
1415 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1416                                 struct pt_regs *regs,
1417                                 struct uprobe_cpu_buffer *ucb, int dsize)
1418 {
1419         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1420 }
1421
1422 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1423                         const char **filename, u64 *probe_offset,
1424                         bool perf_type_tracepoint)
1425 {
1426         const char *pevent = trace_event_name(event->tp_event);
1427         const char *group = event->tp_event->class->system;
1428         struct trace_uprobe *tu;
1429
1430         if (perf_type_tracepoint)
1431                 tu = find_probe_event(pevent, group);
1432         else
1433                 tu = trace_uprobe_primary_from_call(event->tp_event);
1434         if (!tu)
1435                 return -EINVAL;
1436
1437         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1438                                     : BPF_FD_TYPE_UPROBE;
1439         *filename = tu->filename;
1440         *probe_offset = tu->offset;
1441         return 0;
1442 }
1443 #endif  /* CONFIG_PERF_EVENTS */
1444
1445 static int
1446 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1447                       void *data)
1448 {
1449         struct trace_event_file *file = data;
1450
1451         switch (type) {
1452         case TRACE_REG_REGISTER:
1453                 return probe_event_enable(event, file, NULL);
1454
1455         case TRACE_REG_UNREGISTER:
1456                 probe_event_disable(event, file);
1457                 return 0;
1458
1459 #ifdef CONFIG_PERF_EVENTS
1460         case TRACE_REG_PERF_REGISTER:
1461                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1462
1463         case TRACE_REG_PERF_UNREGISTER:
1464                 probe_event_disable(event, NULL);
1465                 return 0;
1466
1467         case TRACE_REG_PERF_OPEN:
1468                 return uprobe_perf_open(event, data);
1469
1470         case TRACE_REG_PERF_CLOSE:
1471                 return uprobe_perf_close(event, data);
1472
1473 #endif
1474         default:
1475                 return 0;
1476         }
1477 }
1478
1479 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1480 {
1481         struct trace_uprobe *tu;
1482         struct uprobe_dispatch_data udd;
1483         struct uprobe_cpu_buffer *ucb;
1484         int dsize, esize;
1485         int ret = 0;
1486
1487
1488         tu = container_of(con, struct trace_uprobe, consumer);
1489         tu->nhit++;
1490
1491         udd.tu = tu;
1492         udd.bp_addr = instruction_pointer(regs);
1493
1494         current->utask->vaddr = (unsigned long) &udd;
1495
1496         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1497                 return 0;
1498
1499         dsize = __get_data_size(&tu->tp, regs);
1500         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1501
1502         ucb = uprobe_buffer_get();
1503         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1504
1505         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1506                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1507
1508 #ifdef CONFIG_PERF_EVENTS
1509         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1510                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1511 #endif
1512         uprobe_buffer_put(ucb);
1513         return ret;
1514 }
1515
1516 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1517                                 unsigned long func, struct pt_regs *regs)
1518 {
1519         struct trace_uprobe *tu;
1520         struct uprobe_dispatch_data udd;
1521         struct uprobe_cpu_buffer *ucb;
1522         int dsize, esize;
1523
1524         tu = container_of(con, struct trace_uprobe, consumer);
1525
1526         udd.tu = tu;
1527         udd.bp_addr = func;
1528
1529         current->utask->vaddr = (unsigned long) &udd;
1530
1531         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1532                 return 0;
1533
1534         dsize = __get_data_size(&tu->tp, regs);
1535         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1536
1537         ucb = uprobe_buffer_get();
1538         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1539
1540         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1541                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1542
1543 #ifdef CONFIG_PERF_EVENTS
1544         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1545                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1546 #endif
1547         uprobe_buffer_put(ucb);
1548         return 0;
1549 }
1550
1551 static struct trace_event_functions uprobe_funcs = {
1552         .trace          = print_uprobe_event
1553 };
1554
1555 static struct trace_event_fields uprobe_fields_array[] = {
1556         { .type = TRACE_FUNCTION_TYPE,
1557           .define_fields = uprobe_event_define_fields },
1558         {}
1559 };
1560
1561 static inline void init_trace_event_call(struct trace_uprobe *tu)
1562 {
1563         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1564         call->event.funcs = &uprobe_funcs;
1565         call->class->fields_array = uprobe_fields_array;
1566
1567         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1568         call->class->reg = trace_uprobe_register;
1569 }
1570
1571 static int register_uprobe_event(struct trace_uprobe *tu)
1572 {
1573         init_trace_event_call(tu);
1574
1575         return trace_probe_register_event_call(&tu->tp);
1576 }
1577
1578 static int unregister_uprobe_event(struct trace_uprobe *tu)
1579 {
1580         return trace_probe_unregister_event_call(&tu->tp);
1581 }
1582
1583 #ifdef CONFIG_PERF_EVENTS
1584 struct trace_event_call *
1585 create_local_trace_uprobe(char *name, unsigned long offs,
1586                           unsigned long ref_ctr_offset, bool is_return)
1587 {
1588         struct trace_uprobe *tu;
1589         struct path path;
1590         int ret;
1591
1592         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1593         if (ret)
1594                 return ERR_PTR(ret);
1595
1596         if (!d_is_reg(path.dentry)) {
1597                 path_put(&path);
1598                 return ERR_PTR(-EINVAL);
1599         }
1600
1601         /*
1602          * local trace_kprobes are not added to dyn_event, so they are never
1603          * searched in find_trace_kprobe(). Therefore, there is no concern of
1604          * duplicated name "DUMMY_EVENT" here.
1605          */
1606         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1607                                 is_return);
1608
1609         if (IS_ERR(tu)) {
1610                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1611                         (int)PTR_ERR(tu));
1612                 path_put(&path);
1613                 return ERR_CAST(tu);
1614         }
1615
1616         tu->offset = offs;
1617         tu->path = path;
1618         tu->ref_ctr_offset = ref_ctr_offset;
1619         tu->filename = kstrdup(name, GFP_KERNEL);
1620         init_trace_event_call(tu);
1621
1622         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1623                 ret = -ENOMEM;
1624                 goto error;
1625         }
1626
1627         return trace_probe_event_call(&tu->tp);
1628 error:
1629         free_trace_uprobe(tu);
1630         return ERR_PTR(ret);
1631 }
1632
1633 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1634 {
1635         struct trace_uprobe *tu;
1636
1637         tu = trace_uprobe_primary_from_call(event_call);
1638
1639         free_trace_uprobe(tu);
1640 }
1641 #endif /* CONFIG_PERF_EVENTS */
1642
1643 /* Make a trace interface for controlling probe points */
1644 static __init int init_uprobe_trace(void)
1645 {
1646         int ret;
1647
1648         ret = dyn_event_register(&trace_uprobe_ops);
1649         if (ret)
1650                 return ret;
1651
1652         ret = tracing_init_dentry();
1653         if (ret)
1654                 return 0;
1655
1656         trace_create_file("uprobe_events", 0644, NULL,
1657                                     NULL, &uprobe_events_ops);
1658         /* Profile interface */
1659         trace_create_file("uprobe_profile", 0444, NULL,
1660                                     NULL, &uprobe_profile_ops);
1661         return 0;
1662 }
1663
1664 fs_initcall(init_uprobe_trace);