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/cleanup.h>
  12 #include <linux/ctype.h>
  13 #include <linux/filter.h>
  14 #include <linux/module.h>
  15 #include <linux/namei.h>
  16 #include <linux/percpu.h>
  17 #include <linux/rculist.h>
  18 #include <linux/security.h>
  19 #include <linux/string.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/uprobes.h>
  22
  23 #include "trace.h"
  24 #include "trace_dynevent.h"
  25 #include "trace_probe.h"
  26 #include "trace_probe_tmpl.h"
  27
  28 #define UPROBE_EVENT_SYSTEM     "uprobes"
  29
  30 struct uprobe_trace_entry_head {
  31         struct trace_entry      ent;
  32         unsigned long           vaddr[];
  33 };
  34
  35 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  36         (sizeof(struct uprobe_trace_entry_head) +       \
  37          sizeof(unsigned long) * (is_return ? 2 : 1))
  38
  39 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  40         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  41
  42 static int trace_uprobe_create(const char *raw_command);
  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                         int argc, const char **argv, 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 uprobe_consumer          consumer;
  63         struct path                     path;
  64         char                            *filename;
  65         struct uprobe                   *uprobe;
  66         unsigned long                   offset;
  67         unsigned long                   ref_ctr_offset;
  68         unsigned long __percpu          *nhits;
  69         struct trace_probe              tp;
  70 };
  71
  72 static bool is_trace_uprobe(struct dyn_event *ev)
  73 {
  74         return ev->ops == &trace_uprobe_ops;
  75 }
  76
  77 static struct trace_uprobe *to_trace_uprobe(struct dyn_event *ev)
  78 {
  79         return container_of(ev, struct trace_uprobe, devent);
  80 }
  81
  82 /**
  83  * for_each_trace_uprobe - iterate over the trace_uprobe list
  84  * @pos:        the struct trace_uprobe * for each entry
  85  * @dpos:       the struct dyn_event * to use as a loop cursor
  86  */
  87 #define for_each_trace_uprobe(pos, dpos)        \
  88         for_each_dyn_event(dpos)                \
  89                 if (is_trace_uprobe(dpos) && (pos = to_trace_uprobe(dpos)))
  90
  91 static int register_uprobe_event(struct trace_uprobe *tu);
  92 static int unregister_uprobe_event(struct trace_uprobe *tu);
  93
  94 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs,
  95                              __u64 *data);
  96 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  97                                 unsigned long func, struct pt_regs *regs,
  98                                 __u64 *data);
  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 = strscpy(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 *edata,
 220                    void *dest, 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         tu->nhits = alloc_percpu(unsigned long);
 346         if (!tu->nhits) {
 347                 ret = -ENOMEM;
 348                 goto error;
 349         }
 350
 351         ret = trace_probe_init(&tu->tp, event, group, true, nargs);
 352         if (ret < 0)
 353                 goto error;
 354
 355         dyn_event_init(&tu->devent, &trace_uprobe_ops);
 356         tu->consumer.handler = uprobe_dispatcher;
 357         if (is_ret)
 358                 tu->consumer.ret_handler = uretprobe_dispatcher;
 359         init_trace_uprobe_filter(tu->tp.event->filter);
 360         return tu;
 361
 362 error:
 363         free_percpu(tu->nhits);
 364         kfree(tu);
 365
 366         return ERR_PTR(ret);
 367 }
 368
 369 static void free_trace_uprobe(struct trace_uprobe *tu)
 370 {
 371         if (!tu)
 372                 return;
 373
 374         path_put(&tu->path);
 375         trace_probe_cleanup(&tu->tp);
 376         kfree(tu->filename);
 377         free_percpu(tu->nhits);
 378         kfree(tu);
 379 }
 380
 381 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 382 {
 383         struct dyn_event *pos;
 384         struct trace_uprobe *tu;
 385
 386         for_each_trace_uprobe(tu, pos)
 387                 if (strcmp(trace_probe_name(&tu->tp), event) == 0 &&
 388                     strcmp(trace_probe_group_name(&tu->tp), group) == 0)
 389                         return tu;
 390
 391         return NULL;
 392 }
 393
 394 /* Unregister a trace_uprobe and probe_event */
 395 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 396 {
 397         int ret;
 398
 399         if (trace_probe_has_sibling(&tu->tp))
 400                 goto unreg;
 401
 402         /* If there's a reference to the dynamic event */
 403         if (trace_event_dyn_busy(trace_probe_event_call(&tu->tp)))
 404                 return -EBUSY;
 405
 406         ret = unregister_uprobe_event(tu);
 407         if (ret)
 408                 return ret;
 409
 410 unreg:
 411         dyn_event_remove(&tu->devent);
 412         trace_probe_unlink(&tu->tp);
 413         free_trace_uprobe(tu);
 414         return 0;
 415 }
 416
 417 static bool trace_uprobe_has_same_uprobe(struct trace_uprobe *orig,
 418                                          struct trace_uprobe *comp)
 419 {
 420         struct trace_probe_event *tpe = orig->tp.event;
 421         struct inode *comp_inode = d_real_inode(comp->path.dentry);
 422         int i;
 423
 424         list_for_each_entry(orig, &tpe->probes, tp.list) {
 425                 if (comp_inode != d_real_inode(orig->path.dentry) ||
 426                     comp->offset != orig->offset)
 427                         continue;
 428
 429                 /*
 430                  * trace_probe_compare_arg_type() ensured that nr_args and
 431                  * each argument name and type are same. Let's compare comm.
 432                  */
 433                 for (i = 0; i < orig->tp.nr_args; i++) {
 434                         if (strcmp(orig->tp.args[i].comm,
 435                                    comp->tp.args[i].comm))
 436                                 break;
 437                 }
 438
 439                 if (i == orig->tp.nr_args)
 440                         return true;
 441         }
 442
 443         return false;
 444 }
 445
 446 static int append_trace_uprobe(struct trace_uprobe *tu, struct trace_uprobe *to)
 447 {
 448         int ret;
 449
 450         ret = trace_probe_compare_arg_type(&tu->tp, &to->tp);
 451         if (ret) {
 452                 /* Note that argument starts index = 2 */
 453                 trace_probe_log_set_index(ret + 1);
 454                 trace_probe_log_err(0, DIFF_ARG_TYPE);
 455                 return -EEXIST;
 456         }
 457         if (trace_uprobe_has_same_uprobe(to, tu)) {
 458                 trace_probe_log_set_index(0);
 459                 trace_probe_log_err(0, SAME_PROBE);
 460                 return -EEXIST;
 461         }
 462
 463         /* Append to existing event */
 464         ret = trace_probe_append(&tu->tp, &to->tp);
 465         if (!ret)
 466                 dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 467
 468         return ret;
 469 }
 470
 471 /*
 472  * Uprobe with multiple reference counter is not allowed. i.e.
 473  * If inode and offset matches, reference counter offset *must*
 474  * match as well. Though, there is one exception: If user is
 475  * replacing old trace_uprobe with new one(same group/event),
 476  * then we allow same uprobe with new reference counter as far
 477  * as the new one does not conflict with any other existing
 478  * ones.
 479  */
 480 static int validate_ref_ctr_offset(struct trace_uprobe *new)
 481 {
 482         struct dyn_event *pos;
 483         struct trace_uprobe *tmp;
 484         struct inode *new_inode = d_real_inode(new->path.dentry);
 485
 486         for_each_trace_uprobe(tmp, pos) {
 487                 if (new_inode == d_real_inode(tmp->path.dentry) &&
 488                     new->offset == tmp->offset &&
 489                     new->ref_ctr_offset != tmp->ref_ctr_offset) {
 490                         pr_warn("Reference counter offset mismatch.");
 491                         return -EINVAL;
 492                 }
 493         }
 494         return 0;
 495 }
 496
 497 /* Register a trace_uprobe and probe_event */
 498 static int register_trace_uprobe(struct trace_uprobe *tu)
 499 {
 500         struct trace_uprobe *old_tu;
 501         int ret;
 502
 503         guard(mutex)(&event_mutex);
 504
 505         ret = validate_ref_ctr_offset(tu);
 506         if (ret)
 507                 return ret;
 508
 509         /* register as an event */
 510         old_tu = find_probe_event(trace_probe_name(&tu->tp),
 511                                   trace_probe_group_name(&tu->tp));
 512         if (old_tu) {
 513                 if (is_ret_probe(tu) != is_ret_probe(old_tu)) {
 514                         trace_probe_log_set_index(0);
 515                         trace_probe_log_err(0, DIFF_PROBE_TYPE);
 516                         return -EEXIST;
 517                 }
 518                 return append_trace_uprobe(tu, old_tu);
 519         }
 520
 521         ret = register_uprobe_event(tu);
 522         if (ret) {
 523                 if (ret == -EEXIST) {
 524                         trace_probe_log_set_index(0);
 525                         trace_probe_log_err(0, EVENT_EXIST);
 526                 } else
 527                         pr_warn("Failed to register probe event(%d)\n", ret);
 528                 return ret;
 529         }
 530
 531         dyn_event_add(&tu->devent, trace_probe_event_call(&tu->tp));
 532
 533         return ret;
 534 }
 535
 536 /*
 537  * Argument syntax:
 538  *  - Add uprobe: p|r[:[GRP/][EVENT]] PATH:OFFSET[%return][(REF)] [FETCHARGS]
 539  */
 540 static int __trace_uprobe_create(int argc, const char **argv)
 541 {
 542         const char *event = NULL, *group = UPROBE_EVENT_SYSTEM;
 543         char *arg, *filename, *rctr, *rctr_end, *tmp;
 544         unsigned long offset, ref_ctr_offset;
 545         char *gbuf __free(kfree) = NULL;
 546         char *buf __free(kfree) = NULL;
 547         enum probe_print_type ptype;
 548         struct trace_uprobe *tu;
 549         bool is_return = false;
 550         struct path path;
 551         int i, ret;
 552
 553         ref_ctr_offset = 0;
 554
 555         switch (argv[0][0]) {
 556         case 'r':
 557                 is_return = true;
 558                 break;
 559         case 'p':
 560                 break;
 561         default:
 562                 return -ECANCELED;
 563         }
 564
 565         if (argc < 2)
 566                 return -ECANCELED;
 567
 568         trace_probe_log_init("trace_uprobe", argc, argv);
 569
 570         if (argc - 2 > MAX_TRACE_ARGS) {
 571                 trace_probe_log_set_index(2);
 572                 trace_probe_log_err(0, TOO_MANY_ARGS);
 573                 return -E2BIG;
 574         }
 575
 576         if (argv[0][1] == ':')
 577                 event = &argv[0][2];
 578
 579         if (!strchr(argv[1], '/'))
 580                 return -ECANCELED;
 581
 582         filename = kstrdup(argv[1], GFP_KERNEL);
 583         if (!filename)
 584                 return -ENOMEM;
 585
 586         /* Find the last occurrence, in case the path contains ':' too. */
 587         arg = strrchr(filename, ':');
 588         if (!arg || !isdigit(arg[1])) {
 589                 kfree(filename);
 590                 return -ECANCELED;
 591         }
 592
 593         trace_probe_log_set_index(1);   /* filename is the 2nd argument */
 594
 595         *arg++ = '\0';
 596         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 597         if (ret) {
 598                 trace_probe_log_err(0, FILE_NOT_FOUND);
 599                 kfree(filename);
 600                 trace_probe_log_clear();
 601                 return ret;
 602         }
 603         if (!d_is_reg(path.dentry)) {
 604                 trace_probe_log_err(0, NO_REGULAR_FILE);
 605                 ret = -EINVAL;
 606                 goto fail_address_parse;
 607         }
 608
 609         /* Parse reference counter offset if specified. */
 610         rctr = strchr(arg, '(');
 611         if (rctr) {
 612                 rctr_end = strchr(rctr, ')');
 613                 if (!rctr_end) {
 614                         ret = -EINVAL;
 615                         rctr_end = rctr + strlen(rctr);
 616                         trace_probe_log_err(rctr_end - filename,
 617                                             REFCNT_OPEN_BRACE);
 618                         goto fail_address_parse;
 619                 } else if (rctr_end[1] != '\0') {
 620                         ret = -EINVAL;
 621                         trace_probe_log_err(rctr_end + 1 - filename,
 622                                             BAD_REFCNT_SUFFIX);
 623                         goto fail_address_parse;
 624                 }
 625
 626                 *rctr++ = '\0';
 627                 *rctr_end = '\0';
 628                 ret = kstrtoul(rctr, 0, &ref_ctr_offset);
 629                 if (ret) {
 630                         trace_probe_log_err(rctr - filename, BAD_REFCNT);
 631                         goto fail_address_parse;
 632                 }
 633         }
 634
 635         /* Check if there is %return suffix */
 636         tmp = strchr(arg, '%');
 637         if (tmp) {
 638                 if (!strcmp(tmp, "%return")) {
 639                         *tmp = '\0';
 640                         is_return = true;
 641                 } else {
 642                         trace_probe_log_err(tmp - filename, BAD_ADDR_SUFFIX);
 643                         ret = -EINVAL;
 644                         goto fail_address_parse;
 645                 }
 646         }
 647
 648         /* Parse uprobe offset. */
 649         ret = kstrtoul(arg, 0, &offset);
 650         if (ret) {
 651                 trace_probe_log_err(arg - filename, BAD_UPROBE_OFFS);
 652                 goto fail_address_parse;
 653         }
 654
 655         /* setup a probe */
 656         trace_probe_log_set_index(0);
 657         if (event) {
 658                 gbuf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
 659                 if (!gbuf)
 660                         goto fail_mem;
 661
 662                 ret = traceprobe_parse_event_name(&event, &group, gbuf,
 663                                                   event - argv[0]);
 664                 if (ret)
 665                         goto fail_address_parse;
 666         }
 667
 668         if (!event) {
 669                 char *tail;
 670                 char *ptr;
 671
 672                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 673                 if (!tail)
 674                         goto fail_mem;
 675
 676                 ptr = strpbrk(tail, ".-_");
 677                 if (ptr)
 678                         *ptr = '\0';
 679
 680                 buf = kmalloc(MAX_EVENT_NAME_LEN, GFP_KERNEL);
 681                 if (!buf)
 682                         goto fail_mem;
 683                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 684                 event = buf;
 685                 kfree(tail);
 686         }
 687
 688         argc -= 2;
 689         argv += 2;
 690
 691         tu = alloc_trace_uprobe(group, event, argc, is_return);
 692         if (IS_ERR(tu)) {
 693                 ret = PTR_ERR(tu);
 694                 /* This must return -ENOMEM otherwise there is a bug */
 695                 WARN_ON_ONCE(ret != -ENOMEM);
 696                 goto fail_address_parse;
 697         }
 698         tu->offset = offset;
 699         tu->ref_ctr_offset = ref_ctr_offset;
 700         tu->path = path;
 701         tu->filename = filename;
 702
 703         /* parse arguments */
 704         for (i = 0; i < argc; i++) {
 705                 struct traceprobe_parse_context *ctx __free(traceprobe_parse_context)
 706                         = kzalloc(sizeof(*ctx), GFP_KERNEL);
 707
 708                 if (!ctx) {
 709                         ret = -ENOMEM;
 710                         goto error;
 711                 }
 712                 ctx->flags = (is_return ? TPARG_FL_RETURN : 0) | TPARG_FL_USER;
 713                 trace_probe_log_set_index(i + 2);
 714                 ret = traceprobe_parse_probe_arg(&tu->tp, i, argv[i], ctx);
 715                 if (ret)
 716                         goto error;
 717         }
 718
 719         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
 720         ret = traceprobe_set_print_fmt(&tu->tp, ptype);
 721         if (ret < 0)
 722                 goto error;
 723
 724         ret = register_trace_uprobe(tu);
 725         if (!ret)
 726                 goto out;
 727
 728 error:
 729         free_trace_uprobe(tu);
 730 out:
 731         trace_probe_log_clear();
 732         return ret;
 733
 734 fail_mem:
 735         ret = -ENOMEM;
 736
 737 fail_address_parse:
 738         trace_probe_log_clear();
 739         path_put(&path);
 740         kfree(filename);
 741
 742         return ret;
 743 }
 744
 745 int trace_uprobe_create(const char *raw_command)
 746 {
 747         return trace_probe_create(raw_command, __trace_uprobe_create);
 748 }
 749
 750 static int create_or_delete_trace_uprobe(const char *raw_command)
 751 {
 752         int ret;
 753
 754         if (raw_command[0] == '-')
 755                 return dyn_event_release(raw_command, &trace_uprobe_ops);
 756
 757         ret = dyn_event_create(raw_command, &trace_uprobe_ops);
 758         return ret == -ECANCELED ? -EINVAL : ret;
 759 }
 760
 761 static int trace_uprobe_release(struct dyn_event *ev)
 762 {
 763         struct trace_uprobe *tu = to_trace_uprobe(ev);
 764
 765         return unregister_trace_uprobe(tu);
 766 }
 767
 768 /* Probes listing interfaces */
 769 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev)
 770 {
 771         struct trace_uprobe *tu = to_trace_uprobe(ev);
 772         char c = is_ret_probe(tu) ? 'r' : 'p';
 773         int i;
 774
 775         seq_printf(m, "%c:%s/%s %s:0x%0*lx", c, trace_probe_group_name(&tu->tp),
 776                         trace_probe_name(&tu->tp), tu->filename,
 777                         (int)(sizeof(void *) * 2), tu->offset);
 778
 779         if (tu->ref_ctr_offset)
 780                 seq_printf(m, "(0x%lx)", tu->ref_ctr_offset);
 781
 782         for (i = 0; i < tu->tp.nr_args; i++)
 783                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 784
 785         seq_putc(m, '\n');
 786         return 0;
 787 }
 788
 789 static int probes_seq_show(struct seq_file *m, void *v)
 790 {
 791         struct dyn_event *ev = v;
 792
 793         if (!is_trace_uprobe(ev))
 794                 return 0;
 795
 796         return trace_uprobe_show(m, ev);
 797 }
 798
 799 static const struct seq_operations probes_seq_op = {
 800         .start  = dyn_event_seq_start,
 801         .next   = dyn_event_seq_next,
 802         .stop   = dyn_event_seq_stop,
 803         .show   = probes_seq_show
 804 };
 805
 806 static int probes_open(struct inode *inode, struct file *file)
 807 {
 808         int ret;
 809
 810         ret = security_locked_down(LOCKDOWN_TRACEFS);
 811         if (ret)
 812                 return ret;
 813
 814         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 815                 ret = dyn_events_release_all(&trace_uprobe_ops);
 816                 if (ret)
 817                         return ret;
 818         }
 819
 820         return seq_open(file, &probes_seq_op);
 821 }
 822
 823 static ssize_t probes_write(struct file *file, const char __user *buffer,
 824                             size_t count, loff_t *ppos)
 825 {
 826         return trace_parse_run_command(file, buffer, count, ppos,
 827                                         create_or_delete_trace_uprobe);
 828 }
 829
 830 static const struct file_operations uprobe_events_ops = {
 831         .owner          = THIS_MODULE,
 832         .open           = probes_open,
 833         .read           = seq_read,
 834         .llseek         = seq_lseek,
 835         .release        = seq_release,
 836         .write          = probes_write,
 837 };
 838
 839 /* Probes profiling interfaces */
 840 static int probes_profile_seq_show(struct seq_file *m, void *v)
 841 {
 842         struct dyn_event *ev = v;
 843         struct trace_uprobe *tu;
 844         unsigned long nhits;
 845         int cpu;
 846
 847         if (!is_trace_uprobe(ev))
 848                 return 0;
 849
 850         tu = to_trace_uprobe(ev);
 851
 852         nhits = 0;
 853         for_each_possible_cpu(cpu) {
 854                 nhits += per_cpu(*tu->nhits, cpu);
 855         }
 856
 857         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 858                    trace_probe_name(&tu->tp), nhits);
 859         return 0;
 860 }
 861
 862 static const struct seq_operations profile_seq_op = {
 863         .start  = dyn_event_seq_start,
 864         .next   = dyn_event_seq_next,
 865         .stop   = dyn_event_seq_stop,
 866         .show   = probes_profile_seq_show
 867 };
 868
 869 static int profile_open(struct inode *inode, struct file *file)
 870 {
 871         int ret;
 872
 873         ret = security_locked_down(LOCKDOWN_TRACEFS);
 874         if (ret)
 875                 return ret;
 876
 877         return seq_open(file, &profile_seq_op);
 878 }
 879
 880 static const struct file_operations uprobe_profile_ops = {
 881         .owner          = THIS_MODULE,
 882         .open           = profile_open,
 883         .read           = seq_read,
 884         .llseek         = seq_lseek,
 885         .release        = seq_release,
 886 };
 887
 888 struct uprobe_cpu_buffer {
 889         struct mutex mutex;
 890         void *buf;
 891         int dsize;
 892 };
 893 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 894 static int uprobe_buffer_refcnt;
 895 #define MAX_UCB_BUFFER_SIZE PAGE_SIZE
 896
 897 static int uprobe_buffer_init(void)
 898 {
 899         int cpu, err_cpu;
 900
 901         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 902         if (uprobe_cpu_buffer == NULL)
 903                 return -ENOMEM;
 904
 905         for_each_possible_cpu(cpu) {
 906                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 907                                                   GFP_KERNEL, 0);
 908                 if (p == NULL) {
 909                         err_cpu = cpu;
 910                         goto err;
 911                 }
 912                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 913                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 914         }
 915
 916         return 0;
 917
 918 err:
 919         for_each_possible_cpu(cpu) {
 920                 if (cpu == err_cpu)
 921                         break;
 922                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 923         }
 924
 925         free_percpu(uprobe_cpu_buffer);
 926         return -ENOMEM;
 927 }
 928
 929 static int uprobe_buffer_enable(void)
 930 {
 931         int ret = 0;
 932
 933         BUG_ON(!mutex_is_locked(&event_mutex));
 934
 935         if (uprobe_buffer_refcnt++ == 0) {
 936                 ret = uprobe_buffer_init();
 937                 if (ret < 0)
 938                         uprobe_buffer_refcnt--;
 939         }
 940
 941         return ret;
 942 }
 943
 944 static void uprobe_buffer_disable(void)
 945 {
 946         int cpu;
 947
 948         BUG_ON(!mutex_is_locked(&event_mutex));
 949
 950         if (--uprobe_buffer_refcnt == 0) {
 951                 for_each_possible_cpu(cpu)
 952                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 953                                                              cpu)->buf);
 954
 955                 free_percpu(uprobe_cpu_buffer);
 956                 uprobe_cpu_buffer = NULL;
 957         }
 958 }
 959
 960 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 961 {
 962         struct uprobe_cpu_buffer *ucb;
 963         int cpu;
 964
 965         cpu = raw_smp_processor_id();
 966         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 967
 968         /*
 969          * Use per-cpu buffers for fastest access, but we might migrate
 970          * so the mutex makes sure we have sole access to it.
 971          */
 972         mutex_lock(&ucb->mutex);
 973
 974         return ucb;
 975 }
 976
 977 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 978 {
 979         if (!ucb)
 980                 return;
 981         mutex_unlock(&ucb->mutex);
 982 }
 983
 984 static struct uprobe_cpu_buffer *prepare_uprobe_buffer(struct trace_uprobe *tu,
 985                                                        struct pt_regs *regs,
 986                                                        struct uprobe_cpu_buffer **ucbp)
 987 {
 988         struct uprobe_cpu_buffer *ucb;
 989         int dsize, esize;
 990
 991         if (*ucbp)
 992                 return *ucbp;
 993
 994         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 995         dsize = __get_data_size(&tu->tp, regs, NULL);
 996
 997         ucb = uprobe_buffer_get();
 998         ucb->dsize = tu->tp.size + dsize;
 999
1000         if (WARN_ON_ONCE(ucb->dsize > MAX_UCB_BUFFER_SIZE)) {
1001                 ucb->dsize = MAX_UCB_BUFFER_SIZE;
1002                 dsize = MAX_UCB_BUFFER_SIZE - tu->tp.size;
1003         }
1004
1005         store_trace_args(ucb->buf, &tu->tp, regs, NULL, esize, dsize);
1006
1007         *ucbp = ucb;
1008         return ucb;
1009 }
1010
1011 static void __uprobe_trace_func(struct trace_uprobe *tu,
1012                                 unsigned long func, struct pt_regs *regs,
1013                                 struct uprobe_cpu_buffer *ucb,
1014                                 struct trace_event_file *trace_file)
1015 {
1016         struct uprobe_trace_entry_head *entry;
1017         struct trace_event_buffer fbuffer;
1018         void *data;
1019         int size, esize;
1020         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1021
1022         WARN_ON(call != trace_file->event_call);
1023
1024         if (trace_trigger_soft_disabled(trace_file))
1025                 return;
1026
1027         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1028         size = esize + ucb->dsize;
1029         entry = trace_event_buffer_reserve(&fbuffer, trace_file, size);
1030         if (!entry)
1031                 return;
1032
1033         if (is_ret_probe(tu)) {
1034                 entry->vaddr[0] = func;
1035                 entry->vaddr[1] = instruction_pointer(regs);
1036                 data = DATAOF_TRACE_ENTRY(entry, true);
1037         } else {
1038                 entry->vaddr[0] = instruction_pointer(regs);
1039                 data = DATAOF_TRACE_ENTRY(entry, false);
1040         }
1041
1042         memcpy(data, ucb->buf, ucb->dsize);
1043
1044         trace_event_buffer_commit(&fbuffer);
1045 }
1046
1047 /* uprobe handler */
1048 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
1049                              struct uprobe_cpu_buffer **ucbp)
1050 {
1051         struct event_file_link *link;
1052         struct uprobe_cpu_buffer *ucb;
1053
1054         if (is_ret_probe(tu))
1055                 return 0;
1056
1057         ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1058
1059         rcu_read_lock();
1060         trace_probe_for_each_link_rcu(link, &tu->tp)
1061                 __uprobe_trace_func(tu, 0, regs, ucb, link->file);
1062         rcu_read_unlock();
1063
1064         return 0;
1065 }
1066
1067 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
1068                                  struct pt_regs *regs,
1069                                  struct uprobe_cpu_buffer **ucbp)
1070 {
1071         struct event_file_link *link;
1072         struct uprobe_cpu_buffer *ucb;
1073
1074         ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1075
1076         rcu_read_lock();
1077         trace_probe_for_each_link_rcu(link, &tu->tp)
1078                 __uprobe_trace_func(tu, func, regs, ucb, link->file);
1079         rcu_read_unlock();
1080 }
1081
1082 /* Event entry printers */
1083 static enum print_line_t
1084 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
1085 {
1086         struct uprobe_trace_entry_head *entry;
1087         struct trace_seq *s = &iter->seq;
1088         struct trace_uprobe *tu;
1089         u8 *data;
1090
1091         entry = (struct uprobe_trace_entry_head *)iter->ent;
1092         tu = trace_uprobe_primary_from_call(
1093                 container_of(event, struct trace_event_call, event));
1094         if (unlikely(!tu))
1095                 goto out;
1096
1097         if (is_ret_probe(tu)) {
1098                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
1099                                  trace_probe_name(&tu->tp),
1100                                  entry->vaddr[1], entry->vaddr[0]);
1101                 data = DATAOF_TRACE_ENTRY(entry, true);
1102         } else {
1103                 trace_seq_printf(s, "%s: (0x%lx)",
1104                                  trace_probe_name(&tu->tp),
1105                                  entry->vaddr[0]);
1106                 data = DATAOF_TRACE_ENTRY(entry, false);
1107         }
1108
1109         if (trace_probe_print_args(s, tu->tp.args, tu->tp.nr_args, data, entry) < 0)
1110                 goto out;
1111
1112         trace_seq_putc(s, '\n');
1113
1114  out:
1115         return trace_handle_return(s);
1116 }
1117
1118 typedef bool (*filter_func_t)(struct uprobe_consumer *self, struct mm_struct *mm);
1119
1120 static int trace_uprobe_enable(struct trace_uprobe *tu, filter_func_t filter)
1121 {
1122         struct inode *inode = d_real_inode(tu->path.dentry);
1123         struct uprobe *uprobe;
1124
1125         tu->consumer.filter = filter;
1126         uprobe = uprobe_register(inode, tu->offset, tu->ref_ctr_offset, &tu->consumer);
1127         if (IS_ERR(uprobe))
1128                 return PTR_ERR(uprobe);
1129
1130         tu->uprobe = uprobe;
1131         return 0;
1132 }
1133
1134 static void __probe_event_disable(struct trace_probe *tp)
1135 {
1136         struct trace_uprobe *tu;
1137         bool sync = false;
1138
1139         tu = container_of(tp, struct trace_uprobe, tp);
1140         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1141
1142         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1143                 if (!tu->uprobe)
1144                         continue;
1145
1146                 uprobe_unregister_nosync(tu->uprobe, &tu->consumer);
1147                 sync = true;
1148                 tu->uprobe = NULL;
1149         }
1150         if (sync)
1151                 uprobe_unregister_sync();
1152 }
1153
1154 static int probe_event_enable(struct trace_event_call *call,
1155                         struct trace_event_file *file, filter_func_t filter)
1156 {
1157         struct trace_probe *tp;
1158         struct trace_uprobe *tu;
1159         bool enabled;
1160         int ret;
1161
1162         tp = trace_probe_primary_from_call(call);
1163         if (WARN_ON_ONCE(!tp))
1164                 return -ENODEV;
1165         enabled = trace_probe_is_enabled(tp);
1166
1167         /* This may also change "enabled" state */
1168         if (file) {
1169                 if (trace_probe_test_flag(tp, TP_FLAG_PROFILE))
1170                         return -EINTR;
1171
1172                 ret = trace_probe_add_file(tp, file);
1173                 if (ret < 0)
1174                         return ret;
1175         } else {
1176                 if (trace_probe_test_flag(tp, TP_FLAG_TRACE))
1177                         return -EINTR;
1178
1179                 trace_probe_set_flag(tp, TP_FLAG_PROFILE);
1180         }
1181
1182         tu = container_of(tp, struct trace_uprobe, tp);
1183         WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
1184
1185         if (enabled)
1186                 return 0;
1187
1188         ret = uprobe_buffer_enable();
1189         if (ret)
1190                 goto err_flags;
1191
1192         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1193                 ret = trace_uprobe_enable(tu, filter);
1194                 if (ret) {
1195                         __probe_event_disable(tp);
1196                         goto err_buffer;
1197                 }
1198         }
1199
1200         return 0;
1201
1202  err_buffer:
1203         uprobe_buffer_disable();
1204
1205  err_flags:
1206         if (file)
1207                 trace_probe_remove_file(tp, file);
1208         else
1209                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1210
1211         return ret;
1212 }
1213
1214 static void probe_event_disable(struct trace_event_call *call,
1215                                 struct trace_event_file *file)
1216 {
1217         struct trace_probe *tp;
1218
1219         tp = trace_probe_primary_from_call(call);
1220         if (WARN_ON_ONCE(!tp))
1221                 return;
1222
1223         if (!trace_probe_is_enabled(tp))
1224                 return;
1225
1226         if (file) {
1227                 if (trace_probe_remove_file(tp, file) < 0)
1228                         return;
1229
1230                 if (trace_probe_is_enabled(tp))
1231                         return;
1232         } else
1233                 trace_probe_clear_flag(tp, TP_FLAG_PROFILE);
1234
1235         __probe_event_disable(tp);
1236         uprobe_buffer_disable();
1237 }
1238
1239 static int uprobe_event_define_fields(struct trace_event_call *event_call)
1240 {
1241         int ret, size;
1242         struct uprobe_trace_entry_head field;
1243         struct trace_uprobe *tu;
1244
1245         tu = trace_uprobe_primary_from_call(event_call);
1246         if (unlikely(!tu))
1247                 return -ENODEV;
1248
1249         if (is_ret_probe(tu)) {
1250                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1251                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1252                 size = SIZEOF_TRACE_ENTRY(true);
1253         } else {
1254                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1255                 size = SIZEOF_TRACE_ENTRY(false);
1256         }
1257
1258         return traceprobe_define_arg_fields(event_call, size, &tu->tp);
1259 }
1260
1261 #ifdef CONFIG_PERF_EVENTS
1262 static bool
1263 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1264 {
1265         struct perf_event *event;
1266
1267         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1268                 if (event->hw.target->mm == mm)
1269                         return true;
1270         }
1271
1272         return false;
1273 }
1274
1275 static inline bool
1276 trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
1277                           struct perf_event *event)
1278 {
1279         return __uprobe_perf_filter(filter, event->hw.target->mm);
1280 }
1281
1282 static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
1283                                        struct perf_event *event)
1284 {
1285         bool done;
1286
1287         write_lock(&filter->rwlock);
1288         if (event->hw.target) {
1289                 list_del(&event->hw.tp_list);
1290                 done = filter->nr_systemwide ||
1291                         (event->hw.target->flags & PF_EXITING) ||
1292                         trace_uprobe_filter_event(filter, event);
1293         } else {
1294                 filter->nr_systemwide--;
1295                 done = filter->nr_systemwide;
1296         }
1297         write_unlock(&filter->rwlock);
1298
1299         return done;
1300 }
1301
1302 /* This returns true if the filter always covers target mm */
1303 static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
1304                                     struct perf_event *event)
1305 {
1306         bool done;
1307
1308         write_lock(&filter->rwlock);
1309         if (event->hw.target) {
1310                 /*
1311                  * event->parent != NULL means copy_process(), we can avoid
1312                  * uprobe_apply(). current->mm must be probed and we can rely
1313                  * on dup_mmap() which preserves the already installed bp's.
1314                  *
1315                  * attr.enable_on_exec means that exec/mmap will install the
1316                  * breakpoints we need.
1317                  */
1318                 done = filter->nr_systemwide ||
1319                         event->parent || event->attr.enable_on_exec ||
1320                         trace_uprobe_filter_event(filter, event);
1321                 list_add(&event->hw.tp_list, &filter->perf_events);
1322         } else {
1323                 done = filter->nr_systemwide;
1324                 filter->nr_systemwide++;
1325         }
1326         write_unlock(&filter->rwlock);
1327
1328         return done;
1329 }
1330
1331 static int uprobe_perf_close(struct trace_event_call *call,
1332                              struct perf_event *event)
1333 {
1334         struct trace_probe *tp;
1335         struct trace_uprobe *tu;
1336         int ret = 0;
1337
1338         tp = trace_probe_primary_from_call(call);
1339         if (WARN_ON_ONCE(!tp))
1340                 return -ENODEV;
1341
1342         tu = container_of(tp, struct trace_uprobe, tp);
1343         if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
1344                 return 0;
1345
1346         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1347                 ret = uprobe_apply(tu->uprobe, &tu->consumer, false);
1348                 if (ret)
1349                         break;
1350         }
1351
1352         return ret;
1353 }
1354
1355 static int uprobe_perf_open(struct trace_event_call *call,
1356                             struct perf_event *event)
1357 {
1358         struct trace_probe *tp;
1359         struct trace_uprobe *tu;
1360         int err = 0;
1361
1362         tp = trace_probe_primary_from_call(call);
1363         if (WARN_ON_ONCE(!tp))
1364                 return -ENODEV;
1365
1366         tu = container_of(tp, struct trace_uprobe, tp);
1367         if (trace_uprobe_filter_add(tu->tp.event->filter, event))
1368                 return 0;
1369
1370         list_for_each_entry(tu, trace_probe_probe_list(tp), tp.list) {
1371                 err = uprobe_apply(tu->uprobe, &tu->consumer, true);
1372                 if (err) {
1373                         uprobe_perf_close(call, event);
1374                         break;
1375                 }
1376         }
1377
1378         return err;
1379 }
1380
1381 static bool uprobe_perf_filter(struct uprobe_consumer *uc, struct mm_struct *mm)
1382 {
1383         struct trace_uprobe_filter *filter;
1384         struct trace_uprobe *tu;
1385         int ret;
1386
1387         tu = container_of(uc, struct trace_uprobe, consumer);
1388         filter = tu->tp.event->filter;
1389
1390         /*
1391          * speculative short-circuiting check to avoid unnecessarily taking
1392          * filter->rwlock below, if the uprobe has system-wide consumer
1393          */
1394         if (READ_ONCE(filter->nr_systemwide))
1395                 return true;
1396
1397         read_lock(&filter->rwlock);
1398         ret = __uprobe_perf_filter(filter, mm);
1399         read_unlock(&filter->rwlock);
1400
1401         return ret;
1402 }
1403
1404 static void __uprobe_perf_func(struct trace_uprobe *tu,
1405                                unsigned long func, struct pt_regs *regs,
1406                                struct uprobe_cpu_buffer **ucbp)
1407 {
1408         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1409         struct uprobe_trace_entry_head *entry;
1410         struct uprobe_cpu_buffer *ucb;
1411         struct hlist_head *head;
1412         void *data;
1413         int size, esize;
1414         int rctx;
1415
1416 #ifdef CONFIG_BPF_EVENTS
1417         if (bpf_prog_array_valid(call)) {
1418                 const struct bpf_prog_array *array;
1419                 u32 ret;
1420
1421                 rcu_read_lock_trace();
1422                 array = rcu_dereference_check(call->prog_array, rcu_read_lock_trace_held());
1423                 ret = bpf_prog_run_array_uprobe(array, regs, bpf_prog_run);
1424                 rcu_read_unlock_trace();
1425                 if (!ret)
1426                         return;
1427         }
1428 #endif /* CONFIG_BPF_EVENTS */
1429
1430         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1431
1432         ucb = prepare_uprobe_buffer(tu, regs, ucbp);
1433         size = esize + ucb->dsize;
1434         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1435         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1436                 return;
1437
1438         preempt_disable();
1439         head = this_cpu_ptr(call->perf_events);
1440         if (hlist_empty(head))
1441                 goto out;
1442
1443         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1444         if (!entry)
1445                 goto out;
1446
1447         if (is_ret_probe(tu)) {
1448                 entry->vaddr[0] = func;
1449                 entry->vaddr[1] = instruction_pointer(regs);
1450                 data = DATAOF_TRACE_ENTRY(entry, true);
1451         } else {
1452                 entry->vaddr[0] = instruction_pointer(regs);
1453                 data = DATAOF_TRACE_ENTRY(entry, false);
1454         }
1455
1456         memcpy(data, ucb->buf, ucb->dsize);
1457
1458         if (size - esize > ucb->dsize)
1459                 memset(data + ucb->dsize, 0, size - esize - ucb->dsize);
1460
1461         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1462                               head, NULL);
1463  out:
1464         preempt_enable();
1465 }
1466
1467 /* uprobe profile handler */
1468 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1469                             struct uprobe_cpu_buffer **ucbp)
1470 {
1471         if (!uprobe_perf_filter(&tu->consumer, current->mm))
1472                 return UPROBE_HANDLER_REMOVE;
1473
1474         if (!is_ret_probe(tu))
1475                 __uprobe_perf_func(tu, 0, regs, ucbp);
1476         return 0;
1477 }
1478
1479 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1480                                 struct pt_regs *regs,
1481                                 struct uprobe_cpu_buffer **ucbp)
1482 {
1483         __uprobe_perf_func(tu, func, regs, ucbp);
1484 }
1485
1486 int bpf_get_uprobe_info(const struct perf_event *event, u32 *fd_type,
1487                         const char **filename, u64 *probe_offset,
1488                         u64 *probe_addr, bool perf_type_tracepoint)
1489 {
1490         const char *pevent = trace_event_name(event->tp_event);
1491         const char *group = event->tp_event->class->system;
1492         struct trace_uprobe *tu;
1493
1494         if (perf_type_tracepoint)
1495                 tu = find_probe_event(pevent, group);
1496         else
1497                 tu = trace_uprobe_primary_from_call(event->tp_event);
1498         if (!tu)
1499                 return -EINVAL;
1500
1501         *fd_type = is_ret_probe(tu) ? BPF_FD_TYPE_URETPROBE
1502                                     : BPF_FD_TYPE_UPROBE;
1503         *filename = tu->filename;
1504         *probe_offset = tu->offset;
1505         *probe_addr = tu->ref_ctr_offset;
1506         return 0;
1507 }
1508 #endif  /* CONFIG_PERF_EVENTS */
1509
1510 static int
1511 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1512                       void *data)
1513 {
1514         struct trace_event_file *file = data;
1515
1516         switch (type) {
1517         case TRACE_REG_REGISTER:
1518                 return probe_event_enable(event, file, NULL);
1519
1520         case TRACE_REG_UNREGISTER:
1521                 probe_event_disable(event, file);
1522                 return 0;
1523
1524 #ifdef CONFIG_PERF_EVENTS
1525         case TRACE_REG_PERF_REGISTER:
1526                 return probe_event_enable(event, NULL, uprobe_perf_filter);
1527
1528         case TRACE_REG_PERF_UNREGISTER:
1529                 probe_event_disable(event, NULL);
1530                 return 0;
1531
1532         case TRACE_REG_PERF_OPEN:
1533                 return uprobe_perf_open(event, data);
1534
1535         case TRACE_REG_PERF_CLOSE:
1536                 return uprobe_perf_close(event, data);
1537
1538 #endif
1539         default:
1540                 return 0;
1541         }
1542 }
1543
1544 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs,
1545                              __u64 *data)
1546 {
1547         struct trace_uprobe *tu;
1548         struct uprobe_dispatch_data udd;
1549         struct uprobe_cpu_buffer *ucb = NULL;
1550         int ret = 0;
1551
1552         tu = container_of(con, struct trace_uprobe, consumer);
1553
1554         this_cpu_inc(*tu->nhits);
1555
1556         udd.tu = tu;
1557         udd.bp_addr = instruction_pointer(regs);
1558
1559         current->utask->vaddr = (unsigned long) &udd;
1560
1561         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1562                 return 0;
1563
1564         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1565                 ret |= uprobe_trace_func(tu, regs, &ucb);
1566
1567 #ifdef CONFIG_PERF_EVENTS
1568         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1569                 ret |= uprobe_perf_func(tu, regs, &ucb);
1570 #endif
1571         uprobe_buffer_put(ucb);
1572         return ret;
1573 }
1574
1575 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1576                                 unsigned long func, struct pt_regs *regs,
1577                                 __u64 *data)
1578 {
1579         struct trace_uprobe *tu;
1580         struct uprobe_dispatch_data udd;
1581         struct uprobe_cpu_buffer *ucb = NULL;
1582
1583         tu = container_of(con, struct trace_uprobe, consumer);
1584
1585         udd.tu = tu;
1586         udd.bp_addr = func;
1587
1588         current->utask->vaddr = (unsigned long) &udd;
1589
1590         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1591                 return 0;
1592
1593         if (trace_probe_test_flag(&tu->tp, TP_FLAG_TRACE))
1594                 uretprobe_trace_func(tu, func, regs, &ucb);
1595
1596 #ifdef CONFIG_PERF_EVENTS
1597         if (trace_probe_test_flag(&tu->tp, TP_FLAG_PROFILE))
1598                 uretprobe_perf_func(tu, func, regs, &ucb);
1599 #endif
1600         uprobe_buffer_put(ucb);
1601         return 0;
1602 }
1603
1604 static struct trace_event_functions uprobe_funcs = {
1605         .trace          = print_uprobe_event
1606 };
1607
1608 static struct trace_event_fields uprobe_fields_array[] = {
1609         { .type = TRACE_FUNCTION_TYPE,
1610           .define_fields = uprobe_event_define_fields },
1611         {}
1612 };
1613
1614 static inline void init_trace_event_call(struct trace_uprobe *tu)
1615 {
1616         struct trace_event_call *call = trace_probe_event_call(&tu->tp);
1617         call->event.funcs = &uprobe_funcs;
1618         call->class->fields_array = uprobe_fields_array;
1619
1620         call->flags = TRACE_EVENT_FL_UPROBE | TRACE_EVENT_FL_CAP_ANY;
1621         call->class->reg = trace_uprobe_register;
1622 }
1623
1624 static int register_uprobe_event(struct trace_uprobe *tu)
1625 {
1626         init_trace_event_call(tu);
1627
1628         return trace_probe_register_event_call(&tu->tp);
1629 }
1630
1631 static int unregister_uprobe_event(struct trace_uprobe *tu)
1632 {
1633         return trace_probe_unregister_event_call(&tu->tp);
1634 }
1635
1636 #ifdef CONFIG_PERF_EVENTS
1637 struct trace_event_call *
1638 create_local_trace_uprobe(char *name, unsigned long offs,
1639                           unsigned long ref_ctr_offset, bool is_return)
1640 {
1641         enum probe_print_type ptype;
1642         struct trace_uprobe *tu;
1643         struct path path;
1644         int ret;
1645
1646         ret = kern_path(name, LOOKUP_FOLLOW, &path);
1647         if (ret)
1648                 return ERR_PTR(ret);
1649
1650         if (!d_is_reg(path.dentry)) {
1651                 path_put(&path);
1652                 return ERR_PTR(-EINVAL);
1653         }
1654
1655         /*
1656          * local trace_kprobes are not added to dyn_event, so they are never
1657          * searched in find_trace_kprobe(). Therefore, there is no concern of
1658          * duplicated name "DUMMY_EVENT" here.
1659          */
1660         tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
1661                                 is_return);
1662
1663         if (IS_ERR(tu)) {
1664                 pr_info("Failed to allocate trace_uprobe.(%d)\n",
1665                         (int)PTR_ERR(tu));
1666                 path_put(&path);
1667                 return ERR_CAST(tu);
1668         }
1669
1670         tu->offset = offs;
1671         tu->path = path;
1672         tu->ref_ctr_offset = ref_ctr_offset;
1673         tu->filename = kstrdup(name, GFP_KERNEL);
1674         if (!tu->filename) {
1675                 ret = -ENOMEM;
1676                 goto error;
1677         }
1678
1679         init_trace_event_call(tu);
1680
1681         ptype = is_ret_probe(tu) ? PROBE_PRINT_RETURN : PROBE_PRINT_NORMAL;
1682         if (traceprobe_set_print_fmt(&tu->tp, ptype) < 0) {
1683                 ret = -ENOMEM;
1684                 goto error;
1685         }
1686
1687         return trace_probe_event_call(&tu->tp);
1688 error:
1689         free_trace_uprobe(tu);
1690         return ERR_PTR(ret);
1691 }
1692
1693 void destroy_local_trace_uprobe(struct trace_event_call *event_call)
1694 {
1695         struct trace_uprobe *tu;
1696
1697         tu = trace_uprobe_primary_from_call(event_call);
1698
1699         free_trace_uprobe(tu);
1700 }
1701 #endif /* CONFIG_PERF_EVENTS */
1702
1703 /* Make a trace interface for controlling probe points */
1704 static __init int init_uprobe_trace(void)
1705 {
1706         int ret;
1707
1708         ret = dyn_event_register(&trace_uprobe_ops);
1709         if (ret)
1710                 return ret;
1711
1712         ret = tracing_init_dentry();
1713         if (ret)
1714                 return 0;
1715
1716         trace_create_file("uprobe_events", TRACE_MODE_WRITE, NULL,
1717                                     NULL, &uprobe_events_ops);
1718         /* Profile interface */
1719         trace_create_file("uprobe_profile", TRACE_MODE_READ, NULL,
1720                                     NULL, &uprobe_profile_ops);
1721         return 0;
1722 }
1723
1724 fs_initcall(init_uprobe_trace);