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         /* If there's a reference to the dynamic event */
 397         if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 398                 return -EBUSY;
 399
 400         ret = unregister_uprobe_event(tu);
 401         if (ret)
 402                 return ret;
 403
 404 unreg:
 405         dyn_event_remove(&tu->devent);
 406         trace_probe_unlink(&tu->tp);
 407         free_trace_uprobe(tu);
 408         return 0;
 409 }
 410
 411 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 412                                          struct trace_uprobe *comp)
 413 {
 414         struct trace_probe_event *tpe = orig->tp.event;
 415         struct trace_probe *pos;
 416         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 417         int i;
 418
 419         list_for_each_entry(pos, &tpe->probes, list) {
 420                 orig = container_of(pos, struct trace_uprobe, tp);
 421                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 422                     comp->offset != orig->offset)
 423                         continue;
 424
 425                 /*
 426                  * trace_probe_compare_arg_type() ensured that nr_args and
 427                  * each argument name and type are same. Let's compare comm.
 428                  */
 429                 for (i = 0; i < orig->tp.nr_args; i++) {
 430                         if (strcmp(orig->tp.args[i].comm,
 431                                    comp->tp.args[i].comm))
 432                                 break;
 433                 }
 434
 435                 if (i == orig->tp.nr_args)
 436                         return true;
 437         }
 438
 439         return false;
 440 }
 441
 442 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 443 {
 444         int ret;
 445
 446         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 447         if (ret) {
 448                 /* Note that argument starts index = 2 */
 449                 trace_probe_log_set_index(ret + 1);
 450                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 451                 return -EEXIST;
 452         }
 453         if (trace_uprobe_has_same_uprobe(to, tu)) {
 454                 trace_probe_log_set_index(0);
 455                 trace_probe_log_err(0, SAME_PROBE);
 456                 return -EEXIST;
 457         }
 458
 459         /* Append to existing event */
 460         ret = trace_probe_append(&tu->tp, &to->tp);
 461         if (!ret)
 462                 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 463
 464         return ret;
 465 }
 466
 467 /*
 468  * Uprobe with multiple reference counter is not allowed. i.e.
 469  * If inode and offset matches, reference counter offset *must*
 470  * match as well. Though, there is one exception: If user is
 471  * replacing old trace_uprobe with new one(same group/event),
 472  * then we allow same uprobe with new reference counter as far
 473  * as the new one does not conflict with any other existing
 474  * ones.
 475  */
 476 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 477 {
 478         struct dyn_event *pos;
 479         struct trace_uprobe *tmp;
 480         struct inode *new_inode = d_real_inode(new->path.dentry);
 481
 482         for_each_trace_uprobe(tmp, pos) {
 483                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 484                     new->offset == tmp->offset &&
 485                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 486                         pr_warn("Reference counter offset mismatch.");
 487                         return -EINVAL;
 488                 }
 489         }
 490         return 0;
 491 }
 492
 493 /* Register a trace_uprobe and probe_event */
 494 static int register_trace_uprobe(struct trace_uprobe *tu)
 495 {
 496         struct trace_uprobe *old_tu;
 497         int ret;
 498
 499         mutex_lock(&event_mutex);
 500
 501         ret = validate_ref_ctr_offset(tu);
 502         if (ret)
 503                 goto end;
 504
 505         /* register as an event */
 506         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 507                                   trace_probe_group_name(&tu->tp));
 508         if (old_tu) {
 509                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 510                         trace_probe_log_set_index(0);
 511                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 512                         ret = -EEXIST;
 513                 } else {
 514                         ret = append_trace_uprobe(tu, old_tu);
 515                 }
 516                 goto end;
 517         }
 518
 519         ret = register_uprobe_event(tu);
 520         if (ret) {
 521                 pr_warn("Failed to register probe event(%d)\n", ret);
 522                 goto end;
 523         }
 524
 525         dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 526
 527 end:
 528         mutex_unlock(&event_mutex);
 529
 530         return ret;
 531 }
 532
 533 /*
 534  * Argument syntax:
 535  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 536  */
 537 static int __trace_uprobe_create(int argc, const char **argv)
 538 {
 539         struct trace_uprobe *tu;
 540         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 541         char *arg, *filename, *rctr, *rctr_end, *tmp;
 542         char buf[MAX_EVENT_NAME_LEN];
 543         struct path path;
 544         unsigned long offset, ref_ctr_offset;
 545         bool is_return = false;
 546         int i, ret;
 547
 548         ret = 0;
 549         ref_ctr_offset = 0;
 550
 551         switch (argv[0][0]) {
 552         case 'r':
 553                 is_return = true;
 554                 break;
 555         case 'p':
 556                 break;
 557         default:
 558                 return -ECANCELED;
 559         }
 560
 561         if (argc < 2)
 562                 return -ECANCELED;
 563
 564         if (argv[0][1] == ':')
 565                 event = &argv[0][2];
 566
 567         if (!strchr(argv[1], '/'))
 568                 return -ECANCELED;
 569
 570         filename = kstrdup(argv[1], GFP_KERNEL);
 571         if (!filename)
 572                 return -ENOMEM;
 573
 574         /* Find the last occurrence, in case the path contains ':' too. */
 575         arg = strrchr(filename, ':');
 576         if (!arg || !isdigit(arg[1])) {
 577                 kfree(filename);
 578                 return -ECANCELED;
 579         }
 580
 581         trace_probe_log_init("trace_uprobe", argc, argv);
 582         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 583
 584         *arg++ = '\0';
 585         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 586         if (ret) {
 587                 trace_probe_log_err(0, FILE_NOT_FOUND);
 588                 kfree(filename);
 589                 trace_probe_log_clear();
 590                 return ret;
 591         }
 592         if (!d_is_reg(path.dentry)) {
 593                 trace_probe_log_err(0, NO_REGULAR_FILE);
 594                 ret = -EINVAL;
 595                 goto fail_address_parse;
 596         }
 597
 598         /* Parse reference counter offset if specified. */
 599         rctr = strchr(arg, '(');
 600         if (rctr) {
 601                 rctr_end = strchr(rctr, ')');
 602                 if (!rctr_end) {
 603                         ret = -EINVAL;
 604                         rctr_end = rctr + strlen(rctr);
 605                         trace_probe_log_err(rctr_end - filename,
 606                                             REFCNT_OPEN_BRACE);
 607                         goto fail_address_parse;
 608                 } else if (rctr_end[1] != '\0') {
 609                         ret = -EINVAL;
 610                         trace_probe_log_err(rctr_end + 1 - filename,
 611                                             BAD_REFCNT_SUFFIX);
 612                         goto fail_address_parse;
 613                 }
 614
 615                 *rctr++ = '\0';
 616                 *rctr_end = '\0';
 617                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 618                 if (ret) {
 619                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 620                         goto fail_address_parse;
 621                 }
 622         }
 623
 624         /* Check if there is %return suffix */
 625         tmp = strchr(arg, '%');
 626         if (tmp) {
 627                 if (!strcmp(tmp, "%return")) {
 628                         *tmp = '\0';
 629                         is_return = true;
 630                 } else {
 631                         trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 632                         ret = -EINVAL;
 633                         goto fail_address_parse;
 634                 }
 635         }
 636
 637         /* Parse uprobe offset. */
 638         ret = kstrtoul(arg, 0, &offset);
 639         if (ret) {
 640                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 641                 goto fail_address_parse;
 642         }
 643
 644         /* setup a probe */
 645         trace_probe_log_set_index(0);
 646         if (event) {
 647                 ret = traceprobe_parse_event_name(&event, &group, buf,
 648                                                   event - argv[0]);
 649                 if (ret)
 650                         goto fail_address_parse;
 651         } else {
 652                 char *tail;
 653                 char *ptr;
 654
 655                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 656                 if (!tail) {
 657                         ret = -ENOMEM;
 658                         goto fail_address_parse;
 659                 }
 660
 661                 ptr = strpbrk(tail, ".-_");
 662                 if (ptr)
 663                         *ptr = '\0';
 664
 665                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 666                 event = buf;
 667                 kfree(tail);
 668         }
 669
 670         argc -= 2;
 671         argv += 2;
 672
 673         tu = alloc_trace_uprobe(group, event, argc, is_return);
 674         if (IS_ERR(tu)) {
 675                 ret = PTR_ERR(tu);
 676                 /* This must return -ENOMEM otherwise there is a bug */
 677                 WARN_ON_ONCE(ret != -ENOMEM);
 678                 goto fail_address_parse;
 679         }
 680         tu->offset = offset;
 681         tu->ref_ctr_offset = ref_ctr_offset;
 682         tu->path = path;
 683         tu->filename = filename;
 684
 685         /* parse arguments */
 686         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 687                 trace_probe_log_set_index(i + 2);
 688                 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i],
 689                                         is_return ? TPARG_FL_RETURN : 0);
 690                 if (ret)
 691                         goto error;
 692         }
 693
 694         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 695         if (ret < 0)
 696                 goto error;
 697
 698         ret = register_trace_uprobe(tu);
 699         if (!ret)
 700                 goto out;
 701
 702 error:
 703         free_trace_uprobe(tu);
 704 out:
 705         trace_probe_log_clear();
 706         return ret;
 707
 708 fail_address_parse:
 709         trace_probe_log_clear();
 710         path_put(&path);
 711         kfree(filename);
 712
 713         return ret;
 714 }
 715
 716 int trace_uprobe_create(const char *raw_command)
 717 {
 718         return trace_probe_create(raw_command, __trace_uprobe_create);
 719 }
 720
 721 static int create_or_delete_trace_uprobe(const char *raw_command)
 722 {
 723         int ret;
 724
 725         if (raw_command[0] == '-')
 726                 return dyn_event_release(raw_command, &trace_uprobe_ops);
 727
 728         ret = trace_uprobe_create(raw_command);
 729         return ret == -ECANCELED ? -EINVAL : ret;
 730 }
 731
 732 static int trace_uprobe_release(struct dyn_event *ev)
 733 {
 734         struct trace_uprobe *tu = to_trace_uprobe(ev);
 735
 736         return unregister_trace_uprobe(tu);
 737 }
 738
 739 /* Probes listing interfaces */
 740 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 741 {
 742         struct trace_uprobe *tu = to_trace_uprobe(ev);
 743         char c = is_ret_probe(tu) ? 'r' : 'p';
 744         int i;
 745
 746         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 747                         trace_probe_name(&tu->tp), tu->filename,
 748                         (int)(sizeof(void *) * 2), tu->offset);
 749
 750         if (tu->ref_ctr_offset)
 751                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 752
 753         for (i = 0; i < tu->tp.nr_args; i++)
 754                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 755
 756         seq_putc(m, '\n');
 757         return 0;
 758 }
 759
 760 static int probes_seq_show(struct seq_file *m, void *v)
 761 {
 762         struct dyn_event *ev = v;
 763
 764         if (!is_trace_uprobe(ev))
 765                 return 0;
 766
 767         return trace_uprobe_show(m, ev);
 768 }
 769
 770 static const struct seq_operations probes_seq_op = {
 771         .start  = dyn_event_seq_start,
 772         .next   = dyn_event_seq_next,
 773         .stop   = dyn_event_seq_stop,
 774         .show   = probes_seq_show
 775 };
 776
 777 static int probes_open(struct inode *inode, struct file *file)
 778 {
 779         int ret;
 780
 781         ret = security_locked_down(LOCKDOWN_TRACEFS);
 782         if (ret)
 783                 return ret;
 784
 785         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 786                 ret = dyn_events_release_all(&trace_uprobe_ops);
 787                 if (ret)
 788                         return ret;
 789         }
 790
 791         return seq_open(file, &probes_seq_op);
 792 }
 793
 794 static ssize_t probes_write(struct file *file, const char __user *buffer,
 795                             size_t count, loff_t *ppos)
 796 {
 797         return trace_parse_run_command(file, buffer, count, ppos,
 798                                         create_or_delete_trace_uprobe);
 799 }
 800
 801 static const struct file_operations uprobe_events_ops = {
 802         .owner          = THIS_MODULE,
 803         .open           = probes_open,
 804         .read           = seq_read,
 805         .llseek         = seq_lseek,
 806         .release        = seq_release,
 807         .write          = probes_write,
 808 };
 809
 810 /* Probes profiling interfaces */
 811 static int probes_profile_seq_show(struct seq_file *m, void *v)
 812 {
 813         struct dyn_event *ev = v;
 814         struct trace_uprobe *tu;
 815
 816         if (!is_trace_uprobe(ev))
 817                 return 0;
 818
 819         tu = to_trace_uprobe(ev);
 820         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 821                         trace_probe_name(&tu->tp), tu->nhit);
 822         return 0;
 823 }
 824
 825 static const struct seq_operations profile_seq_op = {
 826         .start  = dyn_event_seq_start,
 827         .next   = dyn_event_seq_next,
 828         .stop   = dyn_event_seq_stop,
 829         .show   = probes_profile_seq_show
 830 };
 831
 832 static int profile_open(struct inode *inode, struct file *file)
 833 {
 834         int ret;
 835
 836         ret = security_locked_down(LOCKDOWN_TRACEFS);
 837         if (ret)
 838                 return ret;
 839
 840         return seq_open(file, &profile_seq_op);
 841 }
 842
 843 static const struct file_operations uprobe_profile_ops = {
 844         .owner          = THIS_MODULE,
 845         .open           = profile_open,
 846         .read           = seq_read,
 847         .llseek         = seq_lseek,
 848         .release        = seq_release,
 849 };
 850
 851 struct uprobe_cpu_buffer {
 852         struct mutex mutex;
 853         void *buf;
 854 };
 855 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 856 static int uprobe_buffer_refcnt;
 857
 858 static int uprobe_buffer_init(void)
 859 {
 860         int cpu, err_cpu;
 861
 862         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 863         if (uprobe_cpu_buffer == NULL)
 864                 return -ENOMEM;
 865
 866         for_each_possible_cpu(cpu) {
 867                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 868                                                   GFP_KERNEL, 0);
 869                 if (p == NULL) {
 870                         err_cpu = cpu;
 871                         goto err;
 872                 }
 873                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 874                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 875         }
 876
 877         return 0;
 878
 879 err:
 880         for_each_possible_cpu(cpu) {
 881                 if (cpu == err_cpu)
 882                         break;
 883                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 884         }
 885
 886         free_percpu(uprobe_cpu_buffer);
 887         return -ENOMEM;
 888 }
 889
 890 static int uprobe_buffer_enable(void)
 891 {
 892         int ret = 0;
 893
 894         BUG_ON(!mutex_is_locked(&event_mutex));
 895
 896         if (uprobe_buffer_refcnt++ == 0) {
 897                 ret = uprobe_buffer_init();
 898                 if (ret < 0)
 899                         uprobe_buffer_refcnt--;
 900         }
 901
 902         return ret;
 903 }
 904
 905 static void uprobe_buffer_disable(void)
 906 {
 907         int cpu;
 908
 909         BUG_ON(!mutex_is_locked(&event_mutex));
 910
 911         if (--uprobe_buffer_refcnt == 0) {
 912                 for_each_possible_cpu(cpu)
 913                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 914                                                              cpu)->buf);
 915
 916                 free_percpu(uprobe_cpu_buffer);
 917                 uprobe_cpu_buffer = NULL;
 918         }
 919 }
 920
 921 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 922 {
 923         struct uprobe_cpu_buffer *ucb;
 924         int cpu;
 925
 926         cpu = raw_smp_processor_id();
 927         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 928
 929         /*
 930          * Use per-cpu buffers for fastest access, but we might migrate
 931          * so the mutex makes sure we have sole access to it.
 932          */
 933         mutex_lock(&ucb->mutex);
 934
 935         return ucb;
 936 }
 937
 938 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 939 {
 940         mutex_unlock(&ucb->mutex);
 941 }
 942
 943 static void __uprobe_trace_func(struct trace_uprobe *tu,
 944                                 unsigned long func, struct pt_regs *regs,
 945                                 struct uprobe_cpu_buffer *ucb, int dsize,
 946                                 struct trace_event_file *trace_file)
 947 {
 948         struct uprobe_trace_entry_head *entry;
 949         struct trace_buffer *buffer;
 950         struct ring_buffer_event *event;
 951         void *data;
 952         int size, esize;
 953         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 954
 955         WARN_ON(call != trace_file->event_call);
 956
 957         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 958                 return;
 959
 960         if (trace_trigger_soft_disabled(trace_file))
 961                 return;
 962
 963         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 964         size = esize + tu->tp.size + dsize;
 965         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 966                                                 call->event.type, size, 0);
 967         if (!event)
 968                 return;
 969
 970         entry = ring_buffer_event_data(event);
 971         if (is_ret_probe(tu)) {
 972                 entry->vaddr[0] = func;
 973                 entry->vaddr[1] = instruction_pointer(regs);
 974                 data = DATAOF_TRACE_ENTRY(entry, true);
 975         } else {
 976                 entry->vaddr[0] = instruction_pointer(regs);
 977                 data = DATAOF_TRACE_ENTRY(entry, false);
 978         }
 979
 980         memcpy(data, ucb->buf, tu->tp.size + dsize);
 981
 982         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0);
 983 }
 984
 985 /* uprobe handler */
 986 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 987                              struct uprobe_cpu_buffer *ucb, int dsize)
 988 {
 989         struct event_file_link *link;
 990
 991         if (is_ret_probe(tu))
 992                 return 0;
 993
 994         rcu_read_lock();
 995         trace_probe_for_each_link_rcu(link, &tu->tp)
 996                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 997         rcu_read_unlock();
 998
 999         return 0;
1000 }
1001
1002 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1003                                  struct pt_regs *regs,
1004                                  struct uprobe_cpu_buffer *ucb, int dsize)
1005 {
1006         struct event_file_link *link;
1007
1008         rcu_read_lock();
1009         trace_probe_for_each_link_rcu(link, &tu->tp)
1010                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1011         rcu_read_unlock();
1012 }
1013
1014 /* Event entry printers */
1015 static enum print_line_t
1016 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1017 {
1018         struct uprobe_trace_entry_head *entry;
1019         struct trace_seq *s = &iter->seq;
1020         struct trace_uprobe *tu;
1021         u8 *data;
1022
1023         entry = (struct uprobe_trace_entry_head *)iter->ent;
1024         tu = trace_uprobe_primary_from_call(
1025                 container_of(event, struct trace_event_call, event));
1026         if (unlikely(!tu))
1027                 goto out;
1028
1029         if (is_ret_probe(tu)) {
1030                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1031                                  trace_probe_name(&tu->tp),
1032                                  entry->vaddr[1], entry->vaddr[0]);
1033                 data = DATAOF_TRACE_ENTRY(entry, true);
1034         } else {
1035                 trace_seq_printf(s, "%s: (0x%lx)",
1036                                  trace_probe_name(&tu->tp),
1037                                  entry->vaddr[0]);
1038                 data = DATAOF_TRACE_ENTRY(entry, false);
1039         }
1040
1041         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1042                 goto out;
1043
1044         trace_seq_putc(s, '\n');
1045
1046  out:
1047         return trace_handle_return(s);
1048 }
1049
1050 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1051                                 enum uprobe_filter_ctx ctx,
1052                                 struct mm_struct *mm);
1053
1054 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1055 {
1056         int ret;
1057
1058         tu->consumer.filter = filter;
1059         tu->inode = d_real_inode(tu->path.dentry);
1060
1061         if (tu->ref_ctr_offset)
1062                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1063                                 tu->ref_ctr_offset, &tu->consumer);
1064         else
1065                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1066
1067         if (ret)
1068                 tu->inode = NULL;
1069
1070         return ret;
1071 }
1072
1073 static void __probe_event_disable(struct trace_probe *tp)
1074 {
1075         struct trace_probe *pos;
1076         struct trace_uprobe *tu;
1077
1078         tu = container_of(tp, struct trace_uprobe, tp);
1079         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1080
1081         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1082                 tu = container_of(pos, struct trace_uprobe, tp);
1083                 if (!tu->inode)
1084                         continue;
1085
1086                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1087                 tu->inode = NULL;
1088         }
1089 }
1090
1091 static int probe_event_enable(struct trace_event_call *call,
1092                         struct trace_event_file *file, filter_func_t filter)
1093 {
1094         struct trace_probe *pos, *tp;
1095         struct trace_uprobe *tu;
1096         bool enabled;
1097         int ret;
1098
1099         tp = trace_probe_primary_from_call(call);
1100         if (WARN_ON_ONCE(!tp))
1101                 return -ENODEV;
1102         enabled = trace_probe_is_enabled(tp);
1103
1104         /* This may also change "enabled" state */
1105         if (file) {
1106                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1107                         return -EINTR;
1108
1109                 ret = trace_probe_add_file(tp, file);
1110                 if (ret < 0)
1111                         return ret;
1112         } else {
1113                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1114                         return -EINTR;
1115
1116                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1117         }
1118
1119         tu = container_of(tp, struct trace_uprobe, tp);
1120         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1121
1122         if (enabled)
1123                 return 0;
1124
1125         ret = uprobe_buffer_enable();
1126         if (ret)
1127                 goto err_flags;
1128
1129         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1130                 tu = container_of(pos, struct trace_uprobe, tp);
1131                 ret = trace_uprobe_enable(tu, filter);
1132                 if (ret) {
1133                         __probe_event_disable(tp);
1134                         goto err_buffer;
1135                 }
1136         }
1137
1138         return 0;
1139
1140  err_buffer:
1141         uprobe_buffer_disable();
1142
1143  err_flags:
1144         if (file)
1145                 trace_probe_remove_file(tp, file);
1146         else
1147                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1148
1149         return ret;
1150 }
1151
1152 static void probe_event_disable(struct trace_event_call *call,
1153                                 struct trace_event_file *file)
1154 {
1155         struct trace_probe *tp;
1156
1157         tp = trace_probe_primary_from_call(call);
1158         if (WARN_ON_ONCE(!tp))
1159                 return;
1160
1161         if (!trace_probe_is_enabled(tp))
1162                 return;
1163
1164         if (file) {
1165                 if (trace_probe_remove_file(tp, file) < 0)
1166                         return;
1167
1168                 if (trace_probe_is_enabled(tp))
1169                         return;
1170         } else
1171                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1172
1173         __probe_event_disable(tp);
1174         uprobe_buffer_disable();
1175 }
1176
1177 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1178 {
1179         int ret, size;
1180         struct uprobe_trace_entry_head field;
1181         struct trace_uprobe *tu;
1182
1183         tu = trace_uprobe_primary_from_call(event_call);
1184         if (unlikely(!tu))
1185                 return -ENODEV;
1186
1187         if (is_ret_probe(tu)) {
1188                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1189                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1190                 size = SIZEOF_TRACE_ENTRY(true);
1191         } else {
1192                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1193                 size = SIZEOF_TRACE_ENTRY(false);
1194         }
1195
1196         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1197 }
1198
1199 #ifdef CONFIG_PERF_EVENTS
1200 static bool
1201 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1202 {
1203         struct perf_event *event;
1204
1205         if (filter->nr_systemwide)
1206                 return true;
1207
1208         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1209                 if (event->hw.target->mm == mm)
1210                         return true;
1211         }
1212
1213         return false;
1214 }
1215
1216 static inline bool
1217 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1218                           struct perf_event *event)
1219 {
1220         return __uprobe_perf_filter(filter, event->hw.target->mm);
1221 }
1222
1223 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1224                                        struct perf_event *event)
1225 {
1226         bool done;
1227
1228         write_lock(&filter->rwlock);
1229         if (event->hw.target) {
1230                 list_del(&event->hw.tp_list);
1231                 done = filter->nr_systemwide ||
1232                         (event->hw.target->flags & PF_EXITING) ||
1233                         trace_uprobe_filter_event(filter, event);
1234         } else {
1235                 filter->nr_systemwide--;
1236                 done = filter->nr_systemwide;
1237         }
1238         write_unlock(&filter->rwlock);
1239
1240         return done;
1241 }
1242
1243 /* This returns true if the filter always covers target mm */
1244 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1245                                     struct perf_event *event)
1246 {
1247         bool done;
1248
1249         write_lock(&filter->rwlock);
1250         if (event->hw.target) {
1251                 /*
1252                  * event->parent != NULL means copy_process(), we can avoid
1253                  * uprobe_apply(). current->mm must be probed and we can rely
1254                  * on dup_mmap() which preserves the already installed bp's.
1255                  *
1256                  * attr.enable_on_exec means that exec/mmap will install the
1257                  * breakpoints we need.
1258                  */
1259                 done = filter->nr_systemwide ||
1260                         event->parent || event->attr.enable_on_exec ||
1261                         trace_uprobe_filter_event(filter, event);
1262                 list_add(&event->hw.tp_list, &filter->perf_events);
1263         } else {
1264                 done = filter->nr_systemwide;
1265                 filter->nr_systemwide++;
1266         }
1267         write_unlock(&filter->rwlock);
1268
1269         return done;
1270 }
1271
1272 static int uprobe_perf_close(struct trace_event_call *call,
1273                              struct perf_event *event)
1274 {
1275         struct trace_probe *pos, *tp;
1276         struct trace_uprobe *tu;
1277         int ret = 0;
1278
1279         tp = trace_probe_primary_from_call(call);
1280         if (WARN_ON_ONCE(!tp))
1281                 return -ENODEV;
1282
1283         tu = container_of(tp, struct trace_uprobe, tp);
1284         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1285                 return 0;
1286
1287         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1288                 tu = container_of(pos, struct trace_uprobe, tp);
1289                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1290                 if (ret)
1291                         break;
1292         }
1293
1294         return ret;
1295 }
1296
1297 static int uprobe_perf_open(struct trace_event_call *call,
1298                             struct perf_event *event)
1299 {
1300         struct trace_probe *pos, *tp;
1301         struct trace_uprobe *tu;
1302         int err = 0;
1303
1304         tp = trace_probe_primary_from_call(call);
1305         if (WARN_ON_ONCE(!tp))
1306                 return -ENODEV;
1307
1308         tu = container_of(tp, struct trace_uprobe, tp);
1309         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1310                 return 0;
1311
1312         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1313                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1314                 if (err) {
1315                         uprobe_perf_close(call, event);
1316                         break;
1317                 }
1318         }
1319
1320         return err;
1321 }
1322
1323 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1324                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1325 {
1326         struct trace_uprobe_filter *filter;
1327         struct trace_uprobe *tu;
1328         int ret;
1329
1330         tu = container_of(uc, struct trace_uprobe, consumer);
1331         filter = tu->tp.event->filter;
1332
1333         read_lock(&filter->rwlock);
1334         ret = __uprobe_perf_filter(filter, mm);
1335         read_unlock(&filter->rwlock);
1336
1337         return ret;
1338 }
1339
1340 static void __uprobe_perf_func(struct trace_uprobe *tu,
1341                                unsigned long func, struct pt_regs *regs,
1342                                struct uprobe_cpu_buffer *ucb, int dsize)
1343 {
1344         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1345         struct uprobe_trace_entry_head *entry;
1346         struct hlist_head *head;
1347         void *data;
1348         int size, esize;
1349         int rctx;
1350
1351         if (bpf_prog_array_valid(call)) {
1352                 u32 ret;
1353
1354                 preempt_disable();
1355                 ret = trace_call_bpf(call, regs);
1356                 preempt_enable();
1357                 if (!ret)
1358                         return;
1359         }
1360
1361         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1362
1363         size = esize + tu->tp.size + dsize;
1364         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1365         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1366                 return;
1367
1368         preempt_disable();
1369         head = this_cpu_ptr(call->perf_events);
1370         if (hlist_empty(head))
1371                 goto out;
1372
1373         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1374         if (!entry)
1375                 goto out;
1376
1377         if (is_ret_probe(tu)) {
1378                 entry->vaddr[0] = func;
1379                 entry->vaddr[1] = instruction_pointer(regs);
1380                 data = DATAOF_TRACE_ENTRY(entry, true);
1381         } else {
1382                 entry->vaddr[0] = instruction_pointer(regs);
1383                 data = DATAOF_TRACE_ENTRY(entry, false);
1384         }
1385
1386         memcpy(data, ucb->buf, tu->tp.size + dsize);
1387
1388         if (size - esize > tu->tp.size + dsize) {
1389                 int len = tu->tp.size + dsize;
1390
1391                 memset(data + len, 0, size - esize - len);
1392         }
1393
1394         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1395                               head, NULL);
1396  out:
1397         preempt_enable();
1398 }
1399
1400 /* uprobe profile handler */
1401 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1402                             struct uprobe_cpu_buffer *ucb, int dsize)
1403 {
1404         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1405                 return UPROBE_HANDLER_REMOVE;
1406
1407         if (!is_ret_probe(tu))
1408                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1409         return 0;
1410 }
1411
1412 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1413                                 struct pt_regs *regs,
1414                                 struct uprobe_cpu_buffer *ucb, int dsize)
1415 {
1416         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1417 }
1418
1419 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1420                         const char **filename, u64 *probe_offset,
1421                         bool perf_type_tracepoint)
1422 {
1423         const char *pevent = trace_event_name(event->tp_event);
1424         const char *group = event->tp_event->class->system;
1425         struct trace_uprobe *tu;
1426
1427         if (perf_type_tracepoint)
1428                 tu = find_probe_event(pevent, group);
1429         else
1430                 tu = trace_uprobe_primary_from_call(event->tp_event);
1431         if (!tu)
1432                 return -EINVAL;
1433
1434         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1435                                     : BPF_FD_TYPE_UPROBE;
1436         *filename = tu->filename;
1437         *probe_offset = tu->offset;
1438         return 0;
1439 }
1440 #endif  /* CONFIG_PERF_EVENTS */
1441
1442 static int
1443 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1444                       void *data)
1445 {
1446         struct trace_event_file *file = data;
1447
1448         switch (type) {
1449         case TRACE_REG_REGISTER:
1450                 return probe_event_enable(event, file, NULL);
1451
1452         case TRACE_REG_UNREGISTER:
1453                 probe_event_disable(event, file);
1454                 return 0;
1455
1456 #ifdef CONFIG_PERF_EVENTS
1457         case TRACE_REG_PERF_REGISTER:
1458                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1459
1460         case TRACE_REG_PERF_UNREGISTER:
1461                 probe_event_disable(event, NULL);
1462                 return 0;
1463
1464         case TRACE_REG_PERF_OPEN:
1465                 return uprobe_perf_open(event, data);
1466
1467         case TRACE_REG_PERF_CLOSE:
1468                 return uprobe_perf_close(event, data);
1469
1470 #endif
1471         default:
1472                 return 0;
1473         }
1474 }
1475
1476 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1477 {
1478         struct trace_uprobe *tu;
1479         struct uprobe_dispatch_data udd;
1480         struct uprobe_cpu_buffer *ucb;
1481         int dsize, esize;
1482         int ret = 0;
1483
1484
1485         tu = container_of(con, struct trace_uprobe, consumer);
1486         tu->nhit++;
1487
1488         udd.tu = tu;
1489         udd.bp_addr = instruction_pointer(regs);
1490
1491         current->utask->vaddr = (unsigned long) &udd;
1492
1493         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1494                 return 0;
1495
1496         dsize = __get_data_size(&tu->tp, regs);
1497         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1498
1499         ucb = uprobe_buffer_get();
1500         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1501
1502         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1503                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1504
1505 #ifdef CONFIG_PERF_EVENTS
1506         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1507                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1508 #endif
1509         uprobe_buffer_put(ucb);
1510         return ret;
1511 }
1512
1513 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1514                                 unsigned long func, struct pt_regs *regs)
1515 {
1516         struct trace_uprobe *tu;
1517         struct uprobe_dispatch_data udd;
1518         struct uprobe_cpu_buffer *ucb;
1519         int dsize, esize;
1520
1521         tu = container_of(con, struct trace_uprobe, consumer);
1522
1523         udd.tu = tu;
1524         udd.bp_addr = func;
1525
1526         current->utask->vaddr = (unsigned long) &udd;
1527
1528         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1529                 return 0;
1530
1531         dsize = __get_data_size(&tu->tp, regs);
1532         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1533
1534         ucb = uprobe_buffer_get();
1535         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1536
1537         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1538                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1539
1540 #ifdef CONFIG_PERF_EVENTS
1541         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1542                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1543 #endif
1544         uprobe_buffer_put(ucb);
1545         return 0;
1546 }
1547
1548 static struct trace_event_functions uprobe_funcs = {
1549         .trace          = print_uprobe_event
1550 };
1551
1552 static struct trace_event_fields uprobe_fields_array[] = {
1553         { .type = TRACE_FUNCTION_TYPE,
1554           .define_fields = uprobe_event_define_fields },
1555         {}
1556 };
1557
1558 static inline void init_trace_event_call(struct trace_uprobe *tu)
1559 {
1560         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1561         call->event.funcs = &uprobe_funcs;
1562         call->class->fields_array = uprobe_fields_array;
1563
1564         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1565         call->class->reg = trace_uprobe_register;
1566 }
1567
1568 static int register_uprobe_event(struct trace_uprobe *tu)
1569 {
1570         init_trace_event_call(tu);
1571
1572         return trace_probe_register_event_call(&tu->tp);
1573 }
1574
1575 static int unregister_uprobe_event(struct trace_uprobe *tu)
1576 {
1577         return trace_probe_unregister_event_call(&tu->tp);
1578 }
1579
1580 #ifdef CONFIG_PERF_EVENTS
1581 struct trace_event_call *
1582 create_local_trace_uprobe(char *name, unsigned long offs,
1583                           unsigned long ref_ctr_offset, bool is_return)
1584 {
1585         struct trace_uprobe *tu;
1586         struct path path;
1587         int ret;
1588
1589         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1590         if (ret)
1591                 return ERR_PTR(ret);
1592
1593         if (!d_is_reg(path.dentry)) {
1594                 path_put(&path);
1595                 return ERR_PTR(-EINVAL);
1596         }
1597
1598         /*
1599          * local trace_kprobes are not added to dyn_event, so they are never
1600          * searched in find_trace_kprobe(). Therefore, there is no concern of
1601          * duplicated name "DUMMY_EVENT" here.
1602          */
1603         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1604                                 is_return);
1605
1606         if (IS_ERR(tu)) {
1607                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1608                         (int)PTR_ERR(tu));
1609                 path_put(&path);
1610                 return ERR_CAST(tu);
1611         }
1612
1613         tu->offset = offs;
1614         tu->path = path;
1615         tu->ref_ctr_offset = ref_ctr_offset;
1616         tu->filename = kstrdup(name, GFP_KERNEL);
1617         init_trace_event_call(tu);
1618
1619         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1620                 ret = -ENOMEM;
1621                 goto error;
1622         }
1623
1624         return trace_probe_event_call(&tu->tp);
1625 error:
1626         free_trace_uprobe(tu);
1627         return ERR_PTR(ret);
1628 }
1629
1630 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1631 {
1632         struct trace_uprobe *tu;
1633
1634         tu = trace_uprobe_primary_from_call(event_call);
1635
1636         free_trace_uprobe(tu);
1637 }
1638 #endif /* CONFIG_PERF_EVENTS */
1639
1640 /* Make a trace interface for controlling probe points */
1641 static __init int init_uprobe_trace(void)
1642 {
1643         int ret;
1644
1645         ret = dyn_event_register(&trace_uprobe_ops);
1646         if (ret)
1647                 return ret;
1648
1649         ret = tracing_init_dentry();
1650         if (ret)
1651                 return 0;
1652
1653         trace_create_file("uprobe_events", 0644, NULL,
1654                                     NULL, &uprobe_events_ops);
1655         /* Profile interface */
1656         trace_create_file("uprobe_profile", 0444, NULL,
1657                                     NULL, &uprobe_profile_ops);
1658         return 0;
1659 }
1660
1661 fs_initcall(init_uprobe_trace);