kernel/trace/trace_uprobe.c

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