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/ctype.h>
  11 #include <linux/module.h>
  12 #include <linux/uaccess.h>
  13 #include <linux/uprobes.h>
  14 #include <linux/namei.h>
  15 #include <linux/string.h>
  16 #include <linux/rculist.h>
  17
  18 #include "trace_dynevent.h"
  19 #include "trace_probe.h"
  20 #include "trace_probe_tmpl.h"
  21
  22 #define UPROBE_EVENT_SYSTEM     "uprobes"
  23
  24 struct uprobe_trace_entry_head {
  25         struct trace_entry      ent;
  26         unsigned long           vaddr[];
  27 };
  28
  29 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  30         (sizeof(struct uprobe_trace_entry_head) +       \
  31          sizeof(unsigned long) * (is_return ? 2 : 1))
  32
  33 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  34         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  35
  36 struct trace_uprobe_filter {
  37         rwlock_t                rwlock;
  38         int                     nr_systemwide;
  39         struct list_head        perf_events;
  40 };
  41
  42 static int trace_uprobe_create(int argc, const char **argv);
  43 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
  44 static int trace_uprobe_release(struct dyn_event *ev);
  45 static bool trace_uprobe_is_busy(struct dyn_event *ev);
  46 static bool trace_uprobe_match(const char *system, const char *event,
  47                                struct dyn_event *ev);
  48
  49 static struct dyn_event_operations trace_uprobe_ops = {
  50         .create = trace_uprobe_create,
  51         .show = trace_uprobe_show,
  52         .is_busy = trace_uprobe_is_busy,
  53         .free = trace_uprobe_release,
  54         .match = trace_uprobe_match,
  55 };
  56
  57 /*
  58  * uprobe event core functions
  59  */
  60 struct trace_uprobe {
  61         struct dyn_event                devent;
  62         struct trace_uprobe_filter      filter;
  63         struct uprobe_consumer          consumer;
  64         struct path                     path;
  65         struct inode                    *inode;
  66         char                            *filename;
  67         unsigned long                   offset;
  68         unsigned long                   ref_ctr_offset;
  69         unsigned long                   nhit;
  70         struct trace_probe              tp;
  71 };
  72
  73 static bool is_trace_uprobe(struct dyn_event *ev)
  74 {
  75         return ev->ops == &trace_uprobe_ops;
  76 }
  77
  78 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  79 {
  80         return container_of(ev, struct trace_uprobe, devent);
  81 }
  82
  83 /**
  84  * for_each_trace_uprobe - iterate over the trace_uprobe list
  85  * @pos:        the struct trace_uprobe * for each entry
  86  * @dpos:       the struct dyn_event * to use as a loop cursor
  87  */
  88 #define for_each_trace_uprobe(pos, dpos)        \
  89         for_each_dyn_event(dpos)                \
  90                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  91
  92 #define SIZEOF_TRACE_UPROBE(n)                          \
  93         (offsetof(struct trace_uprobe, tp.args) +       \
  94         (sizeof(struct probe_arg) * (n)))
  95
  96 static int register_uprobe_event(struct trace_uprobe *tu);
  97 static int unregister_uprobe_event(struct trace_uprobe *tu);
  98
  99 struct uprobe_dispatch_data {
 100         struct trace_uprobe     *tu;
 101         unsigned long           bp_addr;
 102 };
 103
 104 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
 105 static int uretprobe_dispatcher(struct uprobe_consumer *con,
 106                                 unsigned long func, struct pt_regs *regs);
 107
 108 #ifdef CONFIG_STACK_GROWSUP
 109 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 110 {
 111         return addr - (n * sizeof(long));
 112 }
 113 #else
 114 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
 115 {
 116         return addr + (n * sizeof(long));
 117 }
 118 #endif
 119
 120 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
 121 {
 122         unsigned long ret;
 123         unsigned long addr = user_stack_pointer(regs);
 124
 125         addr = adjust_stack_addr(addr, n);
 126
 127         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 128                 return 0;
 129
 130         return ret;
 131 }
 132
 133 /*
 134  * Uprobes-specific fetch functions
 135  */
 136 static nokprobe_inline int
 137 probe_mem_read(void *dest, void *src, size_t size)
 138 {
 139         void __user *vaddr = (void __force __user *)src;
 140
 141         return copy_from_user(dest, vaddr, size) ? -EFAULT : 0;
 142 }
 143
 144 static nokprobe_inline int
 145 probe_mem_read_user(void *dest, void *src, size_t size)
 146 {
 147         return probe_mem_read(dest, src, size);
 148 }
 149
 150 /*
 151  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 152  * length and relative data location.
 153  */
 154 static nokprobe_inline int
 155 fetch_store_string(unsigned long addr, void *dest, void *base)
 156 {
 157         long ret;
 158         u32 loc = *(u32 *)dest;
 159         int maxlen  = get_loc_len(loc);
 160         u8 *dst = get_loc_data(dest, base);
 161         void __user *src = (void __force __user *) addr;
 162
 163         if (unlikely(!maxlen))
 164                 return -ENOMEM;
 165
 166         if (addr == FETCH_TOKEN_COMM)
 167                 ret = strlcpy(dst, current->comm, maxlen);
 168         else
 169                 ret = strncpy_from_user(dst, src, maxlen);
 170         if (ret >= 0) {
 171                 if (ret == maxlen)
 172                         dst[ret - 1] = '\0';
 173                 else
 174                         /*
 175                          * Include the terminating null byte. In this case it
 176                          * was copied by strncpy_from_user but not accounted
 177                          * for in ret.
 178                          */
 179                         ret++;
 180                 *(u32 *)dest = make_data_loc(ret, (void *)dst - base);
 181         }
 182
 183         return ret;
 184 }
 185
 186 static nokprobe_inline int
 187 fetch_store_string_user(unsigned long addr, void *dest, void *base)
 188 {
 189         return fetch_store_string(addr, dest, base);
 190 }
 191
 192 /* Return the length of string -- including null terminal byte */
 193 static nokprobe_inline int
 194 fetch_store_strlen(unsigned long addr)
 195 {
 196         int len;
 197         void __user *vaddr = (void __force __user *) addr;
 198
 199         if (addr == FETCH_TOKEN_COMM)
 200                 len = strlen(current->comm) + 1;
 201         else
 202                 len = strnlen_user(vaddr, MAX_STRING_SIZE);
 203
 204         return (len > MAX_STRING_SIZE) ? 0 : len;
 205 }
 206
 207 static nokprobe_inline int
 208 fetch_store_strlen_user(unsigned long addr)
 209 {
 210         return fetch_store_strlen(addr);
 211 }
 212
 213 static unsigned long translate_user_vaddr(unsigned long file_offset)
 214 {
 215         unsigned long base_addr;
 216         struct uprobe_dispatch_data *udd;
 217
 218         udd = (void *) current->utask->vaddr;
 219
 220         base_addr = udd->bp_addr - udd->tu->offset;
 221         return base_addr + file_offset;
 222 }
 223
 224 /* Note that we don't verify it, since the code does not come from user space */
 225 static int
 226 process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 227                    void *base)
 228 {
 229         unsigned long val;
 230
 231         /* 1st stage: get value from context */
 232         switch (code->op) {
 233         case FETCH_OP_REG:
 234                 val = regs_get_register(regs, code->param);
 235                 break;
 236         case FETCH_OP_STACK:
 237                 val = get_user_stack_nth(regs, code->param);
 238                 break;
 239         case FETCH_OP_STACKP:
 240                 val = user_stack_pointer(regs);
 241                 break;
 242         case FETCH_OP_RETVAL:
 243                 val = regs_return_value(regs);
 244                 break;
 245         case FETCH_OP_IMM:
 246                 val = code->immediate;
 247                 break;
 248         case FETCH_OP_COMM:
 249                 val = FETCH_TOKEN_COMM;
 250                 break;
 251         case FETCH_OP_FOFFS:
 252                 val = translate_user_vaddr(code->immediate);
 253                 break;
 254         default:
 255                 return -EILSEQ;
 256         }
 257         code++;
 258
 259         return process_fetch_insn_bottom(code, val, dest, base);
 260 }
 261 NOKPROBE_SYMBOL(process_fetch_insn)
 262
 263 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 264 {
 265         rwlock_init(&filter->rwlock);
 266         filter->nr_systemwide = 0;
 267         INIT_LIST_HEAD(&filter->perf_events);
 268 }
 269
 270 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 271 {
 272         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 273 }
 274
 275 static inline bool is_ret_probe(struct trace_uprobe *tu)
 276 {
 277         return tu->consumer.ret_handler != NULL;
 278 }
 279
 280 static bool trace_uprobe_is_busy(struct dyn_event *ev)
 281 {
 282         struct trace_uprobe *tu = to_trace_uprobe(ev);
 283
 284         return trace_probe_is_enabled(&tu->tp);
 285 }
 286
 287 static bool trace_uprobe_match(const char *system, const char *event,
 288                                struct dyn_event *ev)
 289 {
 290         struct trace_uprobe *tu = to_trace_uprobe(ev);
 291
 292         return strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 293             (!system || strcmp(trace_probe_group_name(&tu->tp), system) == 0);
 294 }
 295
 296 static nokprobe_inline struct trace_uprobe *
 297 trace_uprobe_primary_from_call(struct trace_event_call *call)
 298 {
 299         struct trace_probe *tp;
 300
 301         tp = trace_probe_primary_from_call(call);
 302         if (WARN_ON_ONCE(!tp))
 303                 return NULL;
 304
 305         return container_of(tp, struct trace_uprobe, tp);
 306 }
 307
 308 /*
 309  * Allocate new trace_uprobe and initialize it (including uprobes).
 310  */
 311 static struct trace_uprobe *
 312 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 313 {
 314         struct trace_uprobe *tu;
 315         int ret;
 316
 317         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 318         if (!tu)
 319                 return ERR_PTR(-ENOMEM);
 320
 321         ret = trace_probe_init(&tu->tp, event, group);
 322         if (ret < 0)
 323                 goto error;
 324
 325         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 326         tu->consumer.handler = uprobe_dispatcher;
 327         if (is_ret)
 328                 tu->consumer.ret_handler = uretprobe_dispatcher;
 329         init_trace_uprobe_filter(&tu->filter);
 330         return tu;
 331
 332 error:
 333         kfree(tu);
 334
 335         return ERR_PTR(ret);
 336 }
 337
 338 static void free_trace_uprobe(struct trace_uprobe *tu)
 339 {
 340         if (!tu)
 341                 return;
 342
 343         path_put(&tu->path);
 344         trace_probe_cleanup(&tu->tp);
 345         kfree(tu->filename);
 346         kfree(tu);
 347 }
 348
 349 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 350 {
 351         struct dyn_event *pos;
 352         struct trace_uprobe *tu;
 353
 354         for_each_trace_uprobe(tu, pos)
 355                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 356                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 357                         return tu;
 358
 359         return NULL;
 360 }
 361
 362 /* Unregister a trace_uprobe and probe_event */
 363 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 364 {
 365         int ret;
 366
 367         ret = unregister_uprobe_event(tu);
 368         if (ret)
 369                 return ret;
 370
 371         dyn_event_remove(&tu->devent);
 372         free_trace_uprobe(tu);
 373         return 0;
 374 }
 375
 376 /*
 377  * Uprobe with multiple reference counter is not allowed. i.e.
 378  * If inode and offset matches, reference counter offset *must*
 379  * match as well. Though, there is one exception: If user is
 380  * replacing old trace_uprobe with new one(same group/event),
 381  * then we allow same uprobe with new reference counter as far
 382  * as the new one does not conflict with any other existing
 383  * ones.
 384  */
 385 static struct trace_uprobe *find_old_trace_uprobe(struct trace_uprobe *new)
 386 {
 387         struct dyn_event *pos;
 388         struct trace_uprobe *tmp, *old = NULL;
 389         struct inode *new_inode = d_real_inode(new->path.dentry);
 390
 391         old = find_probe_event(trace_probe_name(&new->tp),
 392                                 trace_probe_group_name(&new->tp));
 393
 394         for_each_trace_uprobe(tmp, pos) {
 395                 if ((old ? old != tmp : true) &&
 396                     new_inode == d_real_inode(tmp->path.dentry) &&
 397                     new->offset == tmp->offset &&
 398                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 399                         pr_warn("Reference counter offset mismatch.");
 400                         return ERR_PTR(-EINVAL);
 401                 }
 402         }
 403         return old;
 404 }
 405
 406 /* Register a trace_uprobe and probe_event */
 407 static int register_trace_uprobe(struct trace_uprobe *tu)
 408 {
 409         struct trace_uprobe *old_tu;
 410         int ret;
 411
 412         mutex_lock(&event_mutex);
 413
 414         /* register as an event */
 415         old_tu = find_old_trace_uprobe(tu);
 416         if (IS_ERR(old_tu)) {
 417                 ret = PTR_ERR(old_tu);
 418                 goto end;
 419         }
 420
 421         if (old_tu) {
 422                 /* delete old event */
 423                 ret = unregister_trace_uprobe(old_tu);
 424                 if (ret)
 425                         goto end;
 426         }
 427
 428         ret = register_uprobe_event(tu);
 429         if (ret) {
 430                 pr_warn("Failed to register probe event(%d)\n", ret);
 431                 goto end;
 432         }
 433
 434         dyn_event_add(&tu->devent);
 435
 436 end:
 437         mutex_unlock(&event_mutex);
 438
 439         return ret;
 440 }
 441
 442 /*
 443  * Argument syntax:
 444  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 445  */
 446 static int trace_uprobe_create(int argc, const char **argv)
 447 {
 448         struct trace_uprobe *tu;
 449         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 450         char *arg, *filename, *rctr, *rctr_end, *tmp;
 451         char buf[MAX_EVENT_NAME_LEN];
 452         struct path path;
 453         unsigned long offset, ref_ctr_offset;
 454         bool is_return = false;
 455         int i, ret;
 456
 457         ret = 0;
 458         ref_ctr_offset = 0;
 459
 460         switch (argv[0][0]) {
 461         case 'r':
 462                 is_return = true;
 463                 break;
 464         case 'p':
 465                 break;
 466         default:
 467                 return -ECANCELED;
 468         }
 469
 470         if (argc < 2)
 471                 return -ECANCELED;
 472
 473         if (argv[0][1] == ':')
 474                 event = &argv[0][2];
 475
 476         if (!strchr(argv[1], '/'))
 477                 return -ECANCELED;
 478
 479         filename = kstrdup(argv[1], GFP_KERNEL);
 480         if (!filename)
 481                 return -ENOMEM;
 482
 483         /* Find the last occurrence, in case the path contains ':' too. */
 484         arg = strrchr(filename, ':');
 485         if (!arg || !isdigit(arg[1])) {
 486                 kfree(filename);
 487                 return -ECANCELED;
 488         }
 489
 490         trace_probe_log_init("trace_uprobe", argc, argv);
 491         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 492
 493         *arg++ = '\0';
 494         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 495         if (ret) {
 496                 trace_probe_log_err(0, FILE_NOT_FOUND);
 497                 kfree(filename);
 498                 trace_probe_log_clear();
 499                 return ret;
 500         }
 501         if (!d_is_reg(path.dentry)) {
 502                 trace_probe_log_err(0, NO_REGULAR_FILE);
 503                 ret = -EINVAL;
 504                 goto fail_address_parse;
 505         }
 506
 507         /* Parse reference counter offset if specified. */
 508         rctr = strchr(arg, '(');
 509         if (rctr) {
 510                 rctr_end = strchr(rctr, ')');
 511                 if (!rctr_end) {
 512                         ret = -EINVAL;
 513                         rctr_end = rctr + strlen(rctr);
 514                         trace_probe_log_err(rctr_end - filename,
 515                                             REFCNT_OPEN_BRACE);
 516                         goto fail_address_parse;
 517                 } else if (rctr_end[1] != '\0') {
 518                         ret = -EINVAL;
 519                         trace_probe_log_err(rctr_end + 1 - filename,
 520                                             BAD_REFCNT_SUFFIX);
 521                         goto fail_address_parse;
 522                 }
 523
 524                 *rctr++ = '\0';
 525                 *rctr_end = '\0';
 526                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 527                 if (ret) {
 528                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 529                         goto fail_address_parse;
 530                 }
 531         }
 532
 533         /* Parse uprobe offset. */
 534         ret = kstrtoul(arg, 0, &offset);
 535         if (ret) {
 536                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 537                 goto fail_address_parse;
 538         }
 539
 540         /* setup a probe */
 541         trace_probe_log_set_index(0);
 542         if (event) {
 543                 ret = traceprobe_parse_event_name(&event, &group, buf,
 544                                                   event - argv[0]);
 545                 if (ret)
 546                         goto fail_address_parse;
 547         } else {
 548                 char *tail;
 549                 char *ptr;
 550
 551                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 552                 if (!tail) {
 553                         ret = -ENOMEM;
 554                         goto fail_address_parse;
 555                 }
 556
 557                 ptr = strpbrk(tail, ".-_");
 558                 if (ptr)
 559                         *ptr = '\0';
 560
 561                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 562                 event = buf;
 563                 kfree(tail);
 564         }
 565
 566         argc -= 2;
 567         argv += 2;
 568
 569         tu = alloc_trace_uprobe(group, event, argc, is_return);
 570         if (IS_ERR(tu)) {
 571                 ret = PTR_ERR(tu);
 572                 /* This must return -ENOMEM otherwise there is a bug */
 573                 WARN_ON_ONCE(ret != -ENOMEM);
 574                 goto fail_address_parse;
 575         }
 576         tu->offset = offset;
 577         tu->ref_ctr_offset = ref_ctr_offset;
 578         tu->path = path;
 579         tu->filename = filename;
 580
 581         /* parse arguments */
 582         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 583                 tmp = kstrdup(argv[i], GFP_KERNEL);
 584                 if (!tmp) {
 585                         ret = -ENOMEM;
 586                         goto error;
 587                 }
 588
 589                 trace_probe_log_set_index(i + 2);
 590                 ret = traceprobe_parse_probe_arg(&tu->tp, i, tmp,
 591                                         is_return ? TPARG_FL_RETURN : 0);
 592                 kfree(tmp);
 593                 if (ret)
 594                         goto error;
 595         }
 596
 597         ret = traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu));
 598         if (ret < 0)
 599                 goto error;
 600
 601         ret = register_trace_uprobe(tu);
 602         if (!ret)
 603                 goto out;
 604
 605 error:
 606         free_trace_uprobe(tu);
 607 out:
 608         trace_probe_log_clear();
 609         return ret;
 610
 611 fail_address_parse:
 612         trace_probe_log_clear();
 613         path_put(&path);
 614         kfree(filename);
 615
 616         return ret;
 617 }
 618
 619 static int create_or_delete_trace_uprobe(int argc, char **argv)
 620 {
 621         int ret;
 622
 623         if (argv[0][0] == '-')
 624                 return dyn_event_release(argc, argv, &trace_uprobe_ops);
 625
 626         ret = trace_uprobe_create(argc, (const char **)argv);
 627         return ret == -ECANCELED ? -EINVAL : ret;
 628 }
 629
 630 static int trace_uprobe_release(struct dyn_event *ev)
 631 {
 632         struct trace_uprobe *tu = to_trace_uprobe(ev);
 633
 634         return unregister_trace_uprobe(tu);
 635 }
 636
 637 /* Probes listing interfaces */
 638 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 639 {
 640         struct trace_uprobe *tu = to_trace_uprobe(ev);
 641         char c = is_ret_probe(tu) ? 'r' : 'p';
 642         int i;
 643
 644         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 645                         trace_probe_name(&tu->tp), tu->filename,
 646                         (int)(sizeof(void *) * 2), tu->offset);
 647
 648         if (tu->ref_ctr_offset)
 649                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 650
 651         for (i = 0; i < tu->tp.nr_args; i++)
 652                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 653
 654         seq_putc(m, '\n');
 655         return 0;
 656 }
 657
 658 static int probes_seq_show(struct seq_file *m, void *v)
 659 {
 660         struct dyn_event *ev = v;
 661
 662         if (!is_trace_uprobe(ev))
 663                 return 0;
 664
 665         return trace_uprobe_show(m, ev);
 666 }
 667
 668 static const struct seq_operations probes_seq_op = {
 669         .start  = dyn_event_seq_start,
 670         .next   = dyn_event_seq_next,
 671         .stop   = dyn_event_seq_stop,
 672         .show   = probes_seq_show
 673 };
 674
 675 static int probes_open(struct inode *inode, struct file *file)
 676 {
 677         int ret;
 678
 679         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 680                 ret = dyn_events_release_all(&trace_uprobe_ops);
 681                 if (ret)
 682                         return ret;
 683         }
 684
 685         return seq_open(file, &probes_seq_op);
 686 }
 687
 688 static ssize_t probes_write(struct file *file, const char __user *buffer,
 689                             size_t count, loff_t *ppos)
 690 {
 691         return trace_parse_run_command(file, buffer, count, ppos,
 692                                         create_or_delete_trace_uprobe);
 693 }
 694
 695 static const struct file_operations uprobe_events_ops = {
 696         .owner          = THIS_MODULE,
 697         .open           = probes_open,
 698         .read           = seq_read,
 699         .llseek         = seq_lseek,
 700         .release        = seq_release,
 701         .write          = probes_write,
 702 };
 703
 704 /* Probes profiling interfaces */
 705 static int probes_profile_seq_show(struct seq_file *m, void *v)
 706 {
 707         struct dyn_event *ev = v;
 708         struct trace_uprobe *tu;
 709
 710         if (!is_trace_uprobe(ev))
 711                 return 0;
 712
 713         tu = to_trace_uprobe(ev);
 714         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 715                         trace_probe_name(&tu->tp), tu->nhit);
 716         return 0;
 717 }
 718
 719 static const struct seq_operations profile_seq_op = {
 720         .start  = dyn_event_seq_start,
 721         .next   = dyn_event_seq_next,
 722         .stop   = dyn_event_seq_stop,
 723         .show   = probes_profile_seq_show
 724 };
 725
 726 static int profile_open(struct inode *inode, struct file *file)
 727 {
 728         return seq_open(file, &profile_seq_op);
 729 }
 730
 731 static const struct file_operations uprobe_profile_ops = {
 732         .owner          = THIS_MODULE,
 733         .open           = profile_open,
 734         .read           = seq_read,
 735         .llseek         = seq_lseek,
 736         .release        = seq_release,
 737 };
 738
 739 struct uprobe_cpu_buffer {
 740         struct mutex mutex;
 741         void *buf;
 742 };
 743 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 744 static int uprobe_buffer_refcnt;
 745
 746 static int uprobe_buffer_init(void)
 747 {
 748         int cpu, err_cpu;
 749
 750         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 751         if (uprobe_cpu_buffer == NULL)
 752                 return -ENOMEM;
 753
 754         for_each_possible_cpu(cpu) {
 755                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 756                                                   GFP_KERNEL, 0);
 757                 if (p == NULL) {
 758                         err_cpu = cpu;
 759                         goto err;
 760                 }
 761                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 762                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 763         }
 764
 765         return 0;
 766
 767 err:
 768         for_each_possible_cpu(cpu) {
 769                 if (cpu == err_cpu)
 770                         break;
 771                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 772         }
 773
 774         free_percpu(uprobe_cpu_buffer);
 775         return -ENOMEM;
 776 }
 777
 778 static int uprobe_buffer_enable(void)
 779 {
 780         int ret = 0;
 781
 782         BUG_ON(!mutex_is_locked(&event_mutex));
 783
 784         if (uprobe_buffer_refcnt++ == 0) {
 785                 ret = uprobe_buffer_init();
 786                 if (ret < 0)
 787                         uprobe_buffer_refcnt--;
 788         }
 789
 790         return ret;
 791 }
 792
 793 static void uprobe_buffer_disable(void)
 794 {
 795         int cpu;
 796
 797         BUG_ON(!mutex_is_locked(&event_mutex));
 798
 799         if (--uprobe_buffer_refcnt == 0) {
 800                 for_each_possible_cpu(cpu)
 801                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 802                                                              cpu)->buf);
 803
 804                 free_percpu(uprobe_cpu_buffer);
 805                 uprobe_cpu_buffer = NULL;
 806         }
 807 }
 808
 809 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 810 {
 811         struct uprobe_cpu_buffer *ucb;
 812         int cpu;
 813
 814         cpu = raw_smp_processor_id();
 815         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 816
 817         /*
 818          * Use per-cpu buffers for fastest access, but we might migrate
 819          * so the mutex makes sure we have sole access to it.
 820          */
 821         mutex_lock(&ucb->mutex);
 822
 823         return ucb;
 824 }
 825
 826 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 827 {
 828         mutex_unlock(&ucb->mutex);
 829 }
 830
 831 static void __uprobe_trace_func(struct trace_uprobe *tu,
 832                                 unsigned long func, struct pt_regs *regs,
 833                                 struct uprobe_cpu_buffer *ucb, int dsize,
 834                                 struct trace_event_file *trace_file)
 835 {
 836         struct uprobe_trace_entry_head *entry;
 837         struct ring_buffer_event *event;
 838         struct ring_buffer *buffer;
 839         void *data;
 840         int size, esize;
 841         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
 842
 843         WARN_ON(call != trace_file->event_call);
 844
 845         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 846                 return;
 847
 848         if (trace_trigger_soft_disabled(trace_file))
 849                 return;
 850
 851         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 852         size = esize + tu->tp.size + dsize;
 853         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 854                                                 call->event.type, size, 0, 0);
 855         if (!event)
 856                 return;
 857
 858         entry = ring_buffer_event_data(event);
 859         if (is_ret_probe(tu)) {
 860                 entry->vaddr[0] = func;
 861                 entry->vaddr[1] = instruction_pointer(regs);
 862                 data = DATAOF_TRACE_ENTRY(entry, true);
 863         } else {
 864                 entry->vaddr[0] = instruction_pointer(regs);
 865                 data = DATAOF_TRACE_ENTRY(entry, false);
 866         }
 867
 868         memcpy(data, ucb->buf, tu->tp.size + dsize);
 869
 870         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 871 }
 872
 873 /* uprobe handler */
 874 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 875                              struct uprobe_cpu_buffer *ucb, int dsize)
 876 {
 877         struct event_file_link *link;
 878
 879         if (is_ret_probe(tu))
 880                 return 0;
 881
 882         rcu_read_lock();
 883         trace_probe_for_each_link_rcu(link, &tu->tp)
 884                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 885         rcu_read_unlock();
 886
 887         return 0;
 888 }
 889
 890 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 891                                  struct pt_regs *regs,
 892                                  struct uprobe_cpu_buffer *ucb, int dsize)
 893 {
 894         struct event_file_link *link;
 895
 896         rcu_read_lock();
 897         trace_probe_for_each_link_rcu(link, &tu->tp)
 898                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 899         rcu_read_unlock();
 900 }
 901
 902 /* Event entry printers */
 903 static enum print_line_t
 904 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 905 {
 906         struct uprobe_trace_entry_head *entry;
 907         struct trace_seq *s = &iter->seq;
 908         struct trace_uprobe *tu;
 909         u8 *data;
 910
 911         entry = (struct uprobe_trace_entry_head *)iter->ent;
 912         tu = trace_uprobe_primary_from_call(
 913                 container_of(event, struct trace_event_call, event));
 914         if (unlikely(!tu))
 915                 goto out;
 916
 917         if (is_ret_probe(tu)) {
 918                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 919                                  trace_probe_name(&tu->tp),
 920                                  entry->vaddr[1], entry->vaddr[0]);
 921                 data = DATAOF_TRACE_ENTRY(entry, true);
 922         } else {
 923                 trace_seq_printf(s, "%s: (0x%lx)",
 924                                  trace_probe_name(&tu->tp),
 925                                  entry->vaddr[0]);
 926                 data = DATAOF_TRACE_ENTRY(entry, false);
 927         }
 928
 929         if (print_probe_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
 930                 goto out;
 931
 932         trace_seq_putc(s, '\n');
 933
 934  out:
 935         return trace_handle_return(s);
 936 }
 937
 938 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 939                                 enum uprobe_filter_ctx ctx,
 940                                 struct mm_struct *mm);
 941
 942 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
 943 {
 944         int ret;
 945
 946         tu->consumer.filter = filter;
 947         tu->inode = d_real_inode(tu->path.dentry);
 948
 949         if (tu->ref_ctr_offset)
 950                 ret = uprobe_register_refctr(tu->inode, tu->offset,
 951                                 tu->ref_ctr_offset, &tu->consumer);
 952         else
 953                 ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 954
 955         if (ret)
 956                 tu->inode = NULL;
 957
 958         return ret;
 959 }
 960
 961 static void __probe_event_disable(struct trace_probe *tp)
 962 {
 963         struct trace_probe *pos;
 964         struct trace_uprobe *tu;
 965
 966         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 967                 tu = container_of(pos, struct trace_uprobe, tp);
 968                 if (!tu->inode)
 969                         continue;
 970
 971                 WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 972
 973                 uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 974                 tu->inode = NULL;
 975         }
 976 }
 977
 978 static int probe_event_enable(struct trace_event_call *call,
 979                         struct trace_event_file *file, filter_func_t filter)
 980 {
 981         struct trace_probe *pos, *tp;
 982         struct trace_uprobe *tu;
 983         bool enabled;
 984         int ret;
 985
 986         tp = trace_probe_primary_from_call(call);
 987         if (WARN_ON_ONCE(!tp))
 988                 return -ENODEV;
 989         enabled = trace_probe_is_enabled(tp);
 990
 991         /* This may also change "enabled" state */
 992         if (file) {
 993                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
 994                         return -EINTR;
 995
 996                 ret = trace_probe_add_file(tp, file);
 997                 if (ret < 0)
 998                         return ret;
 999         } else {
1000                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1001                         return -EINTR;
1002
1003                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1004         }
1005
1006         tu = container_of(tp, struct trace_uprobe, tp);
1007         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
1008
1009         if (enabled)
1010                 return 0;
1011
1012         ret = uprobe_buffer_enable();
1013         if (ret)
1014                 goto err_flags;
1015
1016         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1017                 tu = container_of(pos, struct trace_uprobe, tp);
1018                 ret = trace_uprobe_enable(tu, filter);
1019                 if (ret) {
1020                         __probe_event_disable(tp);
1021                         goto err_buffer;
1022                 }
1023         }
1024
1025         return 0;
1026
1027  err_buffer:
1028         uprobe_buffer_disable();
1029
1030  err_flags:
1031         if (file)
1032                 trace_probe_remove_file(tp, file);
1033         else
1034                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1035
1036         return ret;
1037 }
1038
1039 static void probe_event_disable(struct trace_event_call *call,
1040                                 struct trace_event_file *file)
1041 {
1042         struct trace_probe *tp;
1043
1044         tp = trace_probe_primary_from_call(call);
1045         if (WARN_ON_ONCE(!tp))
1046                 return;
1047
1048         if (!trace_probe_is_enabled(tp))
1049                 return;
1050
1051         if (file) {
1052                 if (trace_probe_remove_file(tp, file) < 0)
1053                         return;
1054
1055                 if (trace_probe_is_enabled(tp))
1056                         return;
1057         } else
1058                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1059
1060         __probe_event_disable(tp);
1061         uprobe_buffer_disable();
1062 }
1063
1064 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1065 {
1066         int ret, size;
1067         struct uprobe_trace_entry_head field;
1068         struct trace_uprobe *tu;
1069
1070         tu = trace_uprobe_primary_from_call(event_call);
1071         if (unlikely(!tu))
1072                 return -ENODEV;
1073
1074         if (is_ret_probe(tu)) {
1075                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1076                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1077                 size = SIZEOF_TRACE_ENTRY(true);
1078         } else {
1079                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1080                 size = SIZEOF_TRACE_ENTRY(false);
1081         }
1082
1083         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1084 }
1085
1086 #ifdef CONFIG_PERF_EVENTS
1087 static bool
1088 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1089 {
1090         struct perf_event *event;
1091
1092         if (filter->nr_systemwide)
1093                 return true;
1094
1095         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1096                 if (event->hw.target->mm == mm)
1097                         return true;
1098         }
1099
1100         return false;
1101 }
1102
1103 static inline bool
1104 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1105 {
1106         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1107 }
1108
1109 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1110 {
1111         bool done;
1112
1113         write_lock(&tu->filter.rwlock);
1114         if (event->hw.target) {
1115                 list_del(&event->hw.tp_list);
1116                 done = tu->filter.nr_systemwide ||
1117                         (event->hw.target->flags & PF_EXITING) ||
1118                         uprobe_filter_event(tu, event);
1119         } else {
1120                 tu->filter.nr_systemwide--;
1121                 done = tu->filter.nr_systemwide;
1122         }
1123         write_unlock(&tu->filter.rwlock);
1124
1125         if (!done)
1126                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1127
1128         return 0;
1129 }
1130
1131 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1132 {
1133         bool done;
1134         int err;
1135
1136         write_lock(&tu->filter.rwlock);
1137         if (event->hw.target) {
1138                 /*
1139                  * event->parent != NULL means copy_process(), we can avoid
1140                  * uprobe_apply(). current->mm must be probed and we can rely
1141                  * on dup_mmap() which preserves the already installed bp's.
1142                  *
1143                  * attr.enable_on_exec means that exec/mmap will install the
1144                  * breakpoints we need.
1145                  */
1146                 done = tu->filter.nr_systemwide ||
1147                         event->parent || event->attr.enable_on_exec ||
1148                         uprobe_filter_event(tu, event);
1149                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1150         } else {
1151                 done = tu->filter.nr_systemwide;
1152                 tu->filter.nr_systemwide++;
1153         }
1154         write_unlock(&tu->filter.rwlock);
1155
1156         err = 0;
1157         if (!done) {
1158                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1159                 if (err)
1160                         uprobe_perf_close(tu, event);
1161         }
1162         return err;
1163 }
1164
1165 static int uprobe_perf_multi_call(struct trace_event_call *call,
1166                                   struct perf_event *event,
1167                 int (*op)(struct trace_uprobe *tu, struct perf_event *event))
1168 {
1169         struct trace_probe *pos, *tp;
1170         struct trace_uprobe *tu;
1171         int ret = 0;
1172
1173         tp = trace_probe_primary_from_call(call);
1174         if (WARN_ON_ONCE(!tp))
1175                 return -ENODEV;
1176
1177         list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
1178                 tu = container_of(pos, struct trace_uprobe, tp);
1179                 ret = op(tu, event);
1180                 if (ret)
1181                         break;
1182         }
1183
1184         return ret;
1185 }
1186 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1187                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1188 {
1189         struct trace_uprobe *tu;
1190         int ret;
1191
1192         tu = container_of(uc, struct trace_uprobe, consumer);
1193         read_lock(&tu->filter.rwlock);
1194         ret = __uprobe_perf_filter(&tu->filter, mm);
1195         read_unlock(&tu->filter.rwlock);
1196
1197         return ret;
1198 }
1199
1200 static void __uprobe_perf_func(struct trace_uprobe *tu,
1201                                unsigned long func, struct pt_regs *regs,
1202                                struct uprobe_cpu_buffer *ucb, int dsize)
1203 {
1204         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1205         struct uprobe_trace_entry_head *entry;
1206         struct hlist_head *head;
1207         void *data;
1208         int size, esize;
1209         int rctx;
1210
1211         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1212                 return;
1213
1214         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1215
1216         size = esize + tu->tp.size + dsize;
1217         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1218         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1219                 return;
1220
1221         preempt_disable();
1222         head = this_cpu_ptr(call->perf_events);
1223         if (hlist_empty(head))
1224                 goto out;
1225
1226         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1227         if (!entry)
1228                 goto out;
1229
1230         if (is_ret_probe(tu)) {
1231                 entry->vaddr[0] = func;
1232                 entry->vaddr[1] = instruction_pointer(regs);
1233                 data = DATAOF_TRACE_ENTRY(entry, true);
1234         } else {
1235                 entry->vaddr[0] = instruction_pointer(regs);
1236                 data = DATAOF_TRACE_ENTRY(entry, false);
1237         }
1238
1239         memcpy(data, ucb->buf, tu->tp.size + dsize);
1240
1241         if (size - esize > tu->tp.size + dsize) {
1242                 int len = tu->tp.size + dsize;
1243
1244                 memset(data + len, 0, size - esize - len);
1245         }
1246
1247         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1248                               head, NULL);
1249  out:
1250         preempt_enable();
1251 }
1252
1253 /* uprobe profile handler */
1254 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1255                             struct uprobe_cpu_buffer *ucb, int dsize)
1256 {
1257         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1258                 return UPROBE_HANDLER_REMOVE;
1259
1260         if (!is_ret_probe(tu))
1261                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1262         return 0;
1263 }
1264
1265 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1266                                 struct pt_regs *regs,
1267                                 struct uprobe_cpu_buffer *ucb, int dsize)
1268 {
1269         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1270 }
1271
1272 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1273                         const char **filename, u64 *probe_offset,
1274                         bool perf_type_tracepoint)
1275 {
1276         const char *pevent = trace_event_name(event->tp_event);
1277         const char *group = event->tp_event->class->system;
1278         struct trace_uprobe *tu;
1279
1280         if (perf_type_tracepoint)
1281                 tu = find_probe_event(pevent, group);
1282         else
1283                 tu = event->tp_event->data;
1284         if (!tu)
1285                 return -EINVAL;
1286
1287         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1288                                     : BPF_FD_TYPE_UPROBE;
1289         *filename = tu->filename;
1290         *probe_offset = tu->offset;
1291         return 0;
1292 }
1293 #endif  /* CONFIG_PERF_EVENTS */
1294
1295 static int
1296 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1297                       void *data)
1298 {
1299         struct trace_event_file *file = data;
1300
1301         switch (type) {
1302         case TRACE_REG_REGISTER:
1303                 return probe_event_enable(event, file, NULL);
1304
1305         case TRACE_REG_UNREGISTER:
1306                 probe_event_disable(event, file);
1307                 return 0;
1308
1309 #ifdef CONFIG_PERF_EVENTS
1310         case TRACE_REG_PERF_REGISTER:
1311                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1312
1313         case TRACE_REG_PERF_UNREGISTER:
1314                 probe_event_disable(event, NULL);
1315                 return 0;
1316
1317         case TRACE_REG_PERF_OPEN:
1318                 return uprobe_perf_multi_call(event, data, uprobe_perf_open);
1319
1320         case TRACE_REG_PERF_CLOSE:
1321                 return uprobe_perf_multi_call(event, data, uprobe_perf_close);
1322
1323 #endif
1324         default:
1325                 return 0;
1326         }
1327         return 0;
1328 }
1329
1330 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1331 {
1332         struct trace_uprobe *tu;
1333         struct uprobe_dispatch_data udd;
1334         struct uprobe_cpu_buffer *ucb;
1335         int dsize, esize;
1336         int ret = 0;
1337
1338
1339         tu = container_of(con, struct trace_uprobe, consumer);
1340         tu->nhit++;
1341
1342         udd.tu = tu;
1343         udd.bp_addr = instruction_pointer(regs);
1344
1345         current->utask->vaddr = (unsigned long) &udd;
1346
1347         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1348                 return 0;
1349
1350         dsize = __get_data_size(&tu->tp, regs);
1351         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1352
1353         ucb = uprobe_buffer_get();
1354         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1355
1356         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1357                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1358
1359 #ifdef CONFIG_PERF_EVENTS
1360         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1361                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1362 #endif
1363         uprobe_buffer_put(ucb);
1364         return ret;
1365 }
1366
1367 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1368                                 unsigned long func, struct pt_regs *regs)
1369 {
1370         struct trace_uprobe *tu;
1371         struct uprobe_dispatch_data udd;
1372         struct uprobe_cpu_buffer *ucb;
1373         int dsize, esize;
1374
1375         tu = container_of(con, struct trace_uprobe, consumer);
1376
1377         udd.tu = tu;
1378         udd.bp_addr = func;
1379
1380         current->utask->vaddr = (unsigned long) &udd;
1381
1382         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1383                 return 0;
1384
1385         dsize = __get_data_size(&tu->tp, regs);
1386         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1387
1388         ucb = uprobe_buffer_get();
1389         store_trace_args(ucb->buf, &tu->tp, regs, esize, dsize);
1390
1391         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1392                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1393
1394 #ifdef CONFIG_PERF_EVENTS
1395         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1396                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1397 #endif
1398         uprobe_buffer_put(ucb);
1399         return 0;
1400 }
1401
1402 static struct trace_event_functions uprobe_funcs = {
1403         .trace          = print_uprobe_event
1404 };
1405
1406 static inline void init_trace_event_call(struct trace_uprobe *tu)
1407 {
1408         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1409
1410         call->event.funcs = &uprobe_funcs;
1411         call->class->define_fields = uprobe_event_define_fields;
1412
1413         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1414         call->class->reg = trace_uprobe_register;
1415 }
1416
1417 static int register_uprobe_event(struct trace_uprobe *tu)
1418 {
1419         init_trace_event_call(tu);
1420
1421         return trace_probe_register_event_call(&tu->tp);
1422 }
1423
1424 static int unregister_uprobe_event(struct trace_uprobe *tu)
1425 {
1426         return trace_probe_unregister_event_call(&tu->tp);
1427 }
1428
1429 #ifdef CONFIG_PERF_EVENTS
1430 struct trace_event_call *
1431 create_local_trace_uprobe(char *name, unsigned long offs,
1432                           unsigned long ref_ctr_offset, bool is_return)
1433 {
1434         struct trace_uprobe *tu;
1435         struct path path;
1436         int ret;
1437
1438         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1439         if (ret)
1440                 return ERR_PTR(ret);
1441
1442         if (!d_is_reg(path.dentry)) {
1443                 path_put(&path);
1444                 return ERR_PTR(-EINVAL);
1445         }
1446
1447         /*
1448          * local trace_kprobes are not added to dyn_event, so they are never
1449          * searched in find_trace_kprobe(). Therefore, there is no concern of
1450          * duplicated name "DUMMY_EVENT" here.
1451          */
1452         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1453                                 is_return);
1454
1455         if (IS_ERR(tu)) {
1456                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1457                         (int)PTR_ERR(tu));
1458                 path_put(&path);
1459                 return ERR_CAST(tu);
1460         }
1461
1462         tu->offset = offs;
1463         tu->path = path;
1464         tu->ref_ctr_offset = ref_ctr_offset;
1465         tu->filename = kstrdup(name, GFP_KERNEL);
1466         init_trace_event_call(tu);
1467
1468         if (traceprobe_set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
1469                 ret = -ENOMEM;
1470                 goto error;
1471         }
1472
1473         return trace_probe_event_call(&tu->tp);
1474 error:
1475         free_trace_uprobe(tu);
1476         return ERR_PTR(ret);
1477 }
1478
1479 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1480 {
1481         struct trace_uprobe *tu;
1482
1483         tu = trace_uprobe_primary_from_call(event_call);
1484
1485         free_trace_uprobe(tu);
1486 }
1487 #endif /* CONFIG_PERF_EVENTS */
1488
1489 /* Make a trace interface for controling probe points */
1490 static __init int init_uprobe_trace(void)
1491 {
1492         struct dentry *d_tracer;
1493         int ret;
1494
1495         ret = dyn_event_register(&trace_uprobe_ops);
1496         if (ret)
1497                 return ret;
1498
1499         d_tracer = tracing_init_dentry();
1500         if (IS_ERR(d_tracer))
1501                 return 0;
1502
1503         trace_create_file("uprobe_events", 0644, d_tracer,
1504                                     NULL, &uprobe_events_ops);
1505         /* Profile interface */
1506         trace_create_file("uprobe_profile", 0444, d_tracer,
1507                                     NULL, &uprobe_profile_ops);
1508         return 0;
1509 }
1510
1511 fs_initcall(init_uprobe_trace);