kernel/trace/trace_uprobe.c

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