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 #include <linux/percpu.h>
  21
  22 #include "trace_dynevent.h"
  23 #include "trace_probe.h"
  24 #include "trace_probe_tmpl.h"
  25
  26 #define UPROBE_EVENT_SYSTEM     "uprobes"
  27
  28 struct uprobe_trace_entry_head {
  29         struct trace_entry      ent;
  30         unsigned long           vaddr[];
  31 };
  32
  33 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  34         (sizeof(struct uprobe_trace_entry_head) +       \
  35          sizeof(unsigned long) * (is_return ? 2 : 1))
  36
  37 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  38         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  39
  40 static int trace_uprobe_create(const char *raw_command);
  41 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  42 static int trace_uprobe_release(struct dyn_event *ev);
  43 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  44 static bool trace_uprobe_match(const char *system, const char *event,
  45                         int argc, const char **argv, struct dyn_event *ev);
  46
  47 static struct dyn_event_operations trace_uprobe_ops = {
  48         .create = trace_uprobe_create,
  49         .show = trace_uprobe_show,
  50         .is_busy = trace_uprobe_is_busy,
  51         .free = trace_uprobe_release,
  52         .match = trace_uprobe_match,
  53 };
  54
  55 /*
  56  * uprobe event core functions
  57  */
  58 struct trace_uprobe {
  59         struct dyn_event                devent;
  60         struct uprobe_consumer          consumer;
  61         struct path                     path;
  62         char                            *filename;
  63         struct uprobe                   *uprobe;
  64         unsigned long                   offset;
  65         unsigned long                   ref_ctr_offset;
  66         unsigned long __percpu          *nhits;
  67         struct trace_probe              tp;
  68 };
  69
  70 static bool is_trace_uprobe(struct dyn_event *ev)
  71 {
  72         return ev->ops == &trace_uprobe_ops;
  73 }
  74
  75 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  76 {
  77         return container_of(ev, struct trace_uprobe, devent);
  78 }
  79
  80 /**
  81  * for_each_trace_uprobe - iterate over the trace_uprobe list
  82  * @pos:        the struct trace_uprobe * for each entry
  83  * @dpos:       the struct dyn_event * to use as a loop cursor
  84  */
  85 #define for_each_trace_uprobe(pos, dpos)        \
  86         for_each_dyn_event(dpos)                \
  87                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  88
  89 static int register_uprobe_event(struct trace_uprobe *tu);
  90 static int unregister_uprobe_event(struct trace_uprobe *tu);
  91
  92 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs,
  93                              __u64 *data);
  94 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  95                                 unsigned long func, struct pt_regs *regs,
  96                                 __u64 *data);
  97
  98 #ifdef CONFIG_STACK_GROWSUP
  99 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 100 {
 101         return addr - (n * sizeof(long));
 102 }
 103 #else
 104 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 105 {
 106         return addr + (n * sizeof(long));
 107 }
 108 #endif
 109
 110 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 111 {
 112         unsigned long ret;
 113         unsigned long addr = user_stack_pointer(regs);
 114
 115         addr = adjust_stack_addr(addr, n);
 116
 117         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 118                 return 0;
 119
 120         return ret;
 121 }
 122
 123 /*
 124  * Uprobes-specific fetch functions
 125  */
 126 static nokprobe_inline int
 127 probe_mem_read(void *dest, void *src, size_t size)
 128 {
 129         void __user *vaddr = (void __force __user *)src;
 130
 131         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 132 }
 133
 134 static nokprobe_inline int
 135 probe_mem_read_user(void *dest, void *src, size_t size)
 136 {
 137         return probe_mem_read(dest, src, size);
 138 }
 139
 140 /*
 141  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 142  * length and relative data location.
 143  */
 144 static nokprobe_inline int
 145 fetch_store_string(unsigned long addr, void *dest, void *base)
 146 {
 147         long ret;
 148         u32 loc = *(u32 *)dest;
 149         int maxlen  = get_loc_len(loc);
 150         u8 *dst = get_loc_data(dest, base);
 151         void __user *src = (void __force __user *) addr;
 152
 153         if (unlikely(!maxlen))
 154                 return -ENOMEM;
 155
 156         if (addr == FETCH_TOKEN_COMM)
 157                 ret = strscpy(dst, current->comm, maxlen);
 158         else
 159                 ret = strncpy_from_user(dst, src, maxlen);
 160         if (ret >= 0) {
 161                 if (ret == maxlen)
 162                         dst[ret - 1] = '\0';
 163                 else
 164                         /*
 165                          * Include the terminating null byte. In this case it
 166                          * was copied by strncpy_from_user but not accounted
 167                          * for in ret.
 168                          */
 169                         ret++;
 170                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 171         } else
 172                 *(u32 *)dest = make_data_loc(0, (void *)dst - base);
 173
 174         return ret;
 175 }
 176
 177 static nokprobe_inline int
 178 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 179 {
 180         return fetch_store_string(addr, dest, base);
 181 }
 182
 183 /* Return the length of string -- including null terminal byte */
 184 static nokprobe_inline int
 185 fetch_store_strlen(unsigned long addr)
 186 {
 187         int len;
 188         void __user *vaddr = (void __force __user *) addr;
 189
 190         if (addr == FETCH_TOKEN_COMM)
 191                 len = strlen(current->comm) + 1;
 192         else
 193                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 194
 195         return (len > MAX_STRING_SIZE) ? 0 : len;
 196 }
 197
 198 static nokprobe_inline int
 199 fetch_store_strlen_user(unsigned long addr)
 200 {
 201         return fetch_store_strlen(addr);
 202 }
 203
 204 static unsigned long translate_user_vaddr(unsigned long file_offset)
 205 {
 206         unsigned long base_addr;
 207         struct uprobe_dispatch_data *udd;
 208
 209         udd = (void *) current->utask->vaddr;
 210
 211         base_addr = udd->bp_addr - udd->tu->offset;
 212         return base_addr + file_offset;
 213 }
 214
 215 /* Note that we don't verify it, since the code does not come from user space */
 216 static int
 217 process_fetch_insn(struct fetch_insn *code, void *rec, void *edata,
 218                    void *dest, void *base)
 219 {
 220         struct pt_regs *regs = rec;
 221         unsigned long val;
 222         int ret;
 223
 224         /* 1st stage: get value from context */
 225         switch (code->op) {
 226         case FETCH_OP_REG:
 227                 val = regs_get_register(regs, code->param);
 228                 break;
 229         case FETCH_OP_STACK:
 230                 val = get_user_stack_nth(regs, code->param);
 231                 break;
 232         case FETCH_OP_STACKP:
 233                 val = user_stack_pointer(regs);
 234                 break;
 235         case FETCH_OP_RETVAL:
 236                 val = regs_return_value(regs);
 237                 break;
 238         case FETCH_OP_COMM:
 239                 val = FETCH_TOKEN_COMM;
 240                 break;
 241         case FETCH_OP_FOFFS:
 242                 val = translate_user_vaddr(code->immediate);
 243                 break;
 244         default:
 245                 ret = process_common_fetch_insn(code, &val);
 246                 if (ret < 0)
 247                         return ret;
 248         }
 249         code++;
 250
 251         return process_fetch_insn_bottom(code, val, dest, base);
 252 }
 253 NOKPROBE_SYMBOL(process_fetch_insn)
 254
 255 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 256 {
 257         rwlock_init(&filter->rwlock);
 258         filter->nr_systemwide = 0;
 259         INIT_LIST_HEAD(&filter->perf_events);
 260 }
 261
 262 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 263 {
 264         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 265 }
 266
 267 static inline bool is_ret_probe(struct trace_uprobe *tu)
 268 {
 269         return tu->consumer.ret_handler != NULL;
 270 }
 271
 272 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 273 {
 274         struct trace_uprobe *tu = to_trace_uprobe(ev);
 275
 276         return trace_probe_is_enabled(&tu->tp);
 277 }
 278
 279 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 280                                             int argc, const char **argv)
 281 {
 282         char buf[MAX_ARGSTR_LEN + 1];
 283         int len;
 284
 285         if (!argc)
 286                 return true;
 287
 288         len = strlen(tu->filename);
 289         if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 290                 return false;
 291
 292         if (tu->ref_ctr_offset == 0)
 293                 snprintf(buf, sizeof(buf), "0x%0*lx",
 294                                 (int)(sizeof(void *) * 2), tu->offset);
 295         else
 296                 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 297                                 (int)(sizeof(void *) * 2), tu->offset,
 298                                 tu->ref_ctr_offset);
 299         if (strcmp(buf, &argv[0][len + 1]))
 300                 return false;
 301
 302         argc--; argv++;
 303
 304         return trace_probe_match_command_args(&tu->tp, argc, argv);
 305 }
 306
 307 static bool trace_uprobe_match(const char *system, const char *event,
 308                         int argc, const char **argv, struct dyn_event *ev)
 309 {
 310         struct trace_uprobe *tu = to_trace_uprobe(ev);
 311
 312         return (event[0] == '\0' ||
 313                 strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 314            (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 315            trace_uprobe_match_command_head(tu, argc, argv);
 316 }
 317
 318 static nokprobe_inline struct trace_uprobe *
 319 trace_uprobe_primary_from_call(struct trace_event_call *call)
 320 {
 321         struct trace_probe *tp;
 322
 323         tp = trace_probe_primary_from_call(call);
 324         if (WARN_ON_ONCE(!tp))
 325                 return NULL;
 326
 327         return container_of(tp, struct trace_uprobe, tp);
 328 }
 329
 330 /*
 331  * Allocate new trace_uprobe and initialize it (including uprobes).
 332  */
 333 static struct trace_uprobe *
 334 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 335 {
 336         struct trace_uprobe *tu;
 337         int ret;
 338
 339         tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 340         if (!tu)
 341                 return ERR_PTR(-ENOMEM);
 342
 343         tu->nhits = alloc_percpu(unsigned long);
 344         if (!tu->nhits) {
 345                 ret = -ENOMEM;
 346                 goto error;
 347         }
 348
 349         ret = trace_probe_init(&tu->tp, event, group, true, nargs);
 350         if (ret < 0)
 351                 goto error;
 352
 353         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 354         tu->consumer.handler = uprobe_dispatcher;
 355         if (is_ret)
 356                 tu->consumer.ret_handler = uretprobe_dispatcher;
 357         init_trace_uprobe_filter(tu->tp.event->filter);
 358         return tu;
 359
 360 error:
 361         free_percpu(tu->nhits);
 362         kfree(tu);
 363
 364         return ERR_PTR(ret);
 365 }
 366
 367 static void free_trace_uprobe(struct trace_uprobe *tu)
 368 {
 369         if (!tu)
 370                 return;
 371
 372         path_put(&tu->path);
 373         trace_probe_cleanup(&tu->tp);
 374         kfree(tu->filename);
 375         free_percpu(tu->nhits);
 376         kfree(tu);
 377 }
 378
 379 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 380 {
 381         struct dyn_event *pos;
 382         struct trace_uprobe *tu;
 383
 384         for_each_trace_uprobe(tu, pos)
 385                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 386                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 387                         return tu;
 388
 389         return NULL;
 390 }
 391
 392 /* Unregister a trace_uprobe and probe_event */
 393 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 394 {
 395         int ret;
 396
 397         if (trace_probe_has_sibling(&tu->tp))
 398                 goto unreg;
 399
 400         /* If there's a reference to the dynamic event */
 401         if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 402                 return -EBUSY;
 403
 404         ret = unregister_uprobe_event(tu);
 405         if (ret)
 406                 return ret;
 407
 408 unreg:
 409         dyn_event_remove(&tu->devent);
 410         trace_probe_unlink(&tu->tp);
 411         free_trace_uprobe(tu);
 412         return 0;
 413 }
 414
 415 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 416                                          struct trace_uprobe *comp)
 417 {
 418         struct trace_probe_event *tpe = orig->tp.event;
 419         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 420         int i;
 421
 422         list_for_each_entry(orig, &tpe->probes, tp.list) {
 423                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 424                     comp->offset != orig->offset)
 425                         continue;
 426
 427                 /*
 428                  * trace_probe_compare_arg_type() ensured that nr_args and
 429                  * each argument name and type are same. Let's compare comm.
 430                  */
 431                 for (i = 0; i < orig->tp.nr_args; i++) {
 432                         if (strcmp(orig->tp.args[i].comm,
 433                                    comp->tp.args[i].comm))
 434                                 break;
 435                 }
 436
 437                 if (i == orig->tp.nr_args)
 438                         return true;
 439         }
 440
 441         return false;
 442 }
 443
 444 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 445 {
 446         int ret;
 447
 448         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 449         if (ret) {
 450                 /* Note that argument starts index = 2 */
 451                 trace_probe_log_set_index(ret + 1);
 452                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 453                 return -EEXIST;
 454         }
 455         if (trace_uprobe_has_same_uprobe(to, tu)) {
 456                 trace_probe_log_set_index(0);
 457                 trace_probe_log_err(0, SAME_PROBE);
 458                 return -EEXIST;
 459         }
 460
 461         /* Append to existing event */
 462         ret = trace_probe_append(&tu->tp, &to->tp);
 463         if (!ret)
 464                 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 465
 466         return ret;
 467 }
 468
 469 /*
 470  * Uprobe with multiple reference counter is not allowed. i.e.
 471  * If inode and offset matches, reference counter offset *must*
 472  * match as well. Though, there is one exception: If user is
 473  * replacing old trace_uprobe with new one(same group/event),
 474  * then we allow same uprobe with new reference counter as far
 475  * as the new one does not conflict with any other existing
 476  * ones.
 477  */
 478 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 479 {
 480         struct dyn_event *pos;
 481         struct trace_uprobe *tmp;
 482         struct inode *new_inode = d_real_inode(new->path.dentry);
 483
 484         for_each_trace_uprobe(tmp, pos) {
 485                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 486                     new->offset == tmp->offset &&
 487                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 488                         pr_warn("Reference counter offset mismatch.");
 489                         return -EINVAL;
 490                 }
 491         }
 492         return 0;
 493 }
 494
 495 /* Register a trace_uprobe and probe_event */
 496 static int register_trace_uprobe(struct trace_uprobe *tu)
 497 {
 498         struct trace_uprobe *old_tu;
 499         int ret;
 500
 501         guard(mutex)(&event_mutex);
 502
 503         ret = validate_ref_ctr_offset(tu);
 504         if (ret)
 505                 return ret;
 506
 507         /* register as an event */
 508         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 509                                   trace_probe_group_name(&tu->tp));
 510         if (old_tu) {
 511                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 512                         trace_probe_log_set_index(0);
 513                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 514                         return -EEXIST;
 515                 }
 516                 return append_trace_uprobe(tu, old_tu);
 517         }
 518
 519         ret = register_uprobe_event(tu);
 520         if (ret) {
 521                 if (ret == -EEXIST) {
 522                         trace_probe_log_set_index(0);
 523                         trace_probe_log_err(0, EVENT_EXIST);
 524                 } else
 525                         pr_warn("Failed to register probe event(%d)\n", ret);
 526                 return ret;
 527         }
 528
 529         dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 530
 531         return ret;
 532 }
 533
 534 /*
 535  * Argument syntax:
 536  *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 537  */
 538 static int __trace_uprobe_create(int argc, const char **argv)
 539 {
 540         struct trace_uprobe *tu;
 541         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 542         char *arg, *filename, *rctr, *rctr_end, *tmp;
 543         char buf[MAX_EVENT_NAME_LEN];
 544         char gbuf[MAX_EVENT_NAME_LEN];
 545         enum probe_print_type ptype;
 546         struct path path;
 547         unsigned long offset, ref_ctr_offset;
 548         bool is_return = false;
 549         int i, ret;
 550
 551         ref_ctr_offset = 0;
 552
 553         switch (argv[0][0]) {
 554         case 'r':
 555                 is_return = true;
 556                 break;
 557         case 'p':
 558                 break;
 559         default:
 560                 return -ECANCELED;
 561         }
 562
 563         if (argc < 2)
 564                 return -ECANCELED;
 565
 566         trace_probe_log_init("trace_uprobe", argc, argv);
 567
 568         if (argc - 2 > MAX_TRACE_ARGS) {
 569                 trace_probe_log_set_index(2);
 570                 trace_probe_log_err(0, TOO_MANY_ARGS);
 571                 return -E2BIG;
 572         }
 573
 574         if (argv[0][1] == ':')
 575                 event = &argv[0][2];
 576
 577         if (!strchr(argv[1], '/'))
 578                 return -ECANCELED;
 579
 580         filename = kstrdup(argv[1], GFP_KERNEL);
 581         if (!filename)
 582                 return -ENOMEM;
 583
 584         /* Find the last occurrence, in case the path contains ':' too. */
 585         arg = strrchr(filename, ':');
 586         if (!arg || !isdigit(arg[1])) {
 587                 kfree(filename);
 588                 return -ECANCELED;
 589         }
 590
 591         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 592
 593         *arg++ = '\0';
 594         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 595         if (ret) {
 596                 trace_probe_log_err(0, FILE_NOT_FOUND);
 597                 kfree(filename);
 598                 trace_probe_log_clear();
 599                 return ret;
 600         }
 601         if (!d_is_reg(path.dentry)) {
 602                 trace_probe_log_err(0, NO_REGULAR_FILE);
 603                 ret = -EINVAL;
 604                 goto fail_address_parse;
 605         }
 606
 607         /* Parse reference counter offset if specified. */
 608         rctr = strchr(arg, '(');
 609         if (rctr) {
 610                 rctr_end = strchr(rctr, ')');
 611                 if (!rctr_end) {
 612                         ret = -EINVAL;
 613                         rctr_end = rctr + strlen(rctr);
 614                         trace_probe_log_err(rctr_end - filename,
 615                                             REFCNT_OPEN_BRACE);
 616                         goto fail_address_parse;
 617                 } else if (rctr_end[1] != '\0') {
 618                         ret = -EINVAL;
 619                         trace_probe_log_err(rctr_end + 1 - filename,
 620                                             BAD_REFCNT_SUFFIX);
 621                         goto fail_address_parse;
 622                 }
 623
 624                 *rctr++ = '\0';
 625                 *rctr_end = '\0';
 626                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 627                 if (ret) {
 628                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 629                         goto fail_address_parse;
 630                 }
 631         }
 632
 633         /* Check if there is %return suffix */
 634         tmp = strchr(arg, '%');
 635         if (tmp) {
 636                 if (!strcmp(tmp, "%return")) {
 637                         *tmp = '\0';
 638                         is_return = true;
 639                 } else {
 640                         trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 641                         ret = -EINVAL;
 642                         goto fail_address_parse;
 643                 }
 644         }
 645
 646         /* Parse uprobe offset. */
 647         ret = kstrtoul(arg, 0, &offset);
 648         if (ret) {
 649                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 650                 goto fail_address_parse;
 651         }
 652
 653         /* setup a probe */
 654         trace_probe_log_set_index(0);
 655         if (event) {
 656                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
 657                                                   event - argv[0]);
 658                 if (ret)
 659                         goto fail_address_parse;
 660         }
 661
 662         if (!event) {
 663                 char *tail;
 664                 char *ptr;
 665
 666                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 667                 if (!tail) {
 668                         ret = -ENOMEM;
 669                         goto fail_address_parse;
 670                 }
 671
 672                 ptr = strpbrk(tail, ".-_");
 673                 if (ptr)
 674                         *ptr = '\0';
 675
 676                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 677                 event = buf;
 678                 kfree(tail);
 679         }
 680
 681         argc -= 2;
 682         argv += 2;
 683
 684         tu = alloc_trace_uprobe(group, event, argc, is_return);
 685         if (IS_ERR(tu)) {
 686                 ret = PTR_ERR(tu);
 687                 /* This must return -ENOMEM otherwise there is a bug */
 688                 WARN_ON_ONCE(ret != -ENOMEM);
 689                 goto fail_address_parse;
 690         }
 691         tu->offset = offset;
 692         tu->ref_ctr_offset = ref_ctr_offset;
 693         tu->path = path;
 694         tu->filename = filename;
 695
 696         /* parse arguments */
 697         for (i = 0; i < argc; i++) {
 698                 struct traceprobe_parse_context ctx = {
 699                         .flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
 700                 };
 701
 702                 trace_probe_log_set_index(i + 2);
 703                 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
 704                 traceprobe_finish_parse(&ctx);
 705                 if (ret)
 706                         goto error;
 707         }
 708
 709         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 710         ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 711         if (ret < 0)
 712                 goto error;
 713
 714         ret = register_trace_uprobe(tu);
 715         if (!ret)
 716                 goto out;
 717
 718 error:
 719         free_trace_uprobe(tu);
 720 out:
 721         trace_probe_log_clear();
 722         return ret;
 723
 724 fail_address_parse:
 725         trace_probe_log_clear();
 726         path_put(&path);
 727         kfree(filename);
 728
 729         return ret;
 730 }
 731
 732 int trace_uprobe_create(const char *raw_command)
 733 {
 734         return trace_probe_create(raw_command, __trace_uprobe_create);
 735 }
 736
 737 static int create_or_delete_trace_uprobe(const char *raw_command)
 738 {
 739         int ret;
 740
 741         if (raw_command[0] == '-')
 742                 return dyn_event_release(raw_command, &trace_uprobe_ops);
 743
 744         ret = dyn_event_create(raw_command, &trace_uprobe_ops);
 745         return ret == -ECANCELED ? -EINVAL : ret;
 746 }
 747
 748 static int trace_uprobe_release(struct dyn_event *ev)
 749 {
 750         struct trace_uprobe *tu = to_trace_uprobe(ev);
 751
 752         return unregister_trace_uprobe(tu);
 753 }
 754
 755 /* Probes listing interfaces */
 756 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 757 {
 758         struct trace_uprobe *tu = to_trace_uprobe(ev);
 759         char c = is_ret_probe(tu) ? 'r' : 'p';
 760         int i;
 761
 762         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 763                         trace_probe_name(&tu->tp), tu->filename,
 764                         (int)(sizeof(void *) * 2), tu->offset);
 765
 766         if (tu->ref_ctr_offset)
 767                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 768
 769         for (i = 0; i < tu->tp.nr_args; i++)
 770                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 771
 772         seq_putc(m, '\n');
 773         return 0;
 774 }
 775
 776 static int probes_seq_show(struct seq_file *m, void *v)
 777 {
 778         struct dyn_event *ev = v;
 779
 780         if (!is_trace_uprobe(ev))
 781                 return 0;
 782
 783         return trace_uprobe_show(m, ev);
 784 }
 785
 786 static const struct seq_operations probes_seq_op = {
 787         .start  = dyn_event_seq_start,
 788         .next   = dyn_event_seq_next,
 789         .stop   = dyn_event_seq_stop,
 790         .show   = probes_seq_show
 791 };
 792
 793 static int probes_open(struct inode *inode, struct file *file)
 794 {
 795         int ret;
 796
 797         ret = security_locked_down(LOCKDOWN_TRACEFS);
 798         if (ret)
 799                 return ret;
 800
 801         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 802                 ret = dyn_events_release_all(&trace_uprobe_ops);
 803                 if (ret)
 804                         return ret;
 805         }
 806
 807         return seq_open(file, &probes_seq_op);
 808 }
 809
 810 static ssize_t probes_write(struct file *file, const char __user *buffer,
 811                             size_t count, loff_t *ppos)
 812 {
 813         return trace_parse_run_command(file, buffer, count, ppos,
 814                                         create_or_delete_trace_uprobe);
 815 }
 816
 817 static const struct file_operations uprobe_events_ops = {
 818         .owner          = THIS_MODULE,
 819         .open           = probes_open,
 820         .read           = seq_read,
 821         .llseek         = seq_lseek,
 822         .release        = seq_release,
 823         .write          = probes_write,
 824 };
 825
 826 /* Probes profiling interfaces */
 827 static int probes_profile_seq_show(struct seq_file *m, void *v)
 828 {
 829         struct dyn_event *ev = v;
 830         struct trace_uprobe *tu;
 831         unsigned long nhits;
 832         int cpu;
 833
 834         if (!is_trace_uprobe(ev))
 835                 return 0;
 836
 837         tu = to_trace_uprobe(ev);
 838
 839         nhits = 0;
 840         for_each_possible_cpu(cpu) {
 841                 nhits += per_cpu(*tu->nhits, cpu);
 842         }
 843
 844         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 845                    trace_probe_name(&tu->tp), nhits);
 846         return 0;
 847 }
 848
 849 static const struct seq_operations profile_seq_op = {
 850         .start  = dyn_event_seq_start,
 851         .next   = dyn_event_seq_next,
 852         .stop   = dyn_event_seq_stop,
 853         .show   = probes_profile_seq_show
 854 };
 855
 856 static int profile_open(struct inode *inode, struct file *file)
 857 {
 858         int ret;
 859
 860         ret = security_locked_down(LOCKDOWN_TRACEFS);
 861         if (ret)
 862                 return ret;
 863
 864         return seq_open(file, &profile_seq_op);
 865 }
 866
 867 static const struct file_operations uprobe_profile_ops = {
 868         .owner          = THIS_MODULE,
 869         .open           = profile_open,
 870         .read           = seq_read,
 871         .llseek         = seq_lseek,
 872         .release        = seq_release,
 873 };
 874
 875 struct uprobe_cpu_buffer {
 876         struct mutex mutex;
 877         void *buf;
 878         int dsize;
 879 };
 880 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 881 static int uprobe_buffer_refcnt;
 882 #define MAX_UCB_BUFFER_SIZE PAGE_SIZE
 883
 884 static int uprobe_buffer_init(void)
 885 {
 886         int cpu, err_cpu;
 887
 888         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 889         if (uprobe_cpu_buffer == NULL)
 890                 return -ENOMEM;
 891
 892         for_each_possible_cpu(cpu) {
 893                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 894                                                   GFP_KERNEL, 0);
 895                 if (p == NULL) {
 896                         err_cpu = cpu;
 897                         goto err;
 898                 }
 899                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 900                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 901         }
 902
 903         return 0;
 904
 905 err:
 906         for_each_possible_cpu(cpu) {
 907                 if (cpu == err_cpu)
 908                         break;
 909                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 910         }
 911
 912         free_percpu(uprobe_cpu_buffer);
 913         return -ENOMEM;
 914 }
 915
 916 static int uprobe_buffer_enable(void)
 917 {
 918         int ret = 0;
 919
 920         BUG_ON(!mutex_is_locked(&event_mutex));
 921
 922         if (uprobe_buffer_refcnt++ == 0) {
 923                 ret = uprobe_buffer_init();
 924                 if (ret < 0)
 925                         uprobe_buffer_refcnt--;
 926         }
 927
 928         return ret;
 929 }
 930
 931 static void uprobe_buffer_disable(void)
 932 {
 933         int cpu;
 934
 935         BUG_ON(!mutex_is_locked(&event_mutex));
 936
 937         if (--uprobe_buffer_refcnt == 0) {
 938                 for_each_possible_cpu(cpu)
 939                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 940                                                              cpu)->buf);
 941
 942                 free_percpu(uprobe_cpu_buffer);
 943                 uprobe_cpu_buffer = NULL;
 944         }
 945 }
 946
 947 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 948 {
 949         struct uprobe_cpu_buffer *ucb;
 950         int cpu;
 951
 952         cpu = raw_smp_processor_id();
 953         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 954
 955         /*
 956          * Use per-cpu buffers for fastest access, but we might migrate
 957          * so the mutex makes sure we have sole access to it.
 958          */
 959         mutex_lock(&ucb->mutex);
 960
 961         return ucb;
 962 }
 963
 964 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 965 {
 966         if (!ucb)
 967                 return;
 968         mutex_unlock(&ucb->mutex);
 969 }
 970
 971 static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu,
 972                                                        struct pt_regs *regs,
 973                                                        struct uprobe_cpu_buffer **ucbp)
 974 {
 975         struct uprobe_cpu_buffer *ucb;
 976         int dsize, esize;
 977
 978         if (*ucbp)
 979                 return *ucbp;
 980
 981         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 982         dsize = __get_data_size(&tu->tp, regs, NULL);
 983
 984         ucb = uprobe_buffer_get();
 985         ucb->dsize = tu->tp.size + dsize;
 986
 987         if (WARN_ON_ONCE(ucb->dsize > MAX_UCB_BUFFER_SIZE)) {
 988                 ucb->dsize = MAX_UCB_BUFFER_SIZE;
 989                 dsize = MAX_UCB_BUFFER_SIZE - tu->tp.size;
 990         }
 991
 992         store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize);
 993
 994         *ucbp = ucb;
 995         return ucb;
 996 }
 997
 998 static void __uprobe_trace_func(struct trace_uprobe *tu,
 999                                 unsigned long func, struct pt_regs *regs,
1000                                 struct uprobe_cpu_buffer *ucb,
1001                                 struct trace_event_file *trace_file)
1002 {
1003         struct uprobe_trace_entry_head *entry;
1004         struct trace_event_buffer fbuffer;
1005         void *data;
1006         int size, esize;
1007         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1008
1009         WARN_ON(call != trace_file->event_call);
1010
1011         if (trace_trigger_soft_disabled(trace_file))
1012                 return;
1013
1014         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1015         size = esize + ucb->dsize;
1016         entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
1017         if (!entry)
1018                 return;
1019
1020         if (is_ret_probe(tu)) {
1021                 entry->vaddr[0] = func;
1022                 entry->vaddr[1] = instruction_pointer(regs);
1023                 data = DATAOF_TRACE_ENTRY(entry, true);
1024         } else {
1025                 entry->vaddr[0] = instruction_pointer(regs);
1026                 data = DATAOF_TRACE_ENTRY(entry, false);
1027         }
1028
1029         memcpy(data, ucb->buf, ucb->dsize);
1030
1031         trace_event_buffer_commit(&fbuffer);
1032 }
1033
1034 /* uprobe handler */
1035 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
1036                              struct uprobe_cpu_buffer **ucbp)
1037 {
1038         struct event_file_link *link;
1039         struct uprobe_cpu_buffer *ucb;
1040
1041         if (is_ret_probe(tu))
1042                 return 0;
1043
1044         ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1045
1046         rcu_read_lock();
1047         trace_probe_for_each_link_rcu(link, &tu->tp)
1048                 __uprobe_trace_func(tu, 0, regs, ucb, link->file);
1049         rcu_read_unlock();
1050
1051         return 0;
1052 }
1053
1054 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1055                                  struct pt_regs *regs,
1056                                  struct uprobe_cpu_buffer **ucbp)
1057 {
1058         struct event_file_link *link;
1059         struct uprobe_cpu_buffer *ucb;
1060
1061         ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1062
1063         rcu_read_lock();
1064         trace_probe_for_each_link_rcu(link, &tu->tp)
1065                 __uprobe_trace_func(tu, func, regs, ucb, link->file);
1066         rcu_read_unlock();
1067 }
1068
1069 /* Event entry printers */
1070 static enum print_line_t
1071 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1072 {
1073         struct uprobe_trace_entry_head *entry;
1074         struct trace_seq *s = &iter->seq;
1075         struct trace_uprobe *tu;
1076         u8 *data;
1077
1078         entry = (struct uprobe_trace_entry_head *)iter->ent;
1079         tu = trace_uprobe_primary_from_call(
1080                 container_of(event, struct trace_event_call, event));
1081         if (unlikely(!tu))
1082                 goto out;
1083
1084         if (is_ret_probe(tu)) {
1085                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1086                                  trace_probe_name(&tu->tp),
1087                                  entry->vaddr[1], entry->vaddr[0]);
1088                 data = DATAOF_TRACE_ENTRY(entry, true);
1089         } else {
1090                 trace_seq_printf(s, "%s: (0x%lx)",
1091                                  trace_probe_name(&tu->tp),
1092                                  entry->vaddr[0]);
1093                 data = DATAOF_TRACE_ENTRY(entry, false);
1094         }
1095
1096         if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1097                 goto out;
1098
1099         trace_seq_putc(s, '\n');
1100
1101  out:
1102         return trace_handle_return(s);
1103 }
1104
1105 typedef bool (*filter_func_t)(struct uprobe_consumer *self, struct mm_struct *mm);
1106
1107 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1108 {
1109         struct inode *inode = d_real_inode(tu->path.dentry);
1110         struct uprobe *uprobe;
1111
1112         tu->consumer.filter = filter;
1113         uprobe = uprobe_register(inode, tu->offset, tu->ref_ctr_offset, &tu->consumer);
1114         if (IS_ERR(uprobe))
1115                 return PTR_ERR(uprobe);
1116
1117         tu->uprobe = uprobe;
1118         return 0;
1119 }
1120
1121 static void __probe_event_disable(struct trace_probe *tp)
1122 {
1123         struct trace_uprobe *tu;
1124         bool sync = false;
1125
1126         tu = container_of(tp, struct trace_uprobe, tp);
1127         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1128
1129         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1130                 if (!tu->uprobe)
1131                         continue;
1132
1133                 uprobe_unregister_nosync(tu->uprobe, &tu->consumer);
1134                 sync = true;
1135                 tu->uprobe = NULL;
1136         }
1137         if (sync)
1138                 uprobe_unregister_sync();
1139 }
1140
1141 static int probe_event_enable(struct trace_event_call *call,
1142                         struct trace_event_file *file, filter_func_t filter)
1143 {
1144         struct trace_probe *tp;
1145         struct trace_uprobe *tu;
1146         bool enabled;
1147         int ret;
1148
1149         tp = trace_probe_primary_from_call(call);
1150         if (WARN_ON_ONCE(!tp))
1151                 return -ENODEV;
1152         enabled = trace_probe_is_enabled(tp);
1153
1154         /* This may also change "enabled" state */
1155         if (file) {
1156                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1157                         return -EINTR;
1158
1159                 ret = trace_probe_add_file(tp, file);
1160                 if (ret < 0)
1161                         return ret;
1162         } else {
1163                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1164                         return -EINTR;
1165
1166                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1167         }
1168
1169         tu = container_of(tp, struct trace_uprobe, tp);
1170         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1171
1172         if (enabled)
1173                 return 0;
1174
1175         ret = uprobe_buffer_enable();
1176         if (ret)
1177                 goto err_flags;
1178
1179         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1180                 ret = trace_uprobe_enable(tu, filter);
1181                 if (ret) {
1182                         __probe_event_disable(tp);
1183                         goto err_buffer;
1184                 }
1185         }
1186
1187         return 0;
1188
1189  err_buffer:
1190         uprobe_buffer_disable();
1191
1192  err_flags:
1193         if (file)
1194                 trace_probe_remove_file(tp, file);
1195         else
1196                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1197
1198         return ret;
1199 }
1200
1201 static void probe_event_disable(struct trace_event_call *call,
1202                                 struct trace_event_file *file)
1203 {
1204         struct trace_probe *tp;
1205
1206         tp = trace_probe_primary_from_call(call);
1207         if (WARN_ON_ONCE(!tp))
1208                 return;
1209
1210         if (!trace_probe_is_enabled(tp))
1211                 return;
1212
1213         if (file) {
1214                 if (trace_probe_remove_file(tp, file) < 0)
1215                         return;
1216
1217                 if (trace_probe_is_enabled(tp))
1218                         return;
1219         } else
1220                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1221
1222         __probe_event_disable(tp);
1223         uprobe_buffer_disable();
1224 }
1225
1226 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1227 {
1228         int ret, size;
1229         struct uprobe_trace_entry_head field;
1230         struct trace_uprobe *tu;
1231
1232         tu = trace_uprobe_primary_from_call(event_call);
1233         if (unlikely(!tu))
1234                 return -ENODEV;
1235
1236         if (is_ret_probe(tu)) {
1237                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1238                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1239                 size = SIZEOF_TRACE_ENTRY(true);
1240         } else {
1241                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1242                 size = SIZEOF_TRACE_ENTRY(false);
1243         }
1244
1245         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1246 }
1247
1248 #ifdef CONFIG_PERF_EVENTS
1249 static bool
1250 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1251 {
1252         struct perf_event *event;
1253
1254         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1255                 if (event->hw.target->mm == mm)
1256                         return true;
1257         }
1258
1259         return false;
1260 }
1261
1262 static inline bool
1263 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1264                           struct perf_event *event)
1265 {
1266         return __uprobe_perf_filter(filter, event->hw.target->mm);
1267 }
1268
1269 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1270                                        struct perf_event *event)
1271 {
1272         bool done;
1273
1274         write_lock(&filter->rwlock);
1275         if (event->hw.target) {
1276                 list_del(&event->hw.tp_list);
1277                 done = filter->nr_systemwide ||
1278                         (event->hw.target->flags & PF_EXITING) ||
1279                         trace_uprobe_filter_event(filter, event);
1280         } else {
1281                 filter->nr_systemwide--;
1282                 done = filter->nr_systemwide;
1283         }
1284         write_unlock(&filter->rwlock);
1285
1286         return done;
1287 }
1288
1289 /* This returns true if the filter always covers target mm */
1290 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1291                                     struct perf_event *event)
1292 {
1293         bool done;
1294
1295         write_lock(&filter->rwlock);
1296         if (event->hw.target) {
1297                 /*
1298                  * event->parent != NULL means copy_process(), we can avoid
1299                  * uprobe_apply(). current->mm must be probed and we can rely
1300                  * on dup_mmap() which preserves the already installed bp's.
1301                  *
1302                  * attr.enable_on_exec means that exec/mmap will install the
1303                  * breakpoints we need.
1304                  */
1305                 done = filter->nr_systemwide ||
1306                         event->parent || event->attr.enable_on_exec ||
1307                         trace_uprobe_filter_event(filter, event);
1308                 list_add(&event->hw.tp_list, &filter->perf_events);
1309         } else {
1310                 done = filter->nr_systemwide;
1311                 filter->nr_systemwide++;
1312         }
1313         write_unlock(&filter->rwlock);
1314
1315         return done;
1316 }
1317
1318 static int uprobe_perf_close(struct trace_event_call *call,
1319                              struct perf_event *event)
1320 {
1321         struct trace_probe *tp;
1322         struct trace_uprobe *tu;
1323         int ret = 0;
1324
1325         tp = trace_probe_primary_from_call(call);
1326         if (WARN_ON_ONCE(!tp))
1327                 return -ENODEV;
1328
1329         tu = container_of(tp, struct trace_uprobe, tp);
1330         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1331                 return 0;
1332
1333         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1334                 ret = uprobe_apply(tu->uprobe, &tu->consumer, false);
1335                 if (ret)
1336                         break;
1337         }
1338
1339         return ret;
1340 }
1341
1342 static int uprobe_perf_open(struct trace_event_call *call,
1343                             struct perf_event *event)
1344 {
1345         struct trace_probe *tp;
1346         struct trace_uprobe *tu;
1347         int err = 0;
1348
1349         tp = trace_probe_primary_from_call(call);
1350         if (WARN_ON_ONCE(!tp))
1351                 return -ENODEV;
1352
1353         tu = container_of(tp, struct trace_uprobe, tp);
1354         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1355                 return 0;
1356
1357         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1358                 err = uprobe_apply(tu->uprobe, &tu->consumer, true);
1359                 if (err) {
1360                         uprobe_perf_close(call, event);
1361                         break;
1362                 }
1363         }
1364
1365         return err;
1366 }
1367
1368 static bool uprobe_perf_filter(struct uprobe_consumer *uc, struct mm_struct *mm)
1369 {
1370         struct trace_uprobe_filter *filter;
1371         struct trace_uprobe *tu;
1372         int ret;
1373
1374         tu = container_of(uc, struct trace_uprobe, consumer);
1375         filter = tu->tp.event->filter;
1376
1377         /*
1378          * speculative short-circuiting check to avoid unnecessarily taking
1379          * filter->rwlock below, if the uprobe has system-wide consumer
1380          */
1381         if (READ_ONCE(filter->nr_systemwide))
1382                 return true;
1383
1384         read_lock(&filter->rwlock);
1385         ret = __uprobe_perf_filter(filter, mm);
1386         read_unlock(&filter->rwlock);
1387
1388         return ret;
1389 }
1390
1391 static void __uprobe_perf_func(struct trace_uprobe *tu,
1392                                unsigned long func, struct pt_regs *regs,
1393                                struct uprobe_cpu_buffer **ucbp)
1394 {
1395         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1396         struct uprobe_trace_entry_head *entry;
1397         struct uprobe_cpu_buffer *ucb;
1398         struct hlist_head *head;
1399         void *data;
1400         int size, esize;
1401         int rctx;
1402
1403 #ifdef CONFIG_BPF_EVENTS
1404         if (bpf_prog_array_valid(call)) {
1405                 const struct bpf_prog_array *array;
1406                 u32 ret;
1407
1408                 rcu_read_lock_trace();
1409                 array = rcu_dereference_check(call->prog_array, rcu_read_lock_trace_held());
1410                 ret = bpf_prog_run_array_uprobe(array, regs, bpf_prog_run);
1411                 rcu_read_unlock_trace();
1412                 if (!ret)
1413                         return;
1414         }
1415 #endif /* CONFIG_BPF_EVENTS */
1416
1417         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1418
1419         ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1420         size = esize + ucb->dsize;
1421         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1422         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1423                 return;
1424
1425         preempt_disable();
1426         head = this_cpu_ptr(call->perf_events);
1427         if (hlist_empty(head))
1428                 goto out;
1429
1430         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1431         if (!entry)
1432                 goto out;
1433
1434         if (is_ret_probe(tu)) {
1435                 entry->vaddr[0] = func;
1436                 entry->vaddr[1] = instruction_pointer(regs);
1437                 data = DATAOF_TRACE_ENTRY(entry, true);
1438         } else {
1439                 entry->vaddr[0] = instruction_pointer(regs);
1440                 data = DATAOF_TRACE_ENTRY(entry, false);
1441         }
1442
1443         memcpy(data, ucb->buf, ucb->dsize);
1444
1445         if (size - esize > ucb->dsize)
1446                 memset(data + ucb->dsize, 0, size - esize - ucb->dsize);
1447
1448         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1449                               head, NULL);
1450  out:
1451         preempt_enable();
1452 }
1453
1454 /* uprobe profile handler */
1455 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1456                             struct uprobe_cpu_buffer **ucbp)
1457 {
1458         if (!uprobe_perf_filter(&tu->consumer, current->mm))
1459                 return UPROBE_HANDLER_REMOVE;
1460
1461         if (!is_ret_probe(tu))
1462                 __uprobe_perf_func(tu, 0, regs, ucbp);
1463         return 0;
1464 }
1465
1466 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1467                                 struct pt_regs *regs,
1468                                 struct uprobe_cpu_buffer **ucbp)
1469 {
1470         __uprobe_perf_func(tu, func, regs, ucbp);
1471 }
1472
1473 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1474                         const char **filename, u64 *probe_offset,
1475                         u64 *probe_addr, bool perf_type_tracepoint)
1476 {
1477         const char *pevent = trace_event_name(event->tp_event);
1478         const char *group = event->tp_event->class->system;
1479         struct trace_uprobe *tu;
1480
1481         if (perf_type_tracepoint)
1482                 tu = find_probe_event(pevent, group);
1483         else
1484                 tu = trace_uprobe_primary_from_call(event->tp_event);
1485         if (!tu)
1486                 return -EINVAL;
1487
1488         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1489                                     : BPF_FD_TYPE_UPROBE;
1490         *filename = tu->filename;
1491         *probe_offset = tu->offset;
1492         *probe_addr = tu->ref_ctr_offset;
1493         return 0;
1494 }
1495 #endif  /* CONFIG_PERF_EVENTS */
1496
1497 static int
1498 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1499                       void *data)
1500 {
1501         struct trace_event_file *file = data;
1502
1503         switch (type) {
1504         case TRACE_REG_REGISTER:
1505                 return probe_event_enable(event, file, NULL);
1506
1507         case TRACE_REG_UNREGISTER:
1508                 probe_event_disable(event, file);
1509                 return 0;
1510
1511 #ifdef CONFIG_PERF_EVENTS
1512         case TRACE_REG_PERF_REGISTER:
1513                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1514
1515         case TRACE_REG_PERF_UNREGISTER:
1516                 probe_event_disable(event, NULL);
1517                 return 0;
1518
1519         case TRACE_REG_PERF_OPEN:
1520                 return uprobe_perf_open(event, data);
1521
1522         case TRACE_REG_PERF_CLOSE:
1523                 return uprobe_perf_close(event, data);
1524
1525 #endif
1526         default:
1527                 return 0;
1528         }
1529 }
1530
1531 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs,
1532                              __u64 *data)
1533 {
1534         struct trace_uprobe *tu;
1535         struct uprobe_dispatch_data udd;
1536         struct uprobe_cpu_buffer *ucb = NULL;
1537         int ret = 0;
1538
1539         tu = container_of(con, struct trace_uprobe, consumer);
1540
1541         this_cpu_inc(*tu->nhits);
1542
1543         udd.tu = tu;
1544         udd.bp_addr = instruction_pointer(regs);
1545
1546         current->utask->vaddr = (unsigned long) &udd;
1547
1548         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1549                 return 0;
1550
1551         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1552                 ret |= uprobe_trace_func(tu, regs, &ucb);
1553
1554 #ifdef CONFIG_PERF_EVENTS
1555         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1556                 ret |= uprobe_perf_func(tu, regs, &ucb);
1557 #endif
1558         uprobe_buffer_put(ucb);
1559         return ret;
1560 }
1561
1562 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1563                                 unsigned long func, struct pt_regs *regs,
1564                                 __u64 *data)
1565 {
1566         struct trace_uprobe *tu;
1567         struct uprobe_dispatch_data udd;
1568         struct uprobe_cpu_buffer *ucb = NULL;
1569
1570         tu = container_of(con, struct trace_uprobe, consumer);
1571
1572         udd.tu = tu;
1573         udd.bp_addr = func;
1574
1575         current->utask->vaddr = (unsigned long) &udd;
1576
1577         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1578                 return 0;
1579
1580         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1581                 uretprobe_trace_func(tu, func, regs, &ucb);
1582
1583 #ifdef CONFIG_PERF_EVENTS
1584         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1585                 uretprobe_perf_func(tu, func, regs, &ucb);
1586 #endif
1587         uprobe_buffer_put(ucb);
1588         return 0;
1589 }
1590
1591 static struct trace_event_functions uprobe_funcs = {
1592         .trace          = print_uprobe_event
1593 };
1594
1595 static struct trace_event_fields uprobe_fields_array[] = {
1596         { .type = TRACE_FUNCTION_TYPE,
1597           .define_fields = uprobe_event_define_fields },
1598         {}
1599 };
1600
1601 static inline void init_trace_event_call(struct trace_uprobe *tu)
1602 {
1603         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1604         call->event.funcs = &uprobe_funcs;
1605         call->class->fields_array = uprobe_fields_array;
1606
1607         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1608         call->class->reg = trace_uprobe_register;
1609 }
1610
1611 static int register_uprobe_event(struct trace_uprobe *tu)
1612 {
1613         init_trace_event_call(tu);
1614
1615         return trace_probe_register_event_call(&tu->tp);
1616 }
1617
1618 static int unregister_uprobe_event(struct trace_uprobe *tu)
1619 {
1620         return trace_probe_unregister_event_call(&tu->tp);
1621 }
1622
1623 #ifdef CONFIG_PERF_EVENTS
1624 struct trace_event_call *
1625 create_local_trace_uprobe(char *name, unsigned long offs,
1626                           unsigned long ref_ctr_offset, bool is_return)
1627 {
1628         enum probe_print_type ptype;
1629         struct trace_uprobe *tu;
1630         struct path path;
1631         int ret;
1632
1633         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1634         if (ret)
1635                 return ERR_PTR(ret);
1636
1637         if (!d_is_reg(path.dentry)) {
1638                 path_put(&path);
1639                 return ERR_PTR(-EINVAL);
1640         }
1641
1642         /*
1643          * local trace_kprobes are not added to dyn_event, so they are never
1644          * searched in find_trace_kprobe(). Therefore, there is no concern of
1645          * duplicated name "DUMMY_EVENT" here.
1646          */
1647         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1648                                 is_return);
1649
1650         if (IS_ERR(tu)) {
1651                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1652                         (int)PTR_ERR(tu));
1653                 path_put(&path);
1654                 return ERR_CAST(tu);
1655         }
1656
1657         tu->offset = offs;
1658         tu->path = path;
1659         tu->ref_ctr_offset = ref_ctr_offset;
1660         tu->filename = kstrdup(name, GFP_KERNEL);
1661         if (!tu->filename) {
1662                 ret = -ENOMEM;
1663                 goto error;
1664         }
1665
1666         init_trace_event_call(tu);
1667
1668         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1669         if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1670                 ret = -ENOMEM;
1671                 goto error;
1672         }
1673
1674         return trace_probe_event_call(&tu->tp);
1675 error:
1676         free_trace_uprobe(tu);
1677         return ERR_PTR(ret);
1678 }
1679
1680 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1681 {
1682         struct trace_uprobe *tu;
1683
1684         tu = trace_uprobe_primary_from_call(event_call);
1685
1686         free_trace_uprobe(tu);
1687 }
1688 #endif /* CONFIG_PERF_EVENTS */
1689
1690 /* Make a trace interface for controlling probe points */
1691 static __init int init_uprobe_trace(void)
1692 {
1693         int ret;
1694
1695         ret = dyn_event_register(&trace_uprobe_ops);
1696         if (ret)
1697                 return ret;
1698
1699         ret = tracing_init_dentry();
1700         if (ret)
1701                 return 0;
1702
1703         trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1704                                     NULL, &uprobe_events_ops);
1705         /* Profile interface */
1706         trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1707                                     NULL, &uprobe_profile_ops);
1708         return 0;
1709 }
1710
1711 fs_initcall(init_uprobe_trace);