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/bpf-cgroup.h>
  11 #include <linux/security.h>
  12 #include <linux/ctype.h>
  13 #include <linux/module.h>
  14 #include <linux/uaccess.h>
  15 #include <linux/uprobes.h>
  16 #include <linux/namei.h>
  17 #include <linux/string.h>
  18 #include <linux/rculist.h>
  19 #include <linux/filter.h>
  20
  21 #include "trace_dynevent.h"
  22 #include "trace_probe.h"
  23 #include "trace_probe_tmpl.h"
  24
  25 #define UPROBE_EVENT_SYSTEM     "uprobes"
  26
  27 struct uprobe_trace_entry_head {
  28         struct trace_entry      ent;
  29         unsigned long           vaddr[];
  30 };
  31
  32 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  33         (sizeof(struct uprobe_trace_entry_head) +       \
  34          sizeof(unsigned long) * (is_return ? 2 : 1))
  35
  36 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  37         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  38
  39 static int trace_uprobe_create(const char *raw_command);
  40 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  41 static int trace_uprobe_release(struct dyn_event *ev);
  42 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  43 static bool trace_uprobe_match(const char *system, const char *event,
  44                         int argc, const char **argv, struct dyn_event *ev);
  45
  46 static struct dyn_event_operations trace_uprobe_ops = {
  47         .create = trace_uprobe_create,
  48         .show = trace_uprobe_show,
  49         .is_busy = trace_uprobe_is_busy,
  50         .free = trace_uprobe_release,
  51         .match = trace_uprobe_match,
  52 };
  53
  54 /*
  55  * uprobe event core functions
  56  */
  57 struct trace_uprobe {
  58         struct dyn_event                devent;
  59         struct uprobe_consumer          consumer;
  60         struct path                     path;
  61         struct inode                    *inode;
  62         char                            *filename;
  63         unsigned long                   offset;
  64         unsigned long                   ref_ctr_offset;
  65         unsigned long                   nhit;
  66         struct trace_probe              tp;
  67 };
  68
  69 static bool is_trace_uprobe(struct dyn_event *ev)
  70 {
  71         return ev->ops == &trace_uprobe_ops;
  72 }
  73
  74 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  75 {
  76         return container_of(ev, struct trace_uprobe, devent);
  77 }
  78
  79 /**
  80  * for_each_trace_uprobe - iterate over the trace_uprobe list
  81  * @pos:        the struct trace_uprobe * for each entry
  82  * @dpos:       the struct dyn_event * to use as a loop cursor
  83  */
  84 #define for_each_trace_uprobe(pos, dpos)        \
  85         for_each_dyn_event(dpos)                \
  86                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  87
  88 static int register_uprobe_event(struct trace_uprobe *tu);
  89 static int unregister_uprobe_event(struct trace_uprobe *tu);
  90
  91 struct uprobe_dispatch_data {
  92         struct trace_uprobe     *tu;
  93         unsigned long           bp_addr;
  94 };
  95
  96 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  97 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  98                                 unsigned long func, struct pt_regs *regs);
  99
 100 #ifdef CONFIG_STACK_GROWSUP
 101 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 102 {
 103         return addr - (n * sizeof(long));
 104 }
 105 #else
 106 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 107 {
 108         return addr + (n * sizeof(long));
 109 }
 110 #endif
 111
 112 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 113 {
 114         unsigned long ret;
 115         unsigned long addr = user_stack_pointer(regs);
 116
 117         addr = adjust_stack_addr(addr, n);
 118
 119         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 120                 return 0;
 121
 122         return ret;
 123 }
 124
 125 /*
 126  * Uprobes-specific fetch functions
 127  */
 128 static nokprobe_inline int
 129 probe_mem_read(void *dest, void *src, size_t size)
 130 {
 131         void __user *vaddr = (void __force __user *)src;
 132
 133         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 134 }
 135
 136 static nokprobe_inline int
 137 probe_mem_read_user(void *dest, void *src, size_t size)
 138 {
 139         return probe_mem_read(dest, src, size);
 140 }
 141
 142 /*
 143  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 144  * length and relative data location.
 145  */
 146 static nokprobe_inline int
 147 fetch_store_string(unsigned long addr, void *dest, void *base)
 148 {
 149         long ret;
 150         u32 loc = *(u32 *)dest;
 151         int maxlen  = get_loc_len(loc);
 152         u8 *dst = get_loc_data(dest, base);
 153         void __user *src = (void __force __user *) addr;
 154
 155         if (unlikely(!maxlen))
 156                 return -ENOMEM;
 157
 158         if (addr == FETCH_TOKEN_COMM)
 159                 ret = strlcpy(dst, current->comm, maxlen);
 160         else
 161                 ret = strncpy_from_user(dst, src, maxlen);
 162         if (ret >= 0) {
 163                 if (ret == maxlen)
 164                         dst[ret - 1] = '\0';
 165                 else
 166                         /*
 167                          * Include the terminating null byte. In this case it
 168                          * was copied by strncpy_from_user but not accounted
 169                          * for in ret.
 170                          */
 171                         ret++;
 172                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 173         }
 174
 175         return ret;
 176 }
 177
 178 static nokprobe_inline int
 179 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 180 {
 181         return fetch_store_string(addr, dest, base);
 182 }
 183
 184 /* Return the length of string -- including null terminal byte */
 185 static nokprobe_inline int
 186 fetch_store_strlen(unsigned long addr)
 187 {
 188         int len;
 189         void __user *vaddr = (void __force __user *) addr;
 190
 191         if (addr == FETCH_TOKEN_COMM)
 192                 len = strlen(current->comm) + 1;
 193         else
 194                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 195
 196         return (len > MAX_STRING_SIZE) ? 0 : len;
 197 }
 198
 199 static nokprobe_inline int
 200 fetch_store_strlen_user(unsigned long addr)
 201 {
 202         return fetch_store_strlen(addr);
 203 }
 204
 205 static unsigned long translate_user_vaddr(unsigned long file_offset)
 206 {
 207         unsigned long base_addr;
 208         struct uprobe_dispatch_data *udd;
 209
 210         udd = (void *) current->utask->vaddr;
 211
 212         base_addr = udd->bp_addr - udd->tu->offset;
 213         return base_addr + file_offset;
 214 }
 215
 216 /* Note that we don't verify it, since the code does not come from user space */
 217 static int
 218 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 219                    void *base)
 220 {
 221         struct pt_regs *regs = rec;
 222         unsigned long val;
 223         int ret;
 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_COMM:
 240                 val = FETCH_TOKEN_COMM;
 241                 break;
 242         case FETCH_OP_FOFFS:
 243                 val = translate_user_vaddr(code->immediate);
 244                 break;
 245         default:
 246                 ret = process_common_fetch_insn(code, &val);
 247                 if (ret < 0)
 248                         return ret;
 249         }
 250         code++;
 251
 252         return process_fetch_insn_bottom(code, val, dest, base);
 253 }
 254 NOKPROBE_SYMBOL(process_fetch_insn)
 255
 256 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 257 {
 258         rwlock_init(&filter->rwlock);
 259         filter->nr_systemwide = 0;
 260         INIT_LIST_HEAD(&filter->perf_events);
 261 }
 262
 263 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 264 {
 265         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 266 }
 267
 268 static inline bool is_ret_probe(struct trace_uprobe *tu)
 269 {
 270         return tu->consumer.ret_handler != NULL;
 271 }
 272
 273 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 274 {
 275         struct trace_uprobe *tu = to_trace_uprobe(ev);
 276
 277         return trace_probe_is_enabled(&tu->tp);
 278 }
 279
 280 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 281                                             int argc, const char **argv)
 282 {
 283         char buf[MAX_ARGSTR_LEN + 1];
 284         int len;
 285
 286         if (!argc)
 287                 return true;
 288
 289         len = strlen(tu->filename);
 290         if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 291                 return false;
 292
 293         if (tu->ref_ctr_offset == 0)
 294                 snprintf(buf, sizeof(buf), "0x%0*lx",
 295                                 (int)(sizeof(void *) * 2), tu->offset);
 296         else
 297                 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 298                                 (int)(sizeof(void *) * 2), tu->offset,
 299                                 tu->ref_ctr_offset);
 300         if (strcmp(buf, &argv[0][len + 1]))
 301                 return false;
 302
 303         argc--; argv++;
 304
 305         return trace_probe_match_command_args(&tu->tp, argc, argv);
 306 }
 307
 308 static bool trace_uprobe_match(const char *system, const char *event,
 309                         int argc, const char **argv, struct dyn_event *ev)
 310 {
 311         struct trace_uprobe *tu = to_trace_uprobe(ev);
 312
 313         return (event[0] == '\0' ||
 314                 strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 315            (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 316            trace_uprobe_match_command_head(tu, argc, argv);
 317 }
 318
 319 static nokprobe_inline struct trace_uprobe *
 320 trace_uprobe_primary_from_call(struct trace_event_call *call)
 321 {
 322         struct trace_probe *tp;
 323
 324         tp = trace_probe_primary_from_call(call);
 325         if (WARN_ON_ONCE(!tp))
 326                 return NULL;
 327
 328         return container_of(tp, struct trace_uprobe, tp);
 329 }
 330
 331 /*
 332  * Allocate new trace_uprobe and initialize it (including uprobes).
 333  */
 334 static struct trace_uprobe *
 335 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 336 {
 337         struct trace_uprobe *tu;
 338         int ret;
 339
 340         tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 341         if (!tu)
 342                 return ERR_PTR(-ENOMEM);
 343
 344         ret = trace_probe_init(&tu->tp, event, group, true);
 345         if (ret < 0)
 346                 goto error;
 347
 348         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 349         tu->consumer.handler = uprobe_dispatcher;
 350         if (is_ret)
 351                 tu->consumer.ret_handler = uretprobe_dispatcher;
 352         init_trace_uprobe_filter(tu->tp.event->filter);
 353         return tu;
 354
 355 error:
 356         kfree(tu);
 357
 358         return ERR_PTR(ret);
 359 }
 360
 361 static void free_trace_uprobe(struct trace_uprobe *tu)
 362 {
 363         if (!tu)
 364                 return;
 365
 366         path_put(&tu->path);
 367         trace_probe_cleanup(&tu->tp);
 368         kfree(tu->filename);
 369         kfree(tu);
 370 }
 371
 372 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 373 {
 374         struct dyn_event *pos;
 375         struct trace_uprobe *tu;
 376
 377         for_each_trace_uprobe(tu, pos)
 378                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 379                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 380                         return tu;
 381
 382         return NULL;
 383 }
 384
 385 /* Unregister a trace_uprobe and probe_event */
 386 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 387 {
 388         int ret;
 389
 390         if (trace_probe_has_sibling(&tu->tp))
 391                 goto unreg;
 392
 393         /* If there's a reference to the dynamic event */
 394         if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 395                 return -EBUSY;
 396
 397         ret = unregister_uprobe_event(tu);
 398         if (ret)
 399                 return ret;
 400
 401 unreg:
 402         dyn_event_remove(&tu->devent);
 403         trace_probe_unlink(&tu->tp);
 404         free_trace_uprobe(tu);
 405         return 0;
 406 }
 407
 408 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 409                                          struct trace_uprobe *comp)
 410 {
 411         struct trace_probe_event *tpe = orig->tp.event;
 412         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 413         int i;
 414
 415         list_for_each_entry(orig, &tpe->probes, tp.list) {
 416                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 417                     comp->offset != orig->offset)
 418                         continue;
 419
 420                 /*
 421                  * trace_probe_compare_arg_type() ensured that nr_args and
 422                  * each argument name and type are same. Let's compare comm.
 423                  */
 424                 for (i = 0; i < orig->tp.nr_args; i++) {
 425                         if (strcmp(orig->tp.args[i].comm,
 426                                    comp->tp.args[i].comm))
 427                                 break;
 428                 }
 429
 430                 if (i == orig->tp.nr_args)
 431                         return true;
 432         }
 433
 434         return false;
 435 }
 436
 437 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 438 {
 439         int ret;
 440
 441         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 442         if (ret) {
 443                 /* Note that argument starts index = 2 */
 444                 trace_probe_log_set_index(ret + 1);
 445                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 446                 return -EEXIST;
 447         }
 448         if (trace_uprobe_has_same_uprobe(to, tu)) {
 449                 trace_probe_log_set_index(0);
 450                 trace_probe_log_err(0, SAME_PROBE);
 451                 return -EEXIST;
 452         }
 453
 454         /* Append to existing event */
 455         ret = trace_probe_append(&tu->tp, &to->tp);
 456         if (!ret)
 457                 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 458
 459         return ret;
 460 }
 461
 462 /*
 463  * Uprobe with multiple reference counter is not allowed. i.e.
 464  * If inode and offset matches, reference counter offset *must*
 465  * match as well. Though, there is one exception: If user is
 466  * replacing old trace_uprobe with new one(same group/event),
 467  * then we allow same uprobe with new reference counter as far
 468  * as the new one does not conflict with any other existing
 469  * ones.
 470  */
 471 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 472 {
 473         struct dyn_event *pos;
 474         struct trace_uprobe *tmp;
 475         struct inode *new_inode = d_real_inode(new->path.dentry);
 476
 477         for_each_trace_uprobe(tmp, pos) {
 478                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 479                     new->offset == tmp->offset &&
 480                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 481                         pr_warn("Reference counter offset mismatch.");
 482                         return -EINVAL;
 483                 }
 484         }
 485         return 0;
 486 }
 487
 488 /* Register a trace_uprobe and probe_event */
 489 static int register_trace_uprobe(struct trace_uprobe *tu)
 490 {
 491         struct trace_uprobe *old_tu;
 492         int ret;
 493
 494         mutex_lock(&event_mutex);
 495
 496         ret = validate_ref_ctr_offset(tu);
 497         if (ret)
 498                 goto end;
 499
 500         /* register as an event */
 501         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 502                                   trace_probe_group_name(&tu->tp));
 503         if (old_tu) {
 504                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 505                         trace_probe_log_set_index(0);
 506                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 507                         ret = -EEXIST;
 508                 } else {
 509                         ret = append_trace_uprobe(tu, old_tu);
 510                 }
 511                 goto end;
 512         }
 513
 514         ret = register_uprobe_event(tu);
 515         if (ret) {
 516                 if (ret == -EEXIST) {
 517                         trace_probe_log_set_index(0);
 518                         trace_probe_log_err(0, EVENT_EXIST);
 519                 } else
 520                         pr_warn("Failed to register probe event(%d)\n", ret);
 521                 goto end;
 522         }
 523
 524         dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 525
 526 end:
 527         mutex_unlock(&event_mutex);
 528
 529         return ret;
 530 }
 531
 532 /*
 533  * Argument syntax:
 534  *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 535  */
 536 static int __trace_uprobe_create(int argc, const char **argv)
 537 {
 538         struct trace_uprobe *tu;
 539         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 540         char *arg, *filename, *rctr, *rctr_end, *tmp;
 541         char buf[MAX_EVENT_NAME_LEN];
 542         char gbuf[MAX_EVENT_NAME_LEN];
 543         enum probe_print_type ptype;
 544         struct path path;
 545         unsigned long offset, ref_ctr_offset;
 546         bool is_return = false;
 547         int i, ret;
 548
 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, gbuf,
 648                                                   event - argv[0]);
 649                 if (ret)
 650                         goto fail_address_parse;
 651         }
 652
 653         if (!event) {
 654                 char *tail;
 655                 char *ptr;
 656
 657                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 658                 if (!tail) {
 659                         ret = -ENOMEM;
 660                         goto fail_address_parse;
 661                 }
 662
 663                 ptr = strpbrk(tail, ".-_");
 664                 if (ptr)
 665                         *ptr = '\0';
 666
 667                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 668                 event = buf;
 669                 kfree(tail);
 670         }
 671
 672         argc -= 2;
 673         argv += 2;
 674
 675         tu = alloc_trace_uprobe(group, event, argc, is_return);
 676         if (IS_ERR(tu)) {
 677                 ret = PTR_ERR(tu);
 678                 /* This must return -ENOMEM otherwise there is a bug */
 679                 WARN_ON_ONCE(ret != -ENOMEM);
 680                 goto fail_address_parse;
 681         }
 682         tu->offset = offset;
 683         tu->ref_ctr_offset = ref_ctr_offset;
 684         tu->path = path;
 685         tu->filename = filename;
 686
 687         /* parse arguments */
 688         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 689                 struct traceprobe_parse_context ctx = {
 690                         .flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
 691                 };
 692
 693                 trace_probe_log_set_index(i + 2);
 694                 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
 695                 if (ret)
 696                         goto error;
 697         }
 698
 699         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 700         ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 701         if (ret < 0)
 702                 goto error;
 703
 704         ret = register_trace_uprobe(tu);
 705         if (!ret)
 706                 goto out;
 707
 708 error:
 709         free_trace_uprobe(tu);
 710 out:
 711         trace_probe_log_clear();
 712         return ret;
 713
 714 fail_address_parse:
 715         trace_probe_log_clear();
 716         path_put(&path);
 717         kfree(filename);
 718
 719         return ret;
 720 }
 721
 722 int trace_uprobe_create(const char *raw_command)
 723 {
 724         return trace_probe_create(raw_command, __trace_uprobe_create);
 725 }
 726
 727 static int create_or_delete_trace_uprobe(const char *raw_command)
 728 {
 729         int ret;
 730
 731         if (raw_command[0] == '-')
 732                 return dyn_event_release(raw_command, &trace_uprobe_ops);
 733
 734         ret = trace_uprobe_create(raw_command);
 735         return ret == -ECANCELED ? -EINVAL : ret;
 736 }
 737
 738 static int trace_uprobe_release(struct dyn_event *ev)
 739 {
 740         struct trace_uprobe *tu = to_trace_uprobe(ev);
 741
 742         return unregister_trace_uprobe(tu);
 743 }
 744
 745 /* Probes listing interfaces */
 746 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 747 {
 748         struct trace_uprobe *tu = to_trace_uprobe(ev);
 749         char c = is_ret_probe(tu) ? 'r' : 'p';
 750         int i;
 751
 752         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 753                         trace_probe_name(&tu->tp), tu->filename,
 754                         (int)(sizeof(void *) * 2), tu->offset);
 755
 756         if (tu->ref_ctr_offset)
 757                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 758
 759         for (i = 0; i < tu->tp.nr_args; i++)
 760                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 761
 762         seq_putc(m, '\n');
 763         return 0;
 764 }
 765
 766 static int probes_seq_show(struct seq_file *m, void *v)
 767 {
 768         struct dyn_event *ev = v;
 769
 770         if (!is_trace_uprobe(ev))
 771                 return 0;
 772
 773         return trace_uprobe_show(m, ev);
 774 }
 775
 776 static const struct seq_operations probes_seq_op = {
 777         .start  = dyn_event_seq_start,
 778         .next   = dyn_event_seq_next,
 779         .stop   = dyn_event_seq_stop,
 780         .show   = probes_seq_show
 781 };
 782
 783 static int probes_open(struct inode *inode, struct file *file)
 784 {
 785         int ret;
 786
 787         ret = security_locked_down(LOCKDOWN_TRACEFS);
 788         if (ret)
 789                 return ret;
 790
 791         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 792                 ret = dyn_events_release_all(&trace_uprobe_ops);
 793                 if (ret)
 794                         return ret;
 795         }
 796
 797         return seq_open(file, &probes_seq_op);
 798 }
 799
 800 static ssize_t probes_write(struct file *file, const char __user *buffer,
 801                             size_t count, loff_t *ppos)
 802 {
 803         return trace_parse_run_command(file, buffer, count, ppos,
 804                                         create_or_delete_trace_uprobe);
 805 }
 806
 807 static const struct file_operations uprobe_events_ops = {
 808         .owner          = THIS_MODULE,
 809         .open           = probes_open,
 810         .read           = seq_read,
 811         .llseek         = seq_lseek,
 812         .release        = seq_release,
 813         .write          = probes_write,
 814 };
 815
 816 /* Probes profiling interfaces */
 817 static int probes_profile_seq_show(struct seq_file *m, void *v)
 818 {
 819         struct dyn_event *ev = v;
 820         struct trace_uprobe *tu;
 821
 822         if (!is_trace_uprobe(ev))
 823                 return 0;
 824
 825         tu = to_trace_uprobe(ev);
 826         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 827                         trace_probe_name(&tu->tp), tu->nhit);
 828         return 0;
 829 }
 830
 831 static const struct seq_operations profile_seq_op = {
 832         .start  = dyn_event_seq_start,
 833         .next   = dyn_event_seq_next,
 834         .stop   = dyn_event_seq_stop,
 835         .show   = probes_profile_seq_show
 836 };
 837
 838 static int profile_open(struct inode *inode, struct file *file)
 839 {
 840         int ret;
 841
 842         ret = security_locked_down(LOCKDOWN_TRACEFS);
 843         if (ret)
 844                 return ret;
 845
 846         return seq_open(file, &profile_seq_op);
 847 }
 848
 849 static const struct file_operations uprobe_profile_ops = {
 850         .owner          = THIS_MODULE,
 851         .open           = profile_open,
 852         .read           = seq_read,
 853         .llseek         = seq_lseek,
 854         .release        = seq_release,
 855 };
 856
 857 struct uprobe_cpu_buffer {
 858         struct mutex mutex;
 859         void *buf;
 860 };
 861 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 862 static int uprobe_buffer_refcnt;
 863
 864 static int uprobe_buffer_init(void)
 865 {
 866         int cpu, err_cpu;
 867
 868         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 869         if (uprobe_cpu_buffer == NULL)
 870                 return -ENOMEM;
 871
 872         for_each_possible_cpu(cpu) {
 873                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 874                                                   GFP_KERNEL, 0);
 875                 if (p == NULL) {
 876                         err_cpu = cpu;
 877                         goto err;
 878                 }
 879                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 880                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 881         }
 882
 883         return 0;
 884
 885 err:
 886         for_each_possible_cpu(cpu) {
 887                 if (cpu == err_cpu)
 888                         break;
 889                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 890         }
 891
 892         free_percpu(uprobe_cpu_buffer);
 893         return -ENOMEM;
 894 }
 895
 896 static int uprobe_buffer_enable(void)
 897 {
 898         int ret = 0;
 899
 900         BUG_ON(!mutex_is_locked(&event_mutex));
 901
 902         if (uprobe_buffer_refcnt++ == 0) {
 903                 ret = uprobe_buffer_init();
 904                 if (ret < 0)
 905                         uprobe_buffer_refcnt--;
 906         }
 907
 908         return ret;
 909 }
 910
 911 static void uprobe_buffer_disable(void)
 912 {
 913         int cpu;
 914
 915         BUG_ON(!mutex_is_locked(&event_mutex));
 916
 917         if (--uprobe_buffer_refcnt == 0) {
 918                 for_each_possible_cpu(cpu)
 919                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 920                                                              cpu)->buf);
 921
 922                 free_percpu(uprobe_cpu_buffer);
 923                 uprobe_cpu_buffer = NULL;
 924         }
 925 }
 926
 927 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 928 {
 929         struct uprobe_cpu_buffer *ucb;
 930         int cpu;
 931
 932         cpu = raw_smp_processor_id();
 933         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 934
 935         /*
 936          * Use per-cpu buffers for fastest access, but we might migrate
 937          * so the mutex makes sure we have sole access to it.
 938          */
 939         mutex_lock(&ucb->mutex);
 940
 941         return ucb;
 942 }
 943
 944 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 945 {
 946         mutex_unlock(&ucb->mutex);
 947 }
 948
 949 static void __uprobe_trace_func(struct trace_uprobe *tu,
 950                                 unsigned long func, struct pt_regs *regs,
 951                                 struct uprobe_cpu_buffer *ucb, int dsize,
 952                                 struct trace_event_file *trace_file)
 953 {
 954         struct uprobe_trace_entry_head *entry;
 955         struct trace_event_buffer fbuffer;
 956         void *data;
 957         int size, esize;
 958         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 959
 960         WARN_ON(call != trace_file->event_call);
 961
 962         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 963                 return;
 964
 965         if (trace_trigger_soft_disabled(trace_file))
 966                 return;
 967
 968         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 969         size = esize + tu->tp.size + dsize;
 970         entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 971         if (!entry)
 972                 return;
 973
 974         if (is_ret_probe(tu)) {
 975                 entry->vaddr[0] = func;
 976                 entry->vaddr[1] = instruction_pointer(regs);
 977                 data = DATAOF_TRACE_ENTRY(entry, true);
 978         } else {
 979                 entry->vaddr[0] = instruction_pointer(regs);
 980                 data = DATAOF_TRACE_ENTRY(entry, false);
 981         }
 982
 983         memcpy(data, ucb->buf, tu->tp.size + dsize);
 984
 985         trace_event_buffer_commit(&fbuffer);
 986 }
 987
 988 /* uprobe handler */
 989 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 990                              struct uprobe_cpu_buffer *ucb, int dsize)
 991 {
 992         struct event_file_link *link;
 993
 994         if (is_ret_probe(tu))
 995                 return 0;
 996
 997         rcu_read_lock();
 998         trace_probe_for_each_link_rcu(link, &tu->tp)
 999                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1000         rcu_read_unlock();
1001
1002         return 0;
1003 }
1004
1005 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1006                                  struct pt_regs *regs,
1007                                  struct uprobe_cpu_buffer *ucb, int dsize)
1008 {
1009         struct event_file_link *link;
1010
1011         rcu_read_lock();
1012         trace_probe_for_each_link_rcu(link, &tu->tp)
1013                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1014         rcu_read_unlock();
1015 }
1016
1017 /* Event entry printers */
1018 static enum print_line_t
1019 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1020 {
1021         struct uprobe_trace_entry_head *entry;
1022         struct trace_seq *s = &iter->seq;
1023         struct trace_uprobe *tu;
1024         u8 *data;
1025
1026         entry = (struct uprobe_trace_entry_head *)iter->ent;
1027         tu = trace_uprobe_primary_from_call(
1028                 container_of(event, struct trace_event_call, event));
1029         if (unlikely(!tu))
1030                 goto out;
1031
1032         if (is_ret_probe(tu)) {
1033                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1034                                  trace_probe_name(&tu->tp),
1035                                  entry->vaddr[1], entry->vaddr[0]);
1036                 data = DATAOF_TRACE_ENTRY(entry, true);
1037         } else {
1038                 trace_seq_printf(s, "%s: (0x%lx)",
1039                                  trace_probe_name(&tu->tp),
1040                                  entry->vaddr[0]);
1041                 data = DATAOF_TRACE_ENTRY(entry, false);
1042         }
1043
1044         if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1045                 goto out;
1046
1047         trace_seq_putc(s, '\n');
1048
1049  out:
1050         return trace_handle_return(s);
1051 }
1052
1053 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1054                                 enum uprobe_filter_ctx ctx,
1055                                 struct mm_struct *mm);
1056
1057 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1058 {
1059         int ret;
1060
1061         tu->consumer.filter = filter;
1062         tu->inode = d_real_inode(tu->path.dentry);
1063
1064         if (tu->ref_ctr_offset)
1065                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1066                                 tu->ref_ctr_offset, &tu->consumer);
1067         else
1068                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1069
1070         if (ret)
1071                 tu->inode = NULL;
1072
1073         return ret;
1074 }
1075
1076 static void __probe_event_disable(struct trace_probe *tp)
1077 {
1078         struct trace_uprobe *tu;
1079
1080         tu = container_of(tp, struct trace_uprobe, tp);
1081         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1082
1083         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1084                 if (!tu->inode)
1085                         continue;
1086
1087                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1088                 tu->inode = NULL;
1089         }
1090 }
1091
1092 static int probe_event_enable(struct trace_event_call *call,
1093                         struct trace_event_file *file, filter_func_t filter)
1094 {
1095         struct trace_probe *tp;
1096         struct trace_uprobe *tu;
1097         bool enabled;
1098         int ret;
1099
1100         tp = trace_probe_primary_from_call(call);
1101         if (WARN_ON_ONCE(!tp))
1102                 return -ENODEV;
1103         enabled = trace_probe_is_enabled(tp);
1104
1105         /* This may also change "enabled" state */
1106         if (file) {
1107                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1108                         return -EINTR;
1109
1110                 ret = trace_probe_add_file(tp, file);
1111                 if (ret < 0)
1112                         return ret;
1113         } else {
1114                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1115                         return -EINTR;
1116
1117                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1118         }
1119
1120         tu = container_of(tp, struct trace_uprobe, tp);
1121         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1122
1123         if (enabled)
1124                 return 0;
1125
1126         ret = uprobe_buffer_enable();
1127         if (ret)
1128                 goto err_flags;
1129
1130         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
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 *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(tu, trace_probe_probe_list(tp), tp.list) {
1288                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1289                 if (ret)
1290                         break;
1291         }
1292
1293         return ret;
1294 }
1295
1296 static int uprobe_perf_open(struct trace_event_call *call,
1297                             struct perf_event *event)
1298 {
1299         struct trace_probe *tp;
1300         struct trace_uprobe *tu;
1301         int err = 0;
1302
1303         tp = trace_probe_primary_from_call(call);
1304         if (WARN_ON_ONCE(!tp))
1305                 return -ENODEV;
1306
1307         tu = container_of(tp, struct trace_uprobe, tp);
1308         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1309                 return 0;
1310
1311         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1312                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1313                 if (err) {
1314                         uprobe_perf_close(call, event);
1315                         break;
1316                 }
1317         }
1318
1319         return err;
1320 }
1321
1322 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1323                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1324 {
1325         struct trace_uprobe_filter *filter;
1326         struct trace_uprobe *tu;
1327         int ret;
1328
1329         tu = container_of(uc, struct trace_uprobe, consumer);
1330         filter = tu->tp.event->filter;
1331
1332         read_lock(&filter->rwlock);
1333         ret = __uprobe_perf_filter(filter, mm);
1334         read_unlock(&filter->rwlock);
1335
1336         return ret;
1337 }
1338
1339 static void __uprobe_perf_func(struct trace_uprobe *tu,
1340                                unsigned long func, struct pt_regs *regs,
1341                                struct uprobe_cpu_buffer *ucb, int dsize)
1342 {
1343         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1344         struct uprobe_trace_entry_head *entry;
1345         struct hlist_head *head;
1346         void *data;
1347         int size, esize;
1348         int rctx;
1349
1350 #ifdef CONFIG_BPF_EVENTS
1351         if (bpf_prog_array_valid(call)) {
1352                 u32 ret;
1353
1354                 ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
1355                 if (!ret)
1356                         return;
1357         }
1358 #endif /* CONFIG_BPF_EVENTS */
1359
1360         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1361
1362         size = esize + tu->tp.size + dsize;
1363         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1364         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1365                 return;
1366
1367         preempt_disable();
1368         head = this_cpu_ptr(call->perf_events);
1369         if (hlist_empty(head))
1370                 goto out;
1371
1372         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1373         if (!entry)
1374                 goto out;
1375
1376         if (is_ret_probe(tu)) {
1377                 entry->vaddr[0] = func;
1378                 entry->vaddr[1] = instruction_pointer(regs);
1379                 data = DATAOF_TRACE_ENTRY(entry, true);
1380         } else {
1381                 entry->vaddr[0] = instruction_pointer(regs);
1382                 data = DATAOF_TRACE_ENTRY(entry, false);
1383         }
1384
1385         memcpy(data, ucb->buf, tu->tp.size + dsize);
1386
1387         if (size - esize > tu->tp.size + dsize) {
1388                 int len = tu->tp.size + dsize;
1389
1390                 memset(data + len, 0, size - esize - len);
1391         }
1392
1393         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1394                               head, NULL);
1395  out:
1396         preempt_enable();
1397 }
1398
1399 /* uprobe profile handler */
1400 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1401                             struct uprobe_cpu_buffer *ucb, int dsize)
1402 {
1403         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1404                 return UPROBE_HANDLER_REMOVE;
1405
1406         if (!is_ret_probe(tu))
1407                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1408         return 0;
1409 }
1410
1411 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1412                                 struct pt_regs *regs,
1413                                 struct uprobe_cpu_buffer *ucb, int dsize)
1414 {
1415         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1416 }
1417
1418 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1419                         const char **filename, u64 *probe_offset,
1420                         u64 *probe_addr, bool perf_type_tracepoint)
1421 {
1422         const char *pevent = trace_event_name(event->tp_event);
1423         const char *group = event->tp_event->class->system;
1424         struct trace_uprobe *tu;
1425
1426         if (perf_type_tracepoint)
1427                 tu = find_probe_event(pevent, group);
1428         else
1429                 tu = trace_uprobe_primary_from_call(event->tp_event);
1430         if (!tu)
1431                 return -EINVAL;
1432
1433         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1434                                     : BPF_FD_TYPE_UPROBE;
1435         *filename = tu->filename;
1436         *probe_offset = tu->offset;
1437         *probe_addr = 0;
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         enum probe_print_type ptype;
1586         struct trace_uprobe *tu;
1587         struct path path;
1588         int ret;
1589
1590         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1591         if (ret)
1592                 return ERR_PTR(ret);
1593
1594         if (!d_is_reg(path.dentry)) {
1595                 path_put(&path);
1596                 return ERR_PTR(-EINVAL);
1597         }
1598
1599         /*
1600          * local trace_kprobes are not added to dyn_event, so they are never
1601          * searched in find_trace_kprobe(). Therefore, there is no concern of
1602          * duplicated name "DUMMY_EVENT" here.
1603          */
1604         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1605                                 is_return);
1606
1607         if (IS_ERR(tu)) {
1608                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1609                         (int)PTR_ERR(tu));
1610                 path_put(&path);
1611                 return ERR_CAST(tu);
1612         }
1613
1614         tu->offset = offs;
1615         tu->path = path;
1616         tu->ref_ctr_offset = ref_ctr_offset;
1617         tu->filename = kstrdup(name, GFP_KERNEL);
1618         if (!tu->filename) {
1619                 ret = -ENOMEM;
1620                 goto error;
1621         }
1622
1623         init_trace_event_call(tu);
1624
1625         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1626         if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1627                 ret = -ENOMEM;
1628                 goto error;
1629         }
1630
1631         return trace_probe_event_call(&tu->tp);
1632 error:
1633         free_trace_uprobe(tu);
1634         return ERR_PTR(ret);
1635 }
1636
1637 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1638 {
1639         struct trace_uprobe *tu;
1640
1641         tu = trace_uprobe_primary_from_call(event_call);
1642
1643         free_trace_uprobe(tu);
1644 }
1645 #endif /* CONFIG_PERF_EVENTS */
1646
1647 /* Make a trace interface for controlling probe points */
1648 static __init int init_uprobe_trace(void)
1649 {
1650         int ret;
1651
1652         ret = dyn_event_register(&trace_uprobe_ops);
1653         if (ret)
1654                 return ret;
1655
1656         ret = tracing_init_dentry();
1657         if (ret)
1658                 return 0;
1659
1660         trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1661                                     NULL, &uprobe_events_ops);
1662         /* Profile interface */
1663         trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1664                                     NULL, &uprobe_profile_ops);
1665         return 0;
1666 }
1667
1668 fs_initcall(init_uprobe_trace);