kernel/trace/trace_uprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * uprobes-based tracing events
   4  *
   5  * Copyright (C) IBM Corporation, 2010-2012
   6  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
   7  */
   8 #define pr_fmt(fmt)     "trace_uprobe: " fmt
   9
  10 #include <linux/bpf-cgroup.h>
  11 #include <linux/security.h>
  12 #include <linux/ctype.h>
  13 #include <linux/module.h>
  14 #include <linux/uaccess.h>
  15 #include <linux/uprobes.h>
  16 #include <linux/namei.h>
  17 #include <linux/string.h>
  18 #include <linux/rculist.h>
  19 #include <linux/filter.h>
  20
  21 #include "trace_dynevent.h"
  22 #include "trace_probe.h"
  23 #include "trace_probe_tmpl.h"
  24
  25 #define UPROBE_EVENT_SYSTEM     "uprobes"
  26
  27 struct uprobe_trace_entry_head {
  28         struct trace_entry      ent;
  29         unsigned long           vaddr[];
  30 };
  31
  32 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  33         (sizeof(struct uprobe_trace_entry_head) +       \
  34          sizeof(unsigned long) * (is_return ? 2 : 1))
  35
  36 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  37         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  38
  39 static int trace_uprobe_create(const char *raw_command);
  40 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  41 static int trace_uprobe_release(struct dyn_event *ev);
  42 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  43 static bool trace_uprobe_match(const char *system, const char *event,
  44                         int argc, const char **argv, struct dyn_event *ev);
  45
  46 static struct dyn_event_operations trace_uprobe_ops = {
  47         .create = trace_uprobe_create,
  48         .show = trace_uprobe_show,
  49         .is_busy = trace_uprobe_is_busy,
  50         .free = trace_uprobe_release,
  51         .match = trace_uprobe_match,
  52 };
  53
  54 /*
  55  * uprobe event core functions
  56  */
  57 struct trace_uprobe {
  58         struct dyn_event                devent;
  59         struct uprobe_consumer          consumer;
  60         struct path                     path;
  61         struct inode                    *inode;
  62         char                            *filename;
  63         unsigned long                   offset;
  64         unsigned long                   ref_ctr_offset;
  65         unsigned long                   nhit;
  66         struct trace_probe              tp;
  67 };
  68
  69 static bool is_trace_uprobe(struct dyn_event *ev)
  70 {
  71         return ev->ops == &trace_uprobe_ops;
  72 }
  73
  74 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  75 {
  76         return container_of(ev, struct trace_uprobe, devent);
  77 }
  78
  79 /**
  80  * for_each_trace_uprobe - iterate over the trace_uprobe list
  81  * @pos:        the struct trace_uprobe * for each entry
  82  * @dpos:       the struct dyn_event * to use as a loop cursor
  83  */
  84 #define for_each_trace_uprobe(pos, dpos)        \
  85         for_each_dyn_event(dpos)                \
  86                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  87
  88 static int register_uprobe_event(struct trace_uprobe *tu);
  89 static int unregister_uprobe_event(struct trace_uprobe *tu);
  90
  91 struct uprobe_dispatch_data {
  92         struct trace_uprobe     *tu;
  93         unsigned long           bp_addr;
  94 };
  95
  96 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  97 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  98                                 unsigned long func, struct pt_regs *regs);
  99
 100 #ifdef CONFIG_STACK_GROWSUP
 101 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 102 {
 103         return addr - (n * sizeof(long));
 104 }
 105 #else
 106 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 107 {
 108         return addr + (n * sizeof(long));
 109 }
 110 #endif
 111
 112 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 113 {
 114         unsigned long ret;
 115         unsigned long addr = user_stack_pointer(regs);
 116
 117         addr = adjust_stack_addr(addr, n);
 118
 119         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 120                 return 0;
 121
 122         return ret;
 123 }
 124
 125 /*
 126  * Uprobes-specific fetch functions
 127  */
 128 static nokprobe_inline int
 129 probe_mem_read(void *dest, void *src, size_t size)
 130 {
 131         void __user *vaddr = (void __force __user *)src;
 132
 133         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 134 }
 135
 136 static nokprobe_inline int
 137 probe_mem_read_user(void *dest, void *src, size_t size)
 138 {
 139         return probe_mem_read(dest, src, size);
 140 }
 141
 142 /*
 143  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 144  * length and relative data location.
 145  */
 146 static nokprobe_inline int
 147 fetch_store_string(unsigned long addr, void *dest, void *base)
 148 {
 149         long ret;
 150         u32 loc = *(u32 *)dest;
 151         int maxlen  = get_loc_len(loc);
 152         u8 *dst = get_loc_data(dest, base);
 153         void __user *src = (void __force __user *) addr;
 154
 155         if (unlikely(!maxlen))
 156                 return -ENOMEM;
 157
 158         if (addr == FETCH_TOKEN_COMM)
 159                 ret = strlcpy(dst, current->comm, maxlen);
 160         else
 161                 ret = strncpy_from_user(dst, src, maxlen);
 162         if (ret >= 0) {
 163                 if (ret == maxlen)
 164                         dst[ret - 1] = '\0';
 165                 else
 166                         /*
 167                          * Include the terminating null byte. In this case it
 168                          * was copied by strncpy_from_user but not accounted
 169                          * for in ret.
 170                          */
 171                         ret++;
 172                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 173         } else
 174                 *(u32 *)dest = make_data_loc(0, (void *)dst - base);
 175
 176         return ret;
 177 }
 178
 179 static nokprobe_inline int
 180 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 181 {
 182         return fetch_store_string(addr, dest, base);
 183 }
 184
 185 /* Return the length of string -- including null terminal byte */
 186 static nokprobe_inline int
 187 fetch_store_strlen(unsigned long addr)
 188 {
 189         int len;
 190         void __user *vaddr = (void __force __user *) addr;
 191
 192         if (addr == FETCH_TOKEN_COMM)
 193                 len = strlen(current->comm) + 1;
 194         else
 195                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 196
 197         return (len > MAX_STRING_SIZE) ? 0 : len;
 198 }
 199
 200 static nokprobe_inline int
 201 fetch_store_strlen_user(unsigned long addr)
 202 {
 203         return fetch_store_strlen(addr);
 204 }
 205
 206 static unsigned long translate_user_vaddr(unsigned long file_offset)
 207 {
 208         unsigned long base_addr;
 209         struct uprobe_dispatch_data *udd;
 210
 211         udd = (void *) current->utask->vaddr;
 212
 213         base_addr = udd->bp_addr - udd->tu->offset;
 214         return base_addr + file_offset;
 215 }
 216
 217 /* Note that we don't verify it, since the code does not come from user space */
 218 static int
 219 process_fetch_insn(struct fetch_insn *code, void *rec, void *dest,
 220                    void *base)
 221 {
 222         struct pt_regs *regs = rec;
 223         unsigned long val;
 224         int ret;
 225
 226         /* 1st stage: get value from context */
 227         switch (code->op) {
 228         case FETCH_OP_REG:
 229                 val = regs_get_register(regs, code->param);
 230                 break;
 231         case FETCH_OP_STACK:
 232                 val = get_user_stack_nth(regs, code->param);
 233                 break;
 234         case FETCH_OP_STACKP:
 235                 val = user_stack_pointer(regs);
 236                 break;
 237         case FETCH_OP_RETVAL:
 238                 val = regs_return_value(regs);
 239                 break;
 240         case FETCH_OP_COMM:
 241                 val = FETCH_TOKEN_COMM;
 242                 break;
 243         case FETCH_OP_FOFFS:
 244                 val = translate_user_vaddr(code->immediate);
 245                 break;
 246         default:
 247                 ret = process_common_fetch_insn(code, &val);
 248                 if (ret < 0)
 249                         return ret;
 250         }
 251         code++;
 252
 253         return process_fetch_insn_bottom(code, val, dest, base);
 254 }
 255 NOKPROBE_SYMBOL(process_fetch_insn)
 256
 257 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 258 {
 259         rwlock_init(&filter->rwlock);
 260         filter->nr_systemwide = 0;
 261         INIT_LIST_HEAD(&filter->perf_events);
 262 }
 263
 264 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 265 {
 266         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 267 }
 268
 269 static inline bool is_ret_probe(struct trace_uprobe *tu)
 270 {
 271         return tu->consumer.ret_handler != NULL;
 272 }
 273
 274 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 275 {
 276         struct trace_uprobe *tu = to_trace_uprobe(ev);
 277
 278         return trace_probe_is_enabled(&tu->tp);
 279 }
 280
 281 static bool trace_uprobe_match_command_head(struct trace_uprobe *tu,
 282                                             int argc, const char **argv)
 283 {
 284         char buf[MAX_ARGSTR_LEN + 1];
 285         int len;
 286
 287         if (!argc)
 288                 return true;
 289
 290         len = strlen(tu->filename);
 291         if (strncmp(tu->filename, argv[0], len) || argv[0][len] != ':')
 292                 return false;
 293
 294         if (tu->ref_ctr_offset == 0)
 295                 snprintf(buf, sizeof(buf), "0x%0*lx",
 296                                 (int)(sizeof(void *) * 2), tu->offset);
 297         else
 298                 snprintf(buf, sizeof(buf), "0x%0*lx(0x%lx)",
 299                                 (int)(sizeof(void *) * 2), tu->offset,
 300                                 tu->ref_ctr_offset);
 301         if (strcmp(buf, &argv[0][len + 1]))
 302                 return false;
 303
 304         argc--; argv++;
 305
 306         return trace_probe_match_command_args(&tu->tp, argc, argv);
 307 }
 308
 309 static bool trace_uprobe_match(const char *system, const char *event,
 310                         int argc, const char **argv, struct dyn_event *ev)
 311 {
 312         struct trace_uprobe *tu = to_trace_uprobe(ev);
 313
 314         return (event[0] == '\0' ||
 315                 strcmp(trace_probe_name(&tu->tp), event) == 0) &&
 316            (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0) &&
 317            trace_uprobe_match_command_head(tu, argc, argv);
 318 }
 319
 320 static nokprobe_inline struct trace_uprobe *
 321 trace_uprobe_primary_from_call(struct trace_event_call *call)
 322 {
 323         struct trace_probe *tp;
 324
 325         tp = trace_probe_primary_from_call(call);
 326         if (WARN_ON_ONCE(!tp))
 327                 return NULL;
 328
 329         return container_of(tp, struct trace_uprobe, tp);
 330 }
 331
 332 /*
 333  * Allocate new trace_uprobe and initialize it (including uprobes).
 334  */
 335 static struct trace_uprobe *
 336 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 337 {
 338         struct trace_uprobe *tu;
 339         int ret;
 340
 341         tu = kzalloc(struct_size(tu, tp.args, nargs), GFP_KERNEL);
 342         if (!tu)
 343                 return ERR_PTR(-ENOMEM);
 344
 345         ret = trace_probe_init(&tu->tp, event, group, true);
 346         if (ret < 0)
 347                 goto error;
 348
 349         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 350         tu->consumer.handler = uprobe_dispatcher;
 351         if (is_ret)
 352                 tu->consumer.ret_handler = uretprobe_dispatcher;
 353         init_trace_uprobe_filter(tu->tp.event->filter);
 354         return tu;
 355
 356 error:
 357         kfree(tu);
 358
 359         return ERR_PTR(ret);
 360 }
 361
 362 static void free_trace_uprobe(struct trace_uprobe *tu)
 363 {
 364         if (!tu)
 365                 return;
 366
 367         path_put(&tu->path);
 368         trace_probe_cleanup(&tu->tp);
 369         kfree(tu->filename);
 370         kfree(tu);
 371 }
 372
 373 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 374 {
 375         struct dyn_event *pos;
 376         struct trace_uprobe *tu;
 377
 378         for_each_trace_uprobe(tu, pos)
 379                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 380                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 381                         return tu;
 382
 383         return NULL;
 384 }
 385
 386 /* Unregister a trace_uprobe and probe_event */
 387 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 388 {
 389         int ret;
 390
 391         if (trace_probe_has_sibling(&tu->tp))
 392                 goto unreg;
 393
 394         /* If there's a reference to the dynamic event */
 395         if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 396                 return -EBUSY;
 397
 398         ret = unregister_uprobe_event(tu);
 399         if (ret)
 400                 return ret;
 401
 402 unreg:
 403         dyn_event_remove(&tu->devent);
 404         trace_probe_unlink(&tu->tp);
 405         free_trace_uprobe(tu);
 406         return 0;
 407 }
 408
 409 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 410                                          struct trace_uprobe *comp)
 411 {
 412         struct trace_probe_event *tpe = orig->tp.event;
 413         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 414         int i;
 415
 416         list_for_each_entry(orig, &tpe->probes, tp.list) {
 417                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 418                     comp->offset != orig->offset)
 419                         continue;
 420
 421                 /*
 422                  * trace_probe_compare_arg_type() ensured that nr_args and
 423                  * each argument name and type are same. Let's compare comm.
 424                  */
 425                 for (i = 0; i < orig->tp.nr_args; i++) {
 426                         if (strcmp(orig->tp.args[i].comm,
 427                                    comp->tp.args[i].comm))
 428                                 break;
 429                 }
 430
 431                 if (i == orig->tp.nr_args)
 432                         return true;
 433         }
 434
 435         return false;
 436 }
 437
 438 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 439 {
 440         int ret;
 441
 442         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 443         if (ret) {
 444                 /* Note that argument starts index = 2 */
 445                 trace_probe_log_set_index(ret + 1);
 446                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 447                 return -EEXIST;
 448         }
 449         if (trace_uprobe_has_same_uprobe(to, tu)) {
 450                 trace_probe_log_set_index(0);
 451                 trace_probe_log_err(0, SAME_PROBE);
 452                 return -EEXIST;
 453         }
 454
 455         /* Append to existing event */
 456         ret = trace_probe_append(&tu->tp, &to->tp);
 457         if (!ret)
 458                 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 459
 460         return ret;
 461 }
 462
 463 /*
 464  * Uprobe with multiple reference counter is not allowed. i.e.
 465  * If inode and offset matches, reference counter offset *must*
 466  * match as well. Though, there is one exception: If user is
 467  * replacing old trace_uprobe with new one(same group/event),
 468  * then we allow same uprobe with new reference counter as far
 469  * as the new one does not conflict with any other existing
 470  * ones.
 471  */
 472 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 473 {
 474         struct dyn_event *pos;
 475         struct trace_uprobe *tmp;
 476         struct inode *new_inode = d_real_inode(new->path.dentry);
 477
 478         for_each_trace_uprobe(tmp, pos) {
 479                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 480                     new->offset == tmp->offset &&
 481                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 482                         pr_warn("Reference counter offset mismatch.");
 483                         return -EINVAL;
 484                 }
 485         }
 486         return 0;
 487 }
 488
 489 /* Register a trace_uprobe and probe_event */
 490 static int register_trace_uprobe(struct trace_uprobe *tu)
 491 {
 492         struct trace_uprobe *old_tu;
 493         int ret;
 494
 495         mutex_lock(&event_mutex);
 496
 497         ret = validate_ref_ctr_offset(tu);
 498         if (ret)
 499                 goto end;
 500
 501         /* register as an event */
 502         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 503                                   trace_probe_group_name(&tu->tp));
 504         if (old_tu) {
 505                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 506                         trace_probe_log_set_index(0);
 507                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 508                         ret = -EEXIST;
 509                 } else {
 510                         ret = append_trace_uprobe(tu, old_tu);
 511                 }
 512                 goto end;
 513         }
 514
 515         ret = register_uprobe_event(tu);
 516         if (ret) {
 517                 if (ret == -EEXIST) {
 518                         trace_probe_log_set_index(0);
 519                         trace_probe_log_err(0, EVENT_EXIST);
 520                 } else
 521                         pr_warn("Failed to register probe event(%d)\n", ret);
 522                 goto end;
 523         }
 524
 525         dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 526
 527 end:
 528         mutex_unlock(&event_mutex);
 529
 530         return ret;
 531 }
 532
 533 /*
 534  * Argument syntax:
 535  *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 536  */
 537 static int __trace_uprobe_create(int argc, const char **argv)
 538 {
 539         struct trace_uprobe *tu;
 540         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 541         char *arg, *filename, *rctr, *rctr_end, *tmp;
 542         char buf[MAX_EVENT_NAME_LEN];
 543         char gbuf[MAX_EVENT_NAME_LEN];
 544         enum probe_print_type ptype;
 545         struct path path;
 546         unsigned long offset, ref_ctr_offset;
 547         bool is_return = false;
 548         int i, ret;
 549
 550         ref_ctr_offset = 0;
 551
 552         switch (argv[0][0]) {
 553         case 'r':
 554                 is_return = true;
 555                 break;
 556         case 'p':
 557                 break;
 558         default:
 559                 return -ECANCELED;
 560         }
 561
 562         if (argc < 2)
 563                 return -ECANCELED;
 564
 565         if (argv[0][1] == ':')
 566                 event = &argv[0][2];
 567
 568         if (!strchr(argv[1], '/'))
 569                 return -ECANCELED;
 570
 571         filename = kstrdup(argv[1], GFP_KERNEL);
 572         if (!filename)
 573                 return -ENOMEM;
 574
 575         /* Find the last occurrence, in case the path contains ':' too. */
 576         arg = strrchr(filename, ':');
 577         if (!arg || !isdigit(arg[1])) {
 578                 kfree(filename);
 579                 return -ECANCELED;
 580         }
 581
 582         trace_probe_log_init("trace_uprobe", argc, argv);
 583         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 584
 585         *arg++ = '\0';
 586         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 587         if (ret) {
 588                 trace_probe_log_err(0, FILE_NOT_FOUND);
 589                 kfree(filename);
 590                 trace_probe_log_clear();
 591                 return ret;
 592         }
 593         if (!d_is_reg(path.dentry)) {
 594                 trace_probe_log_err(0, NO_REGULAR_FILE);
 595                 ret = -EINVAL;
 596                 goto fail_address_parse;
 597         }
 598
 599         /* Parse reference counter offset if specified. */
 600         rctr = strchr(arg, '(');
 601         if (rctr) {
 602                 rctr_end = strchr(rctr, ')');
 603                 if (!rctr_end) {
 604                         ret = -EINVAL;
 605                         rctr_end = rctr + strlen(rctr);
 606                         trace_probe_log_err(rctr_end - filename,
 607                                             REFCNT_OPEN_BRACE);
 608                         goto fail_address_parse;
 609                 } else if (rctr_end[1] != '\0') {
 610                         ret = -EINVAL;
 611                         trace_probe_log_err(rctr_end + 1 - filename,
 612                                             BAD_REFCNT_SUFFIX);
 613                         goto fail_address_parse;
 614                 }
 615
 616                 *rctr++ = '\0';
 617                 *rctr_end = '\0';
 618                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 619                 if (ret) {
 620                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 621                         goto fail_address_parse;
 622                 }
 623         }
 624
 625         /* Check if there is %return suffix */
 626         tmp = strchr(arg, '%');
 627         if (tmp) {
 628                 if (!strcmp(tmp, "%return")) {
 629                         *tmp = '\0';
 630                         is_return = true;
 631                 } else {
 632                         trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 633                         ret = -EINVAL;
 634                         goto fail_address_parse;
 635                 }
 636         }
 637
 638         /* Parse uprobe offset. */
 639         ret = kstrtoul(arg, 0, &offset);
 640         if (ret) {
 641                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 642                 goto fail_address_parse;
 643         }
 644
 645         /* setup a probe */
 646         trace_probe_log_set_index(0);
 647         if (event) {
 648                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
 649                                                   event - argv[0]);
 650                 if (ret)
 651                         goto fail_address_parse;
 652         }
 653
 654         if (!event) {
 655                 char *tail;
 656                 char *ptr;
 657
 658                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 659                 if (!tail) {
 660                         ret = -ENOMEM;
 661                         goto fail_address_parse;
 662                 }
 663
 664                 ptr = strpbrk(tail, ".-_");
 665                 if (ptr)
 666                         *ptr = '\0';
 667
 668                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 669                 event = buf;
 670                 kfree(tail);
 671         }
 672
 673         argc -= 2;
 674         argv += 2;
 675
 676         tu = alloc_trace_uprobe(group, event, argc, is_return);
 677         if (IS_ERR(tu)) {
 678                 ret = PTR_ERR(tu);
 679                 /* This must return -ENOMEM otherwise there is a bug */
 680                 WARN_ON_ONCE(ret != -ENOMEM);
 681                 goto fail_address_parse;
 682         }
 683         tu->offset = offset;
 684         tu->ref_ctr_offset = ref_ctr_offset;
 685         tu->path = path;
 686         tu->filename = filename;
 687
 688         /* parse arguments */
 689         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 690                 struct traceprobe_parse_context ctx = {
 691                         .flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER,
 692                 };
 693
 694                 trace_probe_log_set_index(i + 2);
 695                 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], &ctx);
 696                 if (ret)
 697                         goto error;
 698         }
 699
 700         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 701         ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 702         if (ret < 0)
 703                 goto error;
 704
 705         ret = register_trace_uprobe(tu);
 706         if (!ret)
 707                 goto out;
 708
 709 error:
 710         free_trace_uprobe(tu);
 711 out:
 712         trace_probe_log_clear();
 713         return ret;
 714
 715 fail_address_parse:
 716         trace_probe_log_clear();
 717         path_put(&path);
 718         kfree(filename);
 719
 720         return ret;
 721 }
 722
 723 int trace_uprobe_create(const char *raw_command)
 724 {
 725         return trace_probe_create(raw_command, __trace_uprobe_create);
 726 }
 727
 728 static int create_or_delete_trace_uprobe(const char *raw_command)
 729 {
 730         int ret;
 731
 732         if (raw_command[0] == '-')
 733                 return dyn_event_release(raw_command, &trace_uprobe_ops);
 734
 735         ret = trace_uprobe_create(raw_command);
 736         return ret == -ECANCELED ? -EINVAL : ret;
 737 }
 738
 739 static int trace_uprobe_release(struct dyn_event *ev)
 740 {
 741         struct trace_uprobe *tu = to_trace_uprobe(ev);
 742
 743         return unregister_trace_uprobe(tu);
 744 }
 745
 746 /* Probes listing interfaces */
 747 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 748 {
 749         struct trace_uprobe *tu = to_trace_uprobe(ev);
 750         char c = is_ret_probe(tu) ? 'r' : 'p';
 751         int i;
 752
 753         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 754                         trace_probe_name(&tu->tp), tu->filename,
 755                         (int)(sizeof(void *) * 2), tu->offset);
 756
 757         if (tu->ref_ctr_offset)
 758                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 759
 760         for (i = 0; i < tu->tp.nr_args; i++)
 761                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 762
 763         seq_putc(m, '\n');
 764         return 0;
 765 }
 766
 767 static int probes_seq_show(struct seq_file *m, void *v)
 768 {
 769         struct dyn_event *ev = v;
 770
 771         if (!is_trace_uprobe(ev))
 772                 return 0;
 773
 774         return trace_uprobe_show(m, ev);
 775 }
 776
 777 static const struct seq_operations probes_seq_op = {
 778         .start  = dyn_event_seq_start,
 779         .next   = dyn_event_seq_next,
 780         .stop   = dyn_event_seq_stop,
 781         .show   = probes_seq_show
 782 };
 783
 784 static int probes_open(struct inode *inode, struct file *file)
 785 {
 786         int ret;
 787
 788         ret = security_locked_down(LOCKDOWN_TRACEFS);
 789         if (ret)
 790                 return ret;
 791
 792         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 793                 ret = dyn_events_release_all(&trace_uprobe_ops);
 794                 if (ret)
 795                         return ret;
 796         }
 797
 798         return seq_open(file, &probes_seq_op);
 799 }
 800
 801 static ssize_t probes_write(struct file *file, const char __user *buffer,
 802                             size_t count, loff_t *ppos)
 803 {
 804         return trace_parse_run_command(file, buffer, count, ppos,
 805                                         create_or_delete_trace_uprobe);
 806 }
 807
 808 static const struct file_operations uprobe_events_ops = {
 809         .owner          = THIS_MODULE,
 810         .open           = probes_open,
 811         .read           = seq_read,
 812         .llseek         = seq_lseek,
 813         .release        = seq_release,
 814         .write          = probes_write,
 815 };
 816
 817 /* Probes profiling interfaces */
 818 static int probes_profile_seq_show(struct seq_file *m, void *v)
 819 {
 820         struct dyn_event *ev = v;
 821         struct trace_uprobe *tu;
 822
 823         if (!is_trace_uprobe(ev))
 824                 return 0;
 825
 826         tu = to_trace_uprobe(ev);
 827         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 828                         trace_probe_name(&tu->tp), tu->nhit);
 829         return 0;
 830 }
 831
 832 static const struct seq_operations profile_seq_op = {
 833         .start  = dyn_event_seq_start,
 834         .next   = dyn_event_seq_next,
 835         .stop   = dyn_event_seq_stop,
 836         .show   = probes_profile_seq_show
 837 };
 838
 839 static int profile_open(struct inode *inode, struct file *file)
 840 {
 841         int ret;
 842
 843         ret = security_locked_down(LOCKDOWN_TRACEFS);
 844         if (ret)
 845                 return ret;
 846
 847         return seq_open(file, &profile_seq_op);
 848 }
 849
 850 static const struct file_operations uprobe_profile_ops = {
 851         .owner          = THIS_MODULE,
 852         .open           = profile_open,
 853         .read           = seq_read,
 854         .llseek         = seq_lseek,
 855         .release        = seq_release,
 856 };
 857
 858 struct uprobe_cpu_buffer {
 859         struct mutex mutex;
 860         void *buf;
 861 };
 862 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 863 static int uprobe_buffer_refcnt;
 864
 865 static int uprobe_buffer_init(void)
 866 {
 867         int cpu, err_cpu;
 868
 869         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 870         if (uprobe_cpu_buffer == NULL)
 871                 return -ENOMEM;
 872
 873         for_each_possible_cpu(cpu) {
 874                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 875                                                   GFP_KERNEL, 0);
 876                 if (p == NULL) {
 877                         err_cpu = cpu;
 878                         goto err;
 879                 }
 880                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 881                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 882         }
 883
 884         return 0;
 885
 886 err:
 887         for_each_possible_cpu(cpu) {
 888                 if (cpu == err_cpu)
 889                         break;
 890                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 891         }
 892
 893         free_percpu(uprobe_cpu_buffer);
 894         return -ENOMEM;
 895 }
 896
 897 static int uprobe_buffer_enable(void)
 898 {
 899         int ret = 0;
 900
 901         BUG_ON(!mutex_is_locked(&event_mutex));
 902
 903         if (uprobe_buffer_refcnt++ == 0) {
 904                 ret = uprobe_buffer_init();
 905                 if (ret < 0)
 906                         uprobe_buffer_refcnt--;
 907         }
 908
 909         return ret;
 910 }
 911
 912 static void uprobe_buffer_disable(void)
 913 {
 914         int cpu;
 915
 916         BUG_ON(!mutex_is_locked(&event_mutex));
 917
 918         if (--uprobe_buffer_refcnt == 0) {
 919                 for_each_possible_cpu(cpu)
 920                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 921                                                              cpu)->buf);
 922
 923                 free_percpu(uprobe_cpu_buffer);
 924                 uprobe_cpu_buffer = NULL;
 925         }
 926 }
 927
 928 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 929 {
 930         struct uprobe_cpu_buffer *ucb;
 931         int cpu;
 932
 933         cpu = raw_smp_processor_id();
 934         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 935
 936         /*
 937          * Use per-cpu buffers for fastest access, but we might migrate
 938          * so the mutex makes sure we have sole access to it.
 939          */
 940         mutex_lock(&ucb->mutex);
 941
 942         return ucb;
 943 }
 944
 945 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 946 {
 947         mutex_unlock(&ucb->mutex);
 948 }
 949
 950 static void __uprobe_trace_func(struct trace_uprobe *tu,
 951                                 unsigned long func, struct pt_regs *regs,
 952                                 struct uprobe_cpu_buffer *ucb, int dsize,
 953                                 struct trace_event_file *trace_file)
 954 {
 955         struct uprobe_trace_entry_head *entry;
 956         struct trace_event_buffer fbuffer;
 957         void *data;
 958         int size, esize;
 959         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 960
 961         WARN_ON(call != trace_file->event_call);
 962
 963         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 964                 return;
 965
 966         if (trace_trigger_soft_disabled(trace_file))
 967                 return;
 968
 969         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 970         size = esize + tu->tp.size + dsize;
 971         entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
 972         if (!entry)
 973                 return;
 974
 975         if (is_ret_probe(tu)) {
 976                 entry->vaddr[0] = func;
 977                 entry->vaddr[1] = instruction_pointer(regs);
 978                 data = DATAOF_TRACE_ENTRY(entry, true);
 979         } else {
 980                 entry->vaddr[0] = instruction_pointer(regs);
 981                 data = DATAOF_TRACE_ENTRY(entry, false);
 982         }
 983
 984         memcpy(data, ucb->buf, tu->tp.size + dsize);
 985
 986         trace_event_buffer_commit(&fbuffer);
 987 }
 988
 989 /* uprobe handler */
 990 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 991                              struct uprobe_cpu_buffer *ucb, int dsize)
 992 {
 993         struct event_file_link *link;
 994
 995         if (is_ret_probe(tu))
 996                 return 0;
 997
 998         rcu_read_lock();
 999         trace_probe_for_each_link_rcu(link, &tu->tp)
1000                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
1001         rcu_read_unlock();
1002
1003         return 0;
1004 }
1005
1006 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1007                                  struct pt_regs *regs,
1008                                  struct uprobe_cpu_buffer *ucb, int dsize)
1009 {
1010         struct event_file_link *link;
1011
1012         rcu_read_lock();
1013         trace_probe_for_each_link_rcu(link, &tu->tp)
1014                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
1015         rcu_read_unlock();
1016 }
1017
1018 /* Event entry printers */
1019 static enum print_line_t
1020 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1021 {
1022         struct uprobe_trace_entry_head *entry;
1023         struct trace_seq *s = &iter->seq;
1024         struct trace_uprobe *tu;
1025         u8 *data;
1026
1027         entry = (struct uprobe_trace_entry_head *)iter->ent;
1028         tu = trace_uprobe_primary_from_call(
1029                 container_of(event, struct trace_event_call, event));
1030         if (unlikely(!tu))
1031                 goto out;
1032
1033         if (is_ret_probe(tu)) {
1034                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1035                                  trace_probe_name(&tu->tp),
1036                                  entry->vaddr[1], entry->vaddr[0]);
1037                 data = DATAOF_TRACE_ENTRY(entry, true);
1038         } else {
1039                 trace_seq_printf(s, "%s: (0x%lx)",
1040                                  trace_probe_name(&tu->tp),
1041                                  entry->vaddr[0]);
1042                 data = DATAOF_TRACE_ENTRY(entry, false);
1043         }
1044
1045         if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1046                 goto out;
1047
1048         trace_seq_putc(s, '\n');
1049
1050  out:
1051         return trace_handle_return(s);
1052 }
1053
1054 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
1055                                 enum uprobe_filter_ctx ctx,
1056                                 struct mm_struct *mm);
1057
1058 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1059 {
1060         int ret;
1061
1062         tu->consumer.filter = filter;
1063         tu->inode = d_real_inode(tu->path.dentry);
1064
1065         if (tu->ref_ctr_offset)
1066                 ret = uprobe_register_refctr(tu->inode, tu->offset,
1067                                 tu->ref_ctr_offset, &tu->consumer);
1068         else
1069                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
1070
1071         if (ret)
1072                 tu->inode = NULL;
1073
1074         return ret;
1075 }
1076
1077 static void __probe_event_disable(struct trace_probe *tp)
1078 {
1079         struct trace_uprobe *tu;
1080
1081         tu = container_of(tp, struct trace_uprobe, tp);
1082         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1083
1084         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1085                 if (!tu->inode)
1086                         continue;
1087
1088                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
1089                 tu->inode = NULL;
1090         }
1091 }
1092
1093 static int probe_event_enable(struct trace_event_call *call,
1094                         struct trace_event_file *file, filter_func_t filter)
1095 {
1096         struct trace_probe *tp;
1097         struct trace_uprobe *tu;
1098         bool enabled;
1099         int ret;
1100
1101         tp = trace_probe_primary_from_call(call);
1102         if (WARN_ON_ONCE(!tp))
1103                 return -ENODEV;
1104         enabled = trace_probe_is_enabled(tp);
1105
1106         /* This may also change "enabled" state */
1107         if (file) {
1108                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1109                         return -EINTR;
1110
1111                 ret = trace_probe_add_file(tp, file);
1112                 if (ret < 0)
1113                         return ret;
1114         } else {
1115                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1116                         return -EINTR;
1117
1118                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1119         }
1120
1121         tu = container_of(tp, struct trace_uprobe, tp);
1122         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1123
1124         if (enabled)
1125                 return 0;
1126
1127         ret = uprobe_buffer_enable();
1128         if (ret)
1129                 goto err_flags;
1130
1131         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1132                 ret = trace_uprobe_enable(tu, filter);
1133                 if (ret) {
1134                         __probe_event_disable(tp);
1135                         goto err_buffer;
1136                 }
1137         }
1138
1139         return 0;
1140
1141  err_buffer:
1142         uprobe_buffer_disable();
1143
1144  err_flags:
1145         if (file)
1146                 trace_probe_remove_file(tp, file);
1147         else
1148                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1149
1150         return ret;
1151 }
1152
1153 static void probe_event_disable(struct trace_event_call *call,
1154                                 struct trace_event_file *file)
1155 {
1156         struct trace_probe *tp;
1157
1158         tp = trace_probe_primary_from_call(call);
1159         if (WARN_ON_ONCE(!tp))
1160                 return;
1161
1162         if (!trace_probe_is_enabled(tp))
1163                 return;
1164
1165         if (file) {
1166                 if (trace_probe_remove_file(tp, file) < 0)
1167                         return;
1168
1169                 if (trace_probe_is_enabled(tp))
1170                         return;
1171         } else
1172                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1173
1174         __probe_event_disable(tp);
1175         uprobe_buffer_disable();
1176 }
1177
1178 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1179 {
1180         int ret, size;
1181         struct uprobe_trace_entry_head field;
1182         struct trace_uprobe *tu;
1183
1184         tu = trace_uprobe_primary_from_call(event_call);
1185         if (unlikely(!tu))
1186                 return -ENODEV;
1187
1188         if (is_ret_probe(tu)) {
1189                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1190                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1191                 size = SIZEOF_TRACE_ENTRY(true);
1192         } else {
1193                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1194                 size = SIZEOF_TRACE_ENTRY(false);
1195         }
1196
1197         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1198 }
1199
1200 #ifdef CONFIG_PERF_EVENTS
1201 static bool
1202 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1203 {
1204         struct perf_event *event;
1205
1206         if (filter->nr_systemwide)
1207                 return true;
1208
1209         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1210                 if (event->hw.target->mm == mm)
1211                         return true;
1212         }
1213
1214         return false;
1215 }
1216
1217 static inline bool
1218 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1219                           struct perf_event *event)
1220 {
1221         return __uprobe_perf_filter(filter, event->hw.target->mm);
1222 }
1223
1224 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1225                                        struct perf_event *event)
1226 {
1227         bool done;
1228
1229         write_lock(&filter->rwlock);
1230         if (event->hw.target) {
1231                 list_del(&event->hw.tp_list);
1232                 done = filter->nr_systemwide ||
1233                         (event->hw.target->flags & PF_EXITING) ||
1234                         trace_uprobe_filter_event(filter, event);
1235         } else {
1236                 filter->nr_systemwide--;
1237                 done = filter->nr_systemwide;
1238         }
1239         write_unlock(&filter->rwlock);
1240
1241         return done;
1242 }
1243
1244 /* This returns true if the filter always covers target mm */
1245 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1246                                     struct perf_event *event)
1247 {
1248         bool done;
1249
1250         write_lock(&filter->rwlock);
1251         if (event->hw.target) {
1252                 /*
1253                  * event->parent != NULL means copy_process(), we can avoid
1254                  * uprobe_apply(). current->mm must be probed and we can rely
1255                  * on dup_mmap() which preserves the already installed bp's.
1256                  *
1257                  * attr.enable_on_exec means that exec/mmap will install the
1258                  * breakpoints we need.
1259                  */
1260                 done = filter->nr_systemwide ||
1261                         event->parent || event->attr.enable_on_exec ||
1262                         trace_uprobe_filter_event(filter, event);
1263                 list_add(&event->hw.tp_list, &filter->perf_events);
1264         } else {
1265                 done = filter->nr_systemwide;
1266                 filter->nr_systemwide++;
1267         }
1268         write_unlock(&filter->rwlock);
1269
1270         return done;
1271 }
1272
1273 static int uprobe_perf_close(struct trace_event_call *call,
1274                              struct perf_event *event)
1275 {
1276         struct trace_probe *tp;
1277         struct trace_uprobe *tu;
1278         int ret = 0;
1279
1280         tp = trace_probe_primary_from_call(call);
1281         if (WARN_ON_ONCE(!tp))
1282                 return -ENODEV;
1283
1284         tu = container_of(tp, struct trace_uprobe, tp);
1285         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1286                 return 0;
1287
1288         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1289                 ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1290                 if (ret)
1291                         break;
1292         }
1293
1294         return ret;
1295 }
1296
1297 static int uprobe_perf_open(struct trace_event_call *call,
1298                             struct perf_event *event)
1299 {
1300         struct trace_probe *tp;
1301         struct trace_uprobe *tu;
1302         int err = 0;
1303
1304         tp = trace_probe_primary_from_call(call);
1305         if (WARN_ON_ONCE(!tp))
1306                 return -ENODEV;
1307
1308         tu = container_of(tp, struct trace_uprobe, tp);
1309         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1310                 return 0;
1311
1312         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1313                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1314                 if (err) {
1315                         uprobe_perf_close(call, event);
1316                         break;
1317                 }
1318         }
1319
1320         return err;
1321 }
1322
1323 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1324                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1325 {
1326         struct trace_uprobe_filter *filter;
1327         struct trace_uprobe *tu;
1328         int ret;
1329
1330         tu = container_of(uc, struct trace_uprobe, consumer);
1331         filter = tu->tp.event->filter;
1332
1333         read_lock(&filter->rwlock);
1334         ret = __uprobe_perf_filter(filter, mm);
1335         read_unlock(&filter->rwlock);
1336
1337         return ret;
1338 }
1339
1340 static void __uprobe_perf_func(struct trace_uprobe *tu,
1341                                unsigned long func, struct pt_regs *regs,
1342                                struct uprobe_cpu_buffer *ucb, int dsize)
1343 {
1344         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1345         struct uprobe_trace_entry_head *entry;
1346         struct hlist_head *head;
1347         void *data;
1348         int size, esize;
1349         int rctx;
1350
1351 #ifdef CONFIG_BPF_EVENTS
1352         if (bpf_prog_array_valid(call)) {
1353                 u32 ret;
1354
1355                 ret = bpf_prog_run_array_sleepable(call->prog_array, regs, bpf_prog_run);
1356                 if (!ret)
1357                         return;
1358         }
1359 #endif /* CONFIG_BPF_EVENTS */
1360
1361         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1362
1363         size = esize + tu->tp.size + dsize;
1364         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1365         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1366                 return;
1367
1368         preempt_disable();
1369         head = this_cpu_ptr(call->perf_events);
1370         if (hlist_empty(head))
1371                 goto out;
1372
1373         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1374         if (!entry)
1375                 goto out;
1376
1377         if (is_ret_probe(tu)) {
1378                 entry->vaddr[0] = func;
1379                 entry->vaddr[1] = instruction_pointer(regs);
1380                 data = DATAOF_TRACE_ENTRY(entry, true);
1381         } else {
1382                 entry->vaddr[0] = instruction_pointer(regs);
1383                 data = DATAOF_TRACE_ENTRY(entry, false);
1384         }
1385
1386         memcpy(data, ucb->buf, tu->tp.size + dsize);
1387
1388         if (size - esize > tu->tp.size + dsize) {
1389                 int len = tu->tp.size + dsize;
1390
1391                 memset(data + len, 0, size - esize - len);
1392         }
1393
1394         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1395                               head, NULL);
1396  out:
1397         preempt_enable();
1398 }
1399
1400 /* uprobe profile handler */
1401 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1402                             struct uprobe_cpu_buffer *ucb, int dsize)
1403 {
1404         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1405                 return UPROBE_HANDLER_REMOVE;
1406
1407         if (!is_ret_probe(tu))
1408                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1409         return 0;
1410 }
1411
1412 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1413                                 struct pt_regs *regs,
1414                                 struct uprobe_cpu_buffer *ucb, int dsize)
1415 {
1416         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1417 }
1418
1419 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1420                         const char **filename, u64 *probe_offset,
1421                         bool perf_type_tracepoint)
1422 {
1423         const char *pevent = trace_event_name(event->tp_event);
1424         const char *group = event->tp_event->class->system;
1425         struct trace_uprobe *tu;
1426
1427         if (perf_type_tracepoint)
1428                 tu = find_probe_event(pevent, group);
1429         else
1430                 tu = trace_uprobe_primary_from_call(event->tp_event);
1431         if (!tu)
1432                 return -EINVAL;
1433
1434         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1435                                     : BPF_FD_TYPE_UPROBE;
1436         *filename = tu->filename;
1437         *probe_offset = tu->offset;
1438         return 0;
1439 }
1440 #endif  /* CONFIG_PERF_EVENTS */
1441
1442 static int
1443 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1444                       void *data)
1445 {
1446         struct trace_event_file *file = data;
1447
1448         switch (type) {
1449         case TRACE_REG_REGISTER:
1450                 return probe_event_enable(event, file, NULL);
1451
1452         case TRACE_REG_UNREGISTER:
1453                 probe_event_disable(event, file);
1454                 return 0;
1455
1456 #ifdef CONFIG_PERF_EVENTS
1457         case TRACE_REG_PERF_REGISTER:
1458                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1459
1460         case TRACE_REG_PERF_UNREGISTER:
1461                 probe_event_disable(event, NULL);
1462                 return 0;
1463
1464         case TRACE_REG_PERF_OPEN:
1465                 return uprobe_perf_open(event, data);
1466
1467         case TRACE_REG_PERF_CLOSE:
1468                 return uprobe_perf_close(event, data);
1469
1470 #endif
1471         default:
1472                 return 0;
1473         }
1474 }
1475
1476 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1477 {
1478         struct trace_uprobe *tu;
1479         struct uprobe_dispatch_data udd;
1480         struct uprobe_cpu_buffer *ucb;
1481         int dsize, esize;
1482         int ret = 0;
1483
1484
1485         tu = container_of(con, struct trace_uprobe, consumer);
1486         tu->nhit++;
1487
1488         udd.tu = tu;
1489         udd.bp_addr = instruction_pointer(regs);
1490
1491         current->utask->vaddr = (unsigned long) &udd;
1492
1493         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1494                 return 0;
1495
1496         dsize = __get_data_size(&tu->tp, regs);
1497         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1498
1499         ucb = uprobe_buffer_get();
1500         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1501
1502         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1503                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1504
1505 #ifdef CONFIG_PERF_EVENTS
1506         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1507                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1508 #endif
1509         uprobe_buffer_put(ucb);
1510         return ret;
1511 }
1512
1513 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1514                                 unsigned long func, struct pt_regs *regs)
1515 {
1516         struct trace_uprobe *tu;
1517         struct uprobe_dispatch_data udd;
1518         struct uprobe_cpu_buffer *ucb;
1519         int dsize, esize;
1520
1521         tu = container_of(con, struct trace_uprobe, consumer);
1522
1523         udd.tu = tu;
1524         udd.bp_addr = func;
1525
1526         current->utask->vaddr = (unsigned long) &udd;
1527
1528         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1529                 return 0;
1530
1531         dsize = __get_data_size(&tu->tp, regs);
1532         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1533
1534         ucb = uprobe_buffer_get();
1535         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1536
1537         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1538                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1539
1540 #ifdef CONFIG_PERF_EVENTS
1541         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1542                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1543 #endif
1544         uprobe_buffer_put(ucb);
1545         return 0;
1546 }
1547
1548 static struct trace_event_functions uprobe_funcs = {
1549         .trace          = print_uprobe_event
1550 };
1551
1552 static struct trace_event_fields uprobe_fields_array[] = {
1553         { .type = TRACE_FUNCTION_TYPE,
1554           .define_fields = uprobe_event_define_fields },
1555         {}
1556 };
1557
1558 static inline void init_trace_event_call(struct trace_uprobe *tu)
1559 {
1560         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1561         call->event.funcs = &uprobe_funcs;
1562         call->class->fields_array = uprobe_fields_array;
1563
1564         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1565         call->class->reg = trace_uprobe_register;
1566 }
1567
1568 static int register_uprobe_event(struct trace_uprobe *tu)
1569 {
1570         init_trace_event_call(tu);
1571
1572         return trace_probe_register_event_call(&tu->tp);
1573 }
1574
1575 static int unregister_uprobe_event(struct trace_uprobe *tu)
1576 {
1577         return trace_probe_unregister_event_call(&tu->tp);
1578 }
1579
1580 #ifdef CONFIG_PERF_EVENTS
1581 struct trace_event_call *
1582 create_local_trace_uprobe(char *name, unsigned long offs,
1583                           unsigned long ref_ctr_offset, bool is_return)
1584 {
1585         enum probe_print_type ptype;
1586         struct trace_uprobe *tu;
1587         struct path path;
1588         int ret;
1589
1590         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1591         if (ret)
1592                 return ERR_PTR(ret);
1593
1594         if (!d_is_reg(path.dentry)) {
1595                 path_put(&path);
1596                 return ERR_PTR(-EINVAL);
1597         }
1598
1599         /*
1600          * local trace_kprobes are not added to dyn_event, so they are never
1601          * searched in find_trace_kprobe(). Therefore, there is no concern of
1602          * duplicated name "DUMMY_EVENT" here.
1603          */
1604         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1605                                 is_return);
1606
1607         if (IS_ERR(tu)) {
1608                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1609                         (int)PTR_ERR(tu));
1610                 path_put(&path);
1611                 return ERR_CAST(tu);
1612         }
1613
1614         tu->offset = offs;
1615         tu->path = path;
1616         tu->ref_ctr_offset = ref_ctr_offset;
1617         tu->filename = kstrdup(name, GFP_KERNEL);
1618         if (!tu->filename) {
1619                 ret = -ENOMEM;
1620                 goto error;
1621         }
1622
1623         init_trace_event_call(tu);
1624
1625         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1626         if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1627                 ret = -ENOMEM;
1628                 goto error;
1629         }
1630
1631         return trace_probe_event_call(&tu->tp);
1632 error:
1633         free_trace_uprobe(tu);
1634         return ERR_PTR(ret);
1635 }
1636
1637 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1638 {
1639         struct trace_uprobe *tu;
1640
1641         tu = trace_uprobe_primary_from_call(event_call);
1642
1643         free_trace_uprobe(tu);
1644 }
1645 #endif /* CONFIG_PERF_EVENTS */
1646
1647 /* Make a trace interface for controlling probe points */
1648 static __init int init_uprobe_trace(void)
1649 {
1650         int ret;
1651
1652         ret = dyn_event_register(&trace_uprobe_ops);
1653         if (ret)
1654                 return ret;
1655
1656         ret = tracing_init_dentry();
1657         if (ret)
1658                 return 0;
1659
1660         trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1661                                     NULL, &uprobe_events_ops);
1662         /* Profile interface */
1663         trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1664                                     NULL, &uprobe_profile_ops);
1665         return 0;
1666 }
1667
1668 fs_initcall(init_uprobe_trace);