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