kernel/trace/trace_uprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * uprobes-based tracing events
   4  *
   5  * Copyright (C) IBM Corporation, 2010-2012
   6  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
   7  */
   8 #define pr_fmt(fmt)     "trace_uprobe: " fmt
   9
  10 #include <linux/security.h>
  11 #include <linux/ctype.h>
  12 #include <linux/module.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/uprobes.h>
  15 #include <linux/namei.h>
  16 #include <linux/string.h>
  17 #include <linux/rculist.h>
  18
  19 #include "trace_dynevent.h"
  20 #include "trace_probe.h"
  21 #include "trace_probe_tmpl.h"
  22
  23 #define UPROBE_EVENT_SYSTEM     "uprobes"
  24
  25 struct uprobe_trace_entry_head {
  26         struct trace_entry      ent;
  27         unsigned long           vaddr[];
  28 };
  29
  30 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  31         (sizeof(struct uprobe_trace_entry_head) +       \
  32          sizeof(unsigned long) * (is_return ? 2 : 1))
  33
  34 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  35         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  36
  37 static int trace_uprobe_create(int argc, const char **argv);
  38 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  39 static int trace_uprobe_release(struct dyn_event *ev);
  40 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  41 static bool trace_uprobe_match(const char *system, const char *event,
  42                         int argc, const char **argv, struct dyn_event *ev);
  43
  44 static struct dyn_event_operations trace_uprobe_ops = {
  45         .create = trace_uprobe_create,
  46         .show = trace_uprobe_show,
  47         .is_busy = trace_uprobe_is_busy,
  48         .free = trace_uprobe_release,
  49         .match = trace_uprobe_match,
  50 };
  51
  52 /*
  53  * uprobe event core functions
  54  */
  55 struct trace_uprobe {
  56         struct dyn_event                devent;
  57         struct uprobe_consumer          consumer;
  58         struct path                     path;
  59         struct inode                    *inode;
  60         char                            *filename;
  61         unsigned long                   offset;
  62         unsigned long                   ref_ctr_offset;
  63         unsigned long                   nhit;
  64         struct trace_probe              tp;
  65 };
  66
  67 static bool is_trace_uprobe(struct dyn_event *ev)
  68 {
  69         return ev->ops == &trace_uprobe_ops;
  70 }
  71
  72 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  73 {
  74         return container_of(ev, struct trace_uprobe, devent);
  75 }
  76
  77 /**
  78  * for_each_trace_uprobe - iterate over the trace_uprobe list
  79  * @pos:        the struct trace_uprobe * for each entry
  80  * @dpos:       the struct dyn_event * to use as a loop cursor
  81  */
  82 #define for_each_trace_uprobe(pos, dpos)        \
  83         for_each_dyn_event(dpos)                \
  84                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  85
  86 #define SIZEOF_TRACE_UPROBE(n)                          \
  87         (offsetof(struct trace_uprobe, tp.args) +       \
  88         (sizeof(struct probe_arg) * (n)))
  89
  90 static int register_uprobe_event(struct trace_uprobe *tu);
  91 static int unregister_uprobe_event(struct trace_uprobe *tu);
  92
  93 struct uprobe_dispatch_data {
  94         struct trace_uprobe     *tu;
  95         unsigned long           bp_addr;
  96 };
  97
  98 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  99 static int uretprobe_dispatcher(struct uprobe_consumer *con,
 100                                 unsigned long func, struct pt_regs *regs);
 101
 102 #ifdef CONFIG_STACK_GROWSUP
 103 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 104 {
 105         return addr - (n * sizeof(long));
 106 }
 107 #else
 108 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 109 {
 110         return addr + (n * sizeof(long));
 111 }
 112 #endif
 113
 114 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 115 {
 116         unsigned long ret;
 117         unsigned long addr = user_stack_pointer(regs);
 118
 119         addr = adjust_stack_addr(addr, n);
 120
 121         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 122                 return 0;
 123
 124         return ret;
 125 }
 126
 127 /*
 128  * Uprobes-specific fetch functions
 129  */
 130 static nokprobe_inline int
 131 probe_mem_read(void *dest, void *src, size_t size)
 132 {
 133         void __user *vaddr = (void __force __user *)src;
 134
 135         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 136 }
 137
 138 static nokprobe_inline int
 139 probe_mem_read_user(void *dest, void *src, size_t size)
 140 {
 141         return probe_mem_read(dest, src, size);
 142 }
 143
 144 /*
 145  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 146  * length and relative data location.
 147  */
 148 static nokprobe_inline int
 149 fetch_store_string(unsigned long addr, void *dest, void *base)
 150 {
 151         long ret;
 152         u32 loc = *(u32 *)dest;
 153         int maxlen  = get_loc_len(loc);
 154         u8 *dst = get_loc_data(dest, base);
 155         void __user *src = (void __force __user *) addr;
 156
 157         if (unlikely(!maxlen))
 158                 return -ENOMEM;
 159
 160         if (addr == FETCH_TOKEN_COMM)
 161                 ret = strlcpy(dst, current->comm, maxlen);
 162         else
 163                 ret = strncpy_from_user(dst, src, maxlen);
 164         if (ret >= 0) {
 165                 if (ret == maxlen)
 166                         dst[ret - 1] = '\0';
 167                 else
 168                         /*
 169                          * Include the terminating null byte. In this case it
 170                          * was copied by strncpy_from_user but not accounted
 171                          * for in ret.
 172                          */
 173                         ret++;
 174                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 175         }
 176
 177         return ret;
 178 }
 179
 180 static nokprobe_inline int
 181 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 182 {
 183         return fetch_store_string(addr, dest, base);
 184 }
 185
 186 /* Return the length of string -- including null terminal byte */
 187 static nokprobe_inline int
 188 fetch_store_strlen(unsigned long addr)
 189 {
 190         int len;
 191         void __user *vaddr = (void __force __user *) addr;
 192
 193         if (addr == FETCH_TOKEN_COMM)
 194                 len = strlen(current->comm) + 1;
 195         else
 196                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 197
 198         return (len > MAX_STRING_SIZE) ? 0 : len;
 199 }
 200
 201 static nokprobe_inline int
 202 fetch_store_strlen_user(unsigned long addr)
 203 {
 204         return fetch_store_strlen(addr);
 205 }
 206
 207 static unsigned long translate_user_vaddr(unsigned long file_offset)
 208 {
 209         unsigned long base_addr;
 210         struct uprobe_dispatch_data *udd;
 211
 212         udd = (void *) current->utask->vaddr;
 213
 214         base_addr = udd->bp_addr - udd->tu->offset;
 215         return base_addr + file_offset;
 216 }
 217
 218 /* Note that we don't verify it, since the code does not come from user space */
 219 static int
 220 process_fetch_insn(struct fetch_insn *code, 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 [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         /* Parse uprobe offset. */
 621         ret = kstrtoul(arg, 0, &offset);
 622         if (ret) {
 623                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 624                 goto fail_address_parse;
 625         }
 626
 627         /* setup a probe */
 628         trace_probe_log_set_index(0);
 629         if (event) {
 630                 ret = traceprobe_parse_event_name(&event, &group, buf,
 631                                                   event - argv[0]);
 632                 if (ret)
 633                         goto fail_address_parse;
 634         } else {
 635                 char *tail;
 636                 char *ptr;
 637
 638                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 639                 if (!tail) {
 640                         ret = -ENOMEM;
 641                         goto fail_address_parse;
 642                 }
 643
 644                 ptr = strpbrk(tail, ".-_");
 645                 if (ptr)
 646                         *ptr = '\0';
 647
 648                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 649                 event = buf;
 650                 kfree(tail);
 651         }
 652
 653         argc -= 2;
 654         argv += 2;
 655
 656         tu = alloc_trace_uprobe(group, event, argc, is_return);
 657         if (IS_ERR(tu)) {
 658                 ret = PTR_ERR(tu);
 659                 /* This must return -ENOMEM otherwise there is a bug */
 660                 WARN_ON_ONCE(ret != -ENOMEM);
 661                 goto fail_address_parse;
 662         }
 663         tu->offset = offset;
 664         tu->ref_ctr_offset = ref_ctr_offset;
 665         tu->path = path;
 666         tu->filename = filename;
 667
 668         /* parse arguments */
 669         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 670                 tmp = kstrdup(argv[i], GFP_KERNEL);
 671                 if (!tmp) {
 672                         ret = -ENOMEM;
 673                         goto error;
 674                 }
 675
 676                 trace_probe_log_set_index(i + 2);
 677                 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 678                                         is_return ? TPARG_FL_RETURN : 0);
 679                 kfree(tmp);
 680                 if (ret)
 681                         goto error;
 682         }
 683
 684         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 685         if (ret < 0)
 686                 goto error;
 687
 688         ret = register_trace_uprobe(tu);
 689         if (!ret)
 690                 goto out;
 691
 692 error:
 693         free_trace_uprobe(tu);
 694 out:
 695         trace_probe_log_clear();
 696         return ret;
 697
 698 fail_address_parse:
 699         trace_probe_log_clear();
 700         path_put(&path);
 701         kfree(filename);
 702
 703         return ret;
 704 }
 705
 706 static int create_or_delete_trace_uprobe(int argc, char **argv)
 707 {
 708         int ret;
 709
 710         if (argv[0][0] == '-')
 711                 return dyn_event_release(argc, argv, &trace_uprobe_ops);
 712
 713         ret = trace_uprobe_create(argc, (const char **)argv);
 714         return ret == -ECANCELED ? -EINVAL : ret;
 715 }
 716
 717 static int trace_uprobe_release(struct dyn_event *ev)
 718 {
 719         struct trace_uprobe *tu = to_trace_uprobe(ev);
 720
 721         return unregister_trace_uprobe(tu);
 722 }
 723
 724 /* Probes listing interfaces */
 725 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 726 {
 727         struct trace_uprobe *tu = to_trace_uprobe(ev);
 728         char c = is_ret_probe(tu) ? 'r' : 'p';
 729         int i;
 730
 731         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 732                         trace_probe_name(&tu->tp), tu->filename,
 733                         (int)(sizeof(void *) * 2), tu->offset);
 734
 735         if (tu->ref_ctr_offset)
 736                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 737
 738         for (i = 0; i < tu->tp.nr_args; i++)
 739                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 740
 741         seq_putc(m, '\n');
 742         return 0;
 743 }
 744
 745 static int probes_seq_show(struct seq_file *m, void *v)
 746 {
 747         struct dyn_event *ev = v;
 748
 749         if (!is_trace_uprobe(ev))
 750                 return 0;
 751
 752         return trace_uprobe_show(m, ev);
 753 }
 754
 755 static const struct seq_operations probes_seq_op = {
 756         .start  = dyn_event_seq_start,
 757         .next   = dyn_event_seq_next,
 758         .stop   = dyn_event_seq_stop,
 759         .show   = probes_seq_show
 760 };
 761
 762 static int probes_open(struct inode *inode, struct file *file)
 763 {
 764         int ret;
 765
 766         ret = security_locked_down(LOCKDOWN_TRACEFS);
 767         if (ret)
 768                 return ret;
 769
 770         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 771                 ret = dyn_events_release_all(&trace_uprobe_ops);
 772                 if (ret)
 773                         return ret;
 774         }
 775
 776         return seq_open(file, &probes_seq_op);
 777 }
 778
 779 static ssize_t probes_write(struct file *file, const char __user *buffer,
 780                             size_t count, loff_t *ppos)
 781 {
 782         return trace_parse_run_command(file, buffer, count, ppos,
 783                                         create_or_delete_trace_uprobe);
 784 }
 785
 786 static const struct file_operations uprobe_events_ops = {
 787         .owner          = THIS_MODULE,
 788         .open           = probes_open,
 789         .read           = seq_read,
 790         .llseek         = seq_lseek,
 791         .release        = seq_release,
 792         .write          = probes_write,
 793 };
 794
 795 /* Probes profiling interfaces */
 796 static int probes_profile_seq_show(struct seq_file *m, void *v)
 797 {
 798         struct dyn_event *ev = v;
 799         struct trace_uprobe *tu;
 800
 801         if (!is_trace_uprobe(ev))
 802                 return 0;
 803
 804         tu = to_trace_uprobe(ev);
 805         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 806                         trace_probe_name(&tu->tp), tu->nhit);
 807         return 0;
 808 }
 809
 810 static const struct seq_operations profile_seq_op = {
 811         .start  = dyn_event_seq_start,
 812         .next   = dyn_event_seq_next,
 813         .stop   = dyn_event_seq_stop,
 814         .show   = probes_profile_seq_show
 815 };
 816
 817 static int profile_open(struct inode *inode, struct file *file)
 818 {
 819         int ret;
 820
 821         ret = security_locked_down(LOCKDOWN_TRACEFS);
 822         if (ret)
 823                 return ret;
 824
 825         return seq_open(file, &profile_seq_op);
 826 }
 827
 828 static const struct file_operations uprobe_profile_ops = {
 829         .owner          = THIS_MODULE,
 830         .open           = profile_open,
 831         .read           = seq_read,
 832         .llseek         = seq_lseek,
 833         .release        = seq_release,
 834 };
 835
 836 struct uprobe_cpu_buffer {
 837         struct mutex mutex;
 838         void *buf;
 839 };
 840 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 841 static int uprobe_buffer_refcnt;
 842
 843 static int uprobe_buffer_init(void)
 844 {
 845         int cpu, err_cpu;
 846
 847         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 848         if (uprobe_cpu_buffer == NULL)
 849                 return -ENOMEM;
 850
 851         for_each_possible_cpu(cpu) {
 852                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 853                                                   GFP_KERNEL, 0);
 854                 if (p == NULL) {
 855                         err_cpu = cpu;
 856                         goto err;
 857                 }
 858                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 859                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 860         }
 861
 862         return 0;
 863
 864 err:
 865         for_each_possible_cpu(cpu) {
 866                 if (cpu == err_cpu)
 867                         break;
 868                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 869         }
 870
 871         free_percpu(uprobe_cpu_buffer);
 872         return -ENOMEM;
 873 }
 874
 875 static int uprobe_buffer_enable(void)
 876 {
 877         int ret = 0;
 878
 879         BUG_ON(!mutex_is_locked(&event_mutex));
 880
 881         if (uprobe_buffer_refcnt++ == 0) {
 882                 ret = uprobe_buffer_init();
 883                 if (ret < 0)
 884                         uprobe_buffer_refcnt--;
 885         }
 886
 887         return ret;
 888 }
 889
 890 static void uprobe_buffer_disable(void)
 891 {
 892         int cpu;
 893
 894         BUG_ON(!mutex_is_locked(&event_mutex));
 895
 896         if (--uprobe_buffer_refcnt == 0) {
 897                 for_each_possible_cpu(cpu)
 898                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 899                                                              cpu)->buf);
 900
 901                 free_percpu(uprobe_cpu_buffer);
 902                 uprobe_cpu_buffer = NULL;
 903         }
 904 }
 905
 906 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 907 {
 908         struct uprobe_cpu_buffer *ucb;
 909         int cpu;
 910
 911         cpu = raw_smp_processor_id();
 912         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 913
 914         /*
 915          * Use per-cpu buffers for fastest access, but we might migrate
 916          * so the mutex makes sure we have sole access to it.
 917          */
 918         mutex_lock(&ucb->mutex);
 919
 920         return ucb;
 921 }
 922
 923 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 924 {
 925         mutex_unlock(&ucb->mutex);
 926 }
 927
 928 static void __uprobe_trace_func(struct trace_uprobe *tu,
 929                                 unsigned long func, struct pt_regs *regs,
 930                                 struct uprobe_cpu_buffer *ucb, int dsize,
 931                                 struct trace_event_file *trace_file)
 932 {
 933         struct uprobe_trace_entry_head *entry;
 934         struct ring_buffer_event *event;
 935         struct ring_buffer *buffer;
 936         void *data;
 937         int size, esize;
 938         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 939
 940         WARN_ON(call != trace_file->event_call);
 941
 942         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 943                 return;
 944
 945         if (trace_trigger_soft_disabled(trace_file))
 946                 return;
 947
 948         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 949         size = esize + tu->tp.size + dsize;
 950         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 951                                                 call->event.type, size, 0, 0);
 952         if (!event)
 953                 return;
 954
 955         entry = ring_buffer_event_data(event);
 956         if (is_ret_probe(tu)) {
 957                 entry->vaddr[0] = func;
 958                 entry->vaddr[1] = instruction_pointer(regs);
 959                 data = DATAOF_TRACE_ENTRY(entry, true);
 960         } else {
 961                 entry->vaddr[0] = instruction_pointer(regs);
 962                 data = DATAOF_TRACE_ENTRY(entry, false);
 963         }
 964
 965         memcpy(data, ucb->buf, tu->tp.size + dsize);
 966
 967         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 968 }
 969
 970 /* uprobe handler */
 971 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 972                              struct uprobe_cpu_buffer *ucb, int dsize)
 973 {
 974         struct event_file_link *link;
 975
 976         if (is_ret_probe(tu))
 977                 return 0;
 978
 979         rcu_read_lock();
 980         trace_probe_for_each_link_rcu(link, &tu->tp)
 981                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 982         rcu_read_unlock();
 983
 984         return 0;
 985 }
 986
 987 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 988                                  struct pt_regs *regs,
 989                                  struct uprobe_cpu_buffer *ucb, int dsize)
 990 {
 991         struct event_file_link *link;
 992
 993         rcu_read_lock();
 994         trace_probe_for_each_link_rcu(link, &tu->tp)
 995                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 996         rcu_read_unlock();
 997 }
 998
 999 /* Event entry printers */
1000 static enum print_line_t
1001 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1002 {
1003         struct uprobe_trace_entry_head *entry;
1004         struct trace_seq *s = &iter->seq;
1005         struct trace_uprobe *tu;
1006         u8 *data;
1007
1008         entry = (struct uprobe_trace_entry_head *)iter->ent;
1009         tu = trace_uprobe_primary_from_call(
1010                 container_of(event, struct trace_event_call, event));
1011         if (unlikely(!tu))
1012                 goto out;
1013
1014         if (is_ret_probe(tu)) {
1015                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1016                                  trace_probe_name(&tu->tp),
1017                                  entry->vaddr[1], entry->vaddr[0]);
1018                 data = DATAOF_TRACE_ENTRY(entry, true);
1019         } else {
1020                 trace_seq_printf(s, "%s: (0x%lx)",
1021                                  trace_probe_name(&tu->tp),
1022                                  entry->vaddr[0]);
1023                 data = DATAOF_TRACE_ENTRY(entry, false);
1024         }
1025
1026         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1027                 goto out;
1028
1029         trace_seq_putc(s, '\n');
1030
1031  out:
1032         return trace_handle_return(s);
1033 }
1034
1035 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1036                                 enum uprobe_filter_ctx ctx,
1037                                 struct mm_struct *mm);
1038
1039 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1040 {
1041         int ret;
1042
1043         tu->consumer.filter = filter;
1044         tu->inode = d_real_inode(tu->path.dentry);
1045
1046         if (tu->ref_ctr_offset)
1047                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1048                                 tu->ref_ctr_offset, &tu->consumer);
1049         else
1050                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1051
1052         if (ret)
1053                 tu->inode = NULL;
1054
1055         return ret;
1056 }
1057
1058 static void __probe_event_disable(struct trace_probe *tp)
1059 {
1060         struct trace_probe *pos;
1061         struct trace_uprobe *tu;
1062
1063         tu = container_of(tp, struct trace_uprobe, tp);
1064         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1065
1066         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1067                 tu = container_of(pos, struct trace_uprobe, tp);
1068                 if (!tu->inode)
1069                         continue;
1070
1071                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1072                 tu->inode = NULL;
1073         }
1074 }
1075
1076 static int probe_event_enable(struct trace_event_call *call,
1077                         struct trace_event_file *file, filter_func_t filter)
1078 {
1079         struct trace_probe *pos, *tp;
1080         struct trace_uprobe *tu;
1081         bool enabled;
1082         int ret;
1083
1084         tp = trace_probe_primary_from_call(call);
1085         if (WARN_ON_ONCE(!tp))
1086                 return -ENODEV;
1087         enabled = trace_probe_is_enabled(tp);
1088
1089         /* This may also change "enabled" state */
1090         if (file) {
1091                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1092                         return -EINTR;
1093
1094                 ret = trace_probe_add_file(tp, file);
1095                 if (ret < 0)
1096                         return ret;
1097         } else {
1098                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1099                         return -EINTR;
1100
1101                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1102         }
1103
1104         tu = container_of(tp, struct trace_uprobe, tp);
1105         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1106
1107         if (enabled)
1108                 return 0;
1109
1110         ret = uprobe_buffer_enable();
1111         if (ret)
1112                 goto err_flags;
1113
1114         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1115                 tu = container_of(pos, struct trace_uprobe, tp);
1116                 ret = trace_uprobe_enable(tu, filter);
1117                 if (ret) {
1118                         __probe_event_disable(tp);
1119                         goto err_buffer;
1120                 }
1121         }
1122
1123         return 0;
1124
1125  err_buffer:
1126         uprobe_buffer_disable();
1127
1128  err_flags:
1129         if (file)
1130                 trace_probe_remove_file(tp, file);
1131         else
1132                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1133
1134         return ret;
1135 }
1136
1137 static void probe_event_disable(struct trace_event_call *call,
1138                                 struct trace_event_file *file)
1139 {
1140         struct trace_probe *tp;
1141
1142         tp = trace_probe_primary_from_call(call);
1143         if (WARN_ON_ONCE(!tp))
1144                 return;
1145
1146         if (!trace_probe_is_enabled(tp))
1147                 return;
1148
1149         if (file) {
1150                 if (trace_probe_remove_file(tp, file) < 0)
1151                         return;
1152
1153                 if (trace_probe_is_enabled(tp))
1154                         return;
1155         } else
1156                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1157
1158         __probe_event_disable(tp);
1159         uprobe_buffer_disable();
1160 }
1161
1162 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1163 {
1164         int ret, size;
1165         struct uprobe_trace_entry_head field;
1166         struct trace_uprobe *tu;
1167
1168         tu = trace_uprobe_primary_from_call(event_call);
1169         if (unlikely(!tu))
1170                 return -ENODEV;
1171
1172         if (is_ret_probe(tu)) {
1173                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1174                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1175                 size = SIZEOF_TRACE_ENTRY(true);
1176         } else {
1177                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1178                 size = SIZEOF_TRACE_ENTRY(false);
1179         }
1180
1181         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1182 }
1183
1184 #ifdef CONFIG_PERF_EVENTS
1185 static bool
1186 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1187 {
1188         struct perf_event *event;
1189
1190         if (filter->nr_systemwide)
1191                 return true;
1192
1193         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1194                 if (event->hw.target->mm == mm)
1195                         return true;
1196         }
1197
1198         return false;
1199 }
1200
1201 static inline bool
1202 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1203                           struct perf_event *event)
1204 {
1205         return __uprobe_perf_filter(filter, event->hw.target->mm);
1206 }
1207
1208 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1209                                        struct perf_event *event)
1210 {
1211         bool done;
1212
1213         write_lock(&filter->rwlock);
1214         if (event->hw.target) {
1215                 list_del(&event->hw.tp_list);
1216                 done = filter->nr_systemwide ||
1217                         (event->hw.target->flags & PF_EXITING) ||
1218                         trace_uprobe_filter_event(filter, event);
1219         } else {
1220                 filter->nr_systemwide--;
1221                 done = filter->nr_systemwide;
1222         }
1223         write_unlock(&filter->rwlock);
1224
1225         return done;
1226 }
1227
1228 /* This returns true if the filter always covers target mm */
1229 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1230                                     struct perf_event *event)
1231 {
1232         bool done;
1233
1234         write_lock(&filter->rwlock);
1235         if (event->hw.target) {
1236                 /*
1237                  * event->parent != NULL means copy_process(), we can avoid
1238                  * uprobe_apply(). current->mm must be probed and we can rely
1239                  * on dup_mmap() which preserves the already installed bp's.
1240                  *
1241                  * attr.enable_on_exec means that exec/mmap will install the
1242                  * breakpoints we need.
1243                  */
1244                 done = filter->nr_systemwide ||
1245                         event->parent || event->attr.enable_on_exec ||
1246                         trace_uprobe_filter_event(filter, event);
1247                 list_add(&event->hw.tp_list, &filter->perf_events);
1248         } else {
1249                 done = filter->nr_systemwide;
1250                 filter->nr_systemwide++;
1251         }
1252         write_unlock(&filter->rwlock);
1253
1254         return done;
1255 }
1256
1257 static int uprobe_perf_close(struct trace_event_call *call,
1258                              struct perf_event *event)
1259 {
1260         struct trace_probe *pos, *tp;
1261         struct trace_uprobe *tu;
1262         int ret = 0;
1263
1264         tp = trace_probe_primary_from_call(call);
1265         if (WARN_ON_ONCE(!tp))
1266                 return -ENODEV;
1267
1268         tu = container_of(tp, struct trace_uprobe, tp);
1269         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1270                 return 0;
1271
1272         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1273                 tu = container_of(pos, struct trace_uprobe, tp);
1274                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1275                 if (ret)
1276                         break;
1277         }
1278
1279         return ret;
1280 }
1281
1282 static int uprobe_perf_open(struct trace_event_call *call,
1283                             struct perf_event *event)
1284 {
1285         struct trace_probe *pos, *tp;
1286         struct trace_uprobe *tu;
1287         int err = 0;
1288
1289         tp = trace_probe_primary_from_call(call);
1290         if (WARN_ON_ONCE(!tp))
1291                 return -ENODEV;
1292
1293         tu = container_of(tp, struct trace_uprobe, tp);
1294         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1295                 return 0;
1296
1297         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1298                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1299                 if (err) {
1300                         uprobe_perf_close(call, event);
1301                         break;
1302                 }
1303         }
1304
1305         return err;
1306 }
1307
1308 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1309                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1310 {
1311         struct trace_uprobe_filter *filter;
1312         struct trace_uprobe *tu;
1313         int ret;
1314
1315         tu = container_of(uc, struct trace_uprobe, consumer);
1316         filter = tu->tp.event->filter;
1317
1318         read_lock(&filter->rwlock);
1319         ret = __uprobe_perf_filter(filter, mm);
1320         read_unlock(&filter->rwlock);
1321
1322         return ret;
1323 }
1324
1325 static void __uprobe_perf_func(struct trace_uprobe *tu,
1326                                unsigned long func, struct pt_regs *regs,
1327                                struct uprobe_cpu_buffer *ucb, int dsize)
1328 {
1329         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1330         struct uprobe_trace_entry_head *entry;
1331         struct hlist_head *head;
1332         void *data;
1333         int size, esize;
1334         int rctx;
1335
1336         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1337                 return;
1338
1339         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1340
1341         size = esize + tu->tp.size + dsize;
1342         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1343         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1344                 return;
1345
1346         preempt_disable();
1347         head = this_cpu_ptr(call->perf_events);
1348         if (hlist_empty(head))
1349                 goto out;
1350
1351         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1352         if (!entry)
1353                 goto out;
1354
1355         if (is_ret_probe(tu)) {
1356                 entry->vaddr[0] = func;
1357                 entry->vaddr[1] = instruction_pointer(regs);
1358                 data = DATAOF_TRACE_ENTRY(entry, true);
1359         } else {
1360                 entry->vaddr[0] = instruction_pointer(regs);
1361                 data = DATAOF_TRACE_ENTRY(entry, false);
1362         }
1363
1364         memcpy(data, ucb->buf, tu->tp.size + dsize);
1365
1366         if (size - esize > tu->tp.size + dsize) {
1367                 int len = tu->tp.size + dsize;
1368
1369                 memset(data + len, 0, size - esize - len);
1370         }
1371
1372         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1373                               head, NULL);
1374  out:
1375         preempt_enable();
1376 }
1377
1378 /* uprobe profile handler */
1379 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1380                             struct uprobe_cpu_buffer *ucb, int dsize)
1381 {
1382         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1383                 return UPROBE_HANDLER_REMOVE;
1384
1385         if (!is_ret_probe(tu))
1386                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1387         return 0;
1388 }
1389
1390 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1391                                 struct pt_regs *regs,
1392                                 struct uprobe_cpu_buffer *ucb, int dsize)
1393 {
1394         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1395 }
1396
1397 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1398                         const char **filename, u64 *probe_offset,
1399                         bool perf_type_tracepoint)
1400 {
1401         const char *pevent = trace_event_name(event->tp_event);
1402         const char *group = event->tp_event->class->system;
1403         struct trace_uprobe *tu;
1404
1405         if (perf_type_tracepoint)
1406                 tu = find_probe_event(pevent, group);
1407         else
1408                 tu = event->tp_event->data;
1409         if (!tu)
1410                 return -EINVAL;
1411
1412         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1413                                     : BPF_FD_TYPE_UPROBE;
1414         *filename = tu->filename;
1415         *probe_offset = tu->offset;
1416         return 0;
1417 }
1418 #endif  /* CONFIG_PERF_EVENTS */
1419
1420 static int
1421 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1422                       void *data)
1423 {
1424         struct trace_event_file *file = data;
1425
1426         switch (type) {
1427         case TRACE_REG_REGISTER:
1428                 return probe_event_enable(event, file, NULL);
1429
1430         case TRACE_REG_UNREGISTER:
1431                 probe_event_disable(event, file);
1432                 return 0;
1433
1434 #ifdef CONFIG_PERF_EVENTS
1435         case TRACE_REG_PERF_REGISTER:
1436                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1437
1438         case TRACE_REG_PERF_UNREGISTER:
1439                 probe_event_disable(event, NULL);
1440                 return 0;
1441
1442         case TRACE_REG_PERF_OPEN:
1443                 return uprobe_perf_open(event, data);
1444
1445         case TRACE_REG_PERF_CLOSE:
1446                 return uprobe_perf_close(event, data);
1447
1448 #endif
1449         default:
1450                 return 0;
1451         }
1452         return 0;
1453 }
1454
1455 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1456 {
1457         struct trace_uprobe *tu;
1458         struct uprobe_dispatch_data udd;
1459         struct uprobe_cpu_buffer *ucb;
1460         int dsize, esize;
1461         int ret = 0;
1462
1463
1464         tu = container_of(con, struct trace_uprobe, consumer);
1465         tu->nhit++;
1466
1467         udd.tu = tu;
1468         udd.bp_addr = instruction_pointer(regs);
1469
1470         current->utask->vaddr = (unsigned long) &udd;
1471
1472         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1473                 return 0;
1474
1475         dsize = __get_data_size(&tu->tp, regs);
1476         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1477
1478         ucb = uprobe_buffer_get();
1479         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1480
1481         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1482                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1483
1484 #ifdef CONFIG_PERF_EVENTS
1485         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1486                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1487 #endif
1488         uprobe_buffer_put(ucb);
1489         return ret;
1490 }
1491
1492 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1493                                 unsigned long func, struct pt_regs *regs)
1494 {
1495         struct trace_uprobe *tu;
1496         struct uprobe_dispatch_data udd;
1497         struct uprobe_cpu_buffer *ucb;
1498         int dsize, esize;
1499
1500         tu = container_of(con, struct trace_uprobe, consumer);
1501
1502         udd.tu = tu;
1503         udd.bp_addr = func;
1504
1505         current->utask->vaddr = (unsigned long) &udd;
1506
1507         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1508                 return 0;
1509
1510         dsize = __get_data_size(&tu->tp, regs);
1511         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1512
1513         ucb = uprobe_buffer_get();
1514         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1515
1516         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1517                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1518
1519 #ifdef CONFIG_PERF_EVENTS
1520         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1521                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1522 #endif
1523         uprobe_buffer_put(ucb);
1524         return 0;
1525 }
1526
1527 static struct trace_event_functions uprobe_funcs = {
1528         .trace          = print_uprobe_event
1529 };
1530
1531 static inline void init_trace_event_call(struct trace_uprobe *tu)
1532 {
1533         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1534
1535         call->event.funcs = &uprobe_funcs;
1536         call->class->define_fields = uprobe_event_define_fields;
1537
1538         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1539         call->class->reg = trace_uprobe_register;
1540 }
1541
1542 static int register_uprobe_event(struct trace_uprobe *tu)
1543 {
1544         init_trace_event_call(tu);
1545
1546         return trace_probe_register_event_call(&tu->tp);
1547 }
1548
1549 static int unregister_uprobe_event(struct trace_uprobe *tu)
1550 {
1551         return trace_probe_unregister_event_call(&tu->tp);
1552 }
1553
1554 #ifdef CONFIG_PERF_EVENTS
1555 struct trace_event_call *
1556 create_local_trace_uprobe(char *name, unsigned long offs,
1557                           unsigned long ref_ctr_offset, bool is_return)
1558 {
1559         struct trace_uprobe *tu;
1560         struct path path;
1561         int ret;
1562
1563         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1564         if (ret)
1565                 return ERR_PTR(ret);
1566
1567         if (!d_is_reg(path.dentry)) {
1568                 path_put(&path);
1569                 return ERR_PTR(-EINVAL);
1570         }
1571
1572         /*
1573          * local trace_kprobes are not added to dyn_event, so they are never
1574          * searched in find_trace_kprobe(). Therefore, there is no concern of
1575          * duplicated name "DUMMY_EVENT" here.
1576          */
1577         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1578                                 is_return);
1579
1580         if (IS_ERR(tu)) {
1581                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1582                         (int)PTR_ERR(tu));
1583                 path_put(&path);
1584                 return ERR_CAST(tu);
1585         }
1586
1587         tu->offset = offs;
1588         tu->path = path;
1589         tu->ref_ctr_offset = ref_ctr_offset;
1590         tu->filename = kstrdup(name, GFP_KERNEL);
1591         init_trace_event_call(tu);
1592
1593         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1594                 ret = -ENOMEM;
1595                 goto error;
1596         }
1597
1598         return trace_probe_event_call(&tu->tp);
1599 error:
1600         free_trace_uprobe(tu);
1601         return ERR_PTR(ret);
1602 }
1603
1604 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1605 {
1606         struct trace_uprobe *tu;
1607
1608         tu = trace_uprobe_primary_from_call(event_call);
1609
1610         free_trace_uprobe(tu);
1611 }
1612 #endif /* CONFIG_PERF_EVENTS */
1613
1614 /* Make a trace interface for controling probe points */
1615 static __init int init_uprobe_trace(void)
1616 {
1617         struct dentry *d_tracer;
1618         int ret;
1619
1620         ret = dyn_event_register(&trace_uprobe_ops);
1621         if (ret)
1622                 return ret;
1623
1624         d_tracer = tracing_init_dentry();
1625         if (IS_ERR(d_tracer))
1626                 return 0;
1627
1628         trace_create_file("uprobe_events", 0644, d_tracer,
1629                                     NULL, &uprobe_events_ops);
1630         /* Profile interface */
1631         trace_create_file("uprobe_profile", 0444, d_tracer,
1632                                     NULL, &uprobe_profile_ops);
1633         return 0;
1634 }
1635
1636 fs_initcall(init_uprobe_trace);