kernel/trace/trace_kprobe.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Kprobes-based tracing events
   4  *
   5  * Created by Masami Hiramatsu <mhiramat@redhat.com>
   6  *
   7  */
   8 #define pr_fmt(fmt)     "trace_kprobe: " fmt
   9
  10 #include <linux/module.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/rculist.h>
  13 #include <linux/error-injection.h>
  14
  15 #include "trace_kprobe_selftest.h"
  16 #include "trace_probe.h"
  17
  18 #define KPROBE_EVENT_SYSTEM "kprobes"
  19 #define KRETPROBE_MAXACTIVE_MAX 4096
  20
  21 /**
  22  * Kprobe event core functions
  23  */
  24 struct trace_kprobe {
  25         struct list_head        list;
  26         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  27         unsigned long __percpu *nhit;
  28         const char              *symbol;        /* symbol name */
  29         struct trace_probe      tp;
  30 };
  31
  32 #define SIZEOF_TRACE_KPROBE(n)                          \
  33         (offsetof(struct trace_kprobe, tp.args) +       \
  34         (sizeof(struct probe_arg) * (n)))
  35
  36 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  37 {
  38         return tk->rp.handler != NULL;
  39 }
  40
  41 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  42 {
  43         return tk->symbol ? tk->symbol : "unknown";
  44 }
  45
  46 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  47 {
  48         return tk->rp.kp.offset;
  49 }
  50
  51 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  52 {
  53         return !!(kprobe_gone(&tk->rp.kp));
  54 }
  55
  56 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  57                                                  struct module *mod)
  58 {
  59         int len = strlen(mod->name);
  60         const char *name = trace_kprobe_symbol(tk);
  61         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  62 }
  63
  64 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  65 {
  66         return !!strchr(trace_kprobe_symbol(tk), ':');
  67 }
  68
  69 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
  70 {
  71         unsigned long nhit = 0;
  72         int cpu;
  73
  74         for_each_possible_cpu(cpu)
  75                 nhit += *per_cpu_ptr(tk->nhit, cpu);
  76
  77         return nhit;
  78 }
  79
  80 /* Return 0 if it fails to find the symbol address */
  81 static nokprobe_inline
  82 unsigned long trace_kprobe_address(struct trace_kprobe *tk)
  83 {
  84         unsigned long addr;
  85
  86         if (tk->symbol) {
  87                 addr = (unsigned long)
  88                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
  89                 if (addr)
  90                         addr += tk->rp.kp.offset;
  91         } else {
  92                 addr = (unsigned long)tk->rp.kp.addr;
  93         }
  94         return addr;
  95 }
  96
  97 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
  98 {
  99         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 100
 101         return kprobe_on_func_entry(tk->rp.kp.addr,
 102                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
 103                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
 104 }
 105
 106 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 107 {
 108         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 109
 110         return within_error_injection_list(trace_kprobe_address(tk));
 111 }
 112
 113 static int register_kprobe_event(struct trace_kprobe *tk);
 114 static int unregister_kprobe_event(struct trace_kprobe *tk);
 115
 116 static DEFINE_MUTEX(probe_lock);
 117 static LIST_HEAD(probe_list);
 118
 119 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 120 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 121                                 struct pt_regs *regs);
 122
 123 /* Memory fetching by symbol */
 124 struct symbol_cache {
 125         char            *symbol;
 126         long            offset;
 127         unsigned long   addr;
 128 };
 129
 130 unsigned long update_symbol_cache(struct symbol_cache *sc)
 131 {
 132         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 133
 134         if (sc->addr)
 135                 sc->addr += sc->offset;
 136
 137         return sc->addr;
 138 }
 139
 140 void free_symbol_cache(struct symbol_cache *sc)
 141 {
 142         kfree(sc->symbol);
 143         kfree(sc);
 144 }
 145
 146 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 147 {
 148         struct symbol_cache *sc;
 149
 150         if (!sym || strlen(sym) == 0)
 151                 return NULL;
 152
 153         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 154         if (!sc)
 155                 return NULL;
 156
 157         sc->symbol = kstrdup(sym, GFP_KERNEL);
 158         if (!sc->symbol) {
 159                 kfree(sc);
 160                 return NULL;
 161         }
 162         sc->offset = offset;
 163         update_symbol_cache(sc);
 164
 165         return sc;
 166 }
 167
 168 /*
 169  * Kprobes-specific fetch functions
 170  */
 171 #define DEFINE_FETCH_stack(type)                                        \
 172 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 173                                           void *offset, void *dest)     \
 174 {                                                                       \
 175         *(type *)dest = (type)regs_get_kernel_stack_nth(regs,           \
 176                                 (unsigned int)((unsigned long)offset)); \
 177 }                                                                       \
 178 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 179
 180 DEFINE_BASIC_FETCH_FUNCS(stack)
 181 /* No string on the stack entry */
 182 #define fetch_stack_string      NULL
 183 #define fetch_stack_string_size NULL
 184
 185 #define DEFINE_FETCH_memory(type)                                       \
 186 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 187                                           void *addr, void *dest)       \
 188 {                                                                       \
 189         type retval;                                                    \
 190         if (probe_kernel_address(addr, retval))                         \
 191                 *(type *)dest = 0;                                      \
 192         else                                                            \
 193                 *(type *)dest = retval;                                 \
 194 }                                                                       \
 195 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 196
 197 DEFINE_BASIC_FETCH_FUNCS(memory)
 198 /*
 199  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 200  * length and relative data location.
 201  */
 202 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 203                                             void *addr, void *dest)
 204 {
 205         int maxlen = get_rloc_len(*(u32 *)dest);
 206         u8 *dst = get_rloc_data(dest);
 207         long ret;
 208
 209         if (!maxlen)
 210                 return;
 211
 212         /*
 213          * Try to get string again, since the string can be changed while
 214          * probing.
 215          */
 216         ret = strncpy_from_unsafe(dst, addr, maxlen);
 217
 218         if (ret < 0) {  /* Failed to fetch string */
 219                 dst[0] = '\0';
 220                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 221         } else {
 222                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 223         }
 224 }
 225 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 226
 227 /* Return the length of string -- including null terminal byte */
 228 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 229                                                  void *addr, void *dest)
 230 {
 231         mm_segment_t old_fs;
 232         int ret, len = 0;
 233         u8 c;
 234
 235         old_fs = get_fs();
 236         set_fs(KERNEL_DS);
 237         pagefault_disable();
 238
 239         do {
 240                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 241                 len++;
 242         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 243
 244         pagefault_enable();
 245         set_fs(old_fs);
 246
 247         if (ret < 0)    /* Failed to check the length */
 248                 *(u32 *)dest = 0;
 249         else
 250                 *(u32 *)dest = len;
 251 }
 252 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 253
 254 #define DEFINE_FETCH_symbol(type)                                       \
 255 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 256 {                                                                       \
 257         struct symbol_cache *sc = data;                                 \
 258         if (sc->addr)                                                   \
 259                 fetch_memory_##type(regs, (void *)sc->addr, dest);      \
 260         else                                                            \
 261                 *(type *)dest = 0;                                      \
 262 }                                                                       \
 263 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 264
 265 DEFINE_BASIC_FETCH_FUNCS(symbol)
 266 DEFINE_FETCH_symbol(string)
 267 DEFINE_FETCH_symbol(string_size)
 268
 269 /* kprobes don't support file_offset fetch methods */
 270 #define fetch_file_offset_u8            NULL
 271 #define fetch_file_offset_u16           NULL
 272 #define fetch_file_offset_u32           NULL
 273 #define fetch_file_offset_u64           NULL
 274 #define fetch_file_offset_string        NULL
 275 #define fetch_file_offset_string_size   NULL
 276
 277 /* Fetch type information table */
 278 static const struct fetch_type kprobes_fetch_type_table[] = {
 279         /* Special types */
 280         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 281                                         sizeof(u32), 1, "__data_loc char[]"),
 282         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 283                                         string_size, sizeof(u32), 0, "u32"),
 284         /* Basic types */
 285         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 286         ASSIGN_FETCH_TYPE(u16, u16, 0),
 287         ASSIGN_FETCH_TYPE(u32, u32, 0),
 288         ASSIGN_FETCH_TYPE(u64, u64, 0),
 289         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 290         ASSIGN_FETCH_TYPE(s16, u16, 1),
 291         ASSIGN_FETCH_TYPE(s32, u32, 1),
 292         ASSIGN_FETCH_TYPE(s64, u64, 1),
 293         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 294         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 295         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 296         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 297
 298         ASSIGN_FETCH_TYPE_END
 299 };
 300
 301 /*
 302  * Allocate new trace_probe and initialize it (including kprobes).
 303  */
 304 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 305                                              const char *event,
 306                                              void *addr,
 307                                              const char *symbol,
 308                                              unsigned long offs,
 309                                              int maxactive,
 310                                              int nargs, bool is_return)
 311 {
 312         struct trace_kprobe *tk;
 313         int ret = -ENOMEM;
 314
 315         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 316         if (!tk)
 317                 return ERR_PTR(ret);
 318
 319         tk->nhit = alloc_percpu(unsigned long);
 320         if (!tk->nhit)
 321                 goto error;
 322
 323         if (symbol) {
 324                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 325                 if (!tk->symbol)
 326                         goto error;
 327                 tk->rp.kp.symbol_name = tk->symbol;
 328                 tk->rp.kp.offset = offs;
 329         } else
 330                 tk->rp.kp.addr = addr;
 331
 332         if (is_return)
 333                 tk->rp.handler = kretprobe_dispatcher;
 334         else
 335                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 336
 337         tk->rp.maxactive = maxactive;
 338
 339         if (!event || !is_good_name(event)) {
 340                 ret = -EINVAL;
 341                 goto error;
 342         }
 343
 344         tk->tp.call.class = &tk->tp.class;
 345         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 346         if (!tk->tp.call.name)
 347                 goto error;
 348
 349         if (!group || !is_good_name(group)) {
 350                 ret = -EINVAL;
 351                 goto error;
 352         }
 353
 354         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 355         if (!tk->tp.class.system)
 356                 goto error;
 357
 358         INIT_LIST_HEAD(&tk->list);
 359         INIT_LIST_HEAD(&tk->tp.files);
 360         return tk;
 361 error:
 362         kfree(tk->tp.call.name);
 363         kfree(tk->symbol);
 364         free_percpu(tk->nhit);
 365         kfree(tk);
 366         return ERR_PTR(ret);
 367 }
 368
 369 static void free_trace_kprobe(struct trace_kprobe *tk)
 370 {
 371         int i;
 372
 373         for (i = 0; i < tk->tp.nr_args; i++)
 374                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 375
 376         kfree(tk->tp.call.class->system);
 377         kfree(tk->tp.call.name);
 378         kfree(tk->symbol);
 379         free_percpu(tk->nhit);
 380         kfree(tk);
 381 }
 382
 383 static struct trace_kprobe *find_trace_kprobe(const char *event,
 384                                               const char *group)
 385 {
 386         struct trace_kprobe *tk;
 387
 388         list_for_each_entry(tk, &probe_list, list)
 389                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 390                     strcmp(tk->tp.call.class->system, group) == 0)
 391                         return tk;
 392         return NULL;
 393 }
 394
 395 static inline int __enable_trace_kprobe(struct trace_kprobe *tk)
 396 {
 397         int ret = 0;
 398
 399         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 400                 if (trace_kprobe_is_return(tk))
 401                         ret = enable_kretprobe(&tk->rp);
 402                 else
 403                         ret = enable_kprobe(&tk->rp.kp);
 404         }
 405
 406         return ret;
 407 }
 408
 409 /*
 410  * Enable trace_probe
 411  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 412  */
 413 static int
 414 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 415 {
 416         struct event_file_link *link;
 417         int ret = 0;
 418
 419         if (file) {
 420                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 421                 if (!link) {
 422                         ret = -ENOMEM;
 423                         goto out;
 424                 }
 425
 426                 link->file = file;
 427                 list_add_tail_rcu(&link->list, &tk->tp.files);
 428
 429                 tk->tp.flags |= TP_FLAG_TRACE;
 430                 ret = __enable_trace_kprobe(tk);
 431                 if (ret) {
 432                         list_del_rcu(&link->list);
 433                         kfree(link);
 434                         tk->tp.flags &= ~TP_FLAG_TRACE;
 435                 }
 436
 437         } else {
 438                 tk->tp.flags |= TP_FLAG_PROFILE;
 439                 ret = __enable_trace_kprobe(tk);
 440                 if (ret)
 441                         tk->tp.flags &= ~TP_FLAG_PROFILE;
 442         }
 443  out:
 444         return ret;
 445 }
 446
 447 /*
 448  * Disable trace_probe
 449  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 450  */
 451 static int
 452 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 453 {
 454         struct event_file_link *link = NULL;
 455         int wait = 0;
 456         int ret = 0;
 457
 458         if (file) {
 459                 link = find_event_file_link(&tk->tp, file);
 460                 if (!link) {
 461                         ret = -EINVAL;
 462                         goto out;
 463                 }
 464
 465                 list_del_rcu(&link->list);
 466                 wait = 1;
 467                 if (!list_empty(&tk->tp.files))
 468                         goto out;
 469
 470                 tk->tp.flags &= ~TP_FLAG_TRACE;
 471         } else
 472                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 473
 474         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 475                 if (trace_kprobe_is_return(tk))
 476                         disable_kretprobe(&tk->rp);
 477                 else
 478                         disable_kprobe(&tk->rp.kp);
 479                 wait = 1;
 480         }
 481
 482         /*
 483          * if tk is not added to any list, it must be a local trace_kprobe
 484          * created with perf_event_open. We don't need to wait for these
 485          * trace_kprobes
 486          */
 487         if (list_empty(&tk->list))
 488                 wait = 0;
 489  out:
 490         if (wait) {
 491                 /*
 492                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 493                  * to ensure disabled (all running handlers are finished).
 494                  * This is not only for kfree(), but also the caller,
 495                  * trace_remove_event_call() supposes it for releasing
 496                  * event_call related objects, which will be accessed in
 497                  * the kprobe_trace_func/kretprobe_trace_func.
 498                  */
 499                 synchronize_sched();
 500                 kfree(link);    /* Ignored if link == NULL */
 501         }
 502
 503         return ret;
 504 }
 505
 506 #if defined(CONFIG_KPROBES_ON_FTRACE) && \
 507         !defined(CONFIG_KPROBE_EVENTS_ON_NOTRACE)
 508 static bool within_notrace_func(struct trace_kprobe *tk)
 509 {
 510         unsigned long offset, size, addr;
 511
 512         addr = trace_kprobe_address(tk);
 513         if (!addr || !kallsyms_lookup_size_offset(addr, &size, &offset))
 514                 return false;
 515
 516         return !ftrace_location_range(addr - offset, addr - offset + size);
 517 }
 518 #else
 519 #define within_notrace_func(tk) (false)
 520 #endif
 521
 522 /* Internal register function - just handle k*probes and flags */
 523 static int __register_trace_kprobe(struct trace_kprobe *tk)
 524 {
 525         int i, ret;
 526
 527         if (trace_probe_is_registered(&tk->tp))
 528                 return -EINVAL;
 529
 530         if (within_notrace_func(tk)) {
 531                 pr_warn("Could not probe notrace function %s\n",
 532                         trace_kprobe_symbol(tk));
 533                 return -EINVAL;
 534         }
 535
 536         for (i = 0; i < tk->tp.nr_args; i++)
 537                 traceprobe_update_arg(&tk->tp.args[i]);
 538
 539         /* Set/clear disabled flag according to tp->flag */
 540         if (trace_probe_is_enabled(&tk->tp))
 541                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 542         else
 543                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 544
 545         if (trace_kprobe_is_return(tk))
 546                 ret = register_kretprobe(&tk->rp);
 547         else
 548                 ret = register_kprobe(&tk->rp.kp);
 549
 550         if (ret == 0)
 551                 tk->tp.flags |= TP_FLAG_REGISTERED;
 552         else {
 553                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 554                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 555                         ret = 0;
 556                 } else if (ret == -EILSEQ) {
 557                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 558                                 tk->rp.kp.addr);
 559                         ret = -EINVAL;
 560                 }
 561         }
 562
 563         return ret;
 564 }
 565
 566 /* Internal unregister function - just handle k*probes and flags */
 567 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 568 {
 569         if (trace_probe_is_registered(&tk->tp)) {
 570                 if (trace_kprobe_is_return(tk))
 571                         unregister_kretprobe(&tk->rp);
 572                 else
 573                         unregister_kprobe(&tk->rp.kp);
 574                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 575                 /* Cleanup kprobe for reuse */
 576                 if (tk->rp.kp.symbol_name)
 577                         tk->rp.kp.addr = NULL;
 578         }
 579 }
 580
 581 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 582 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 583 {
 584         /* Enabled event can not be unregistered */
 585         if (trace_probe_is_enabled(&tk->tp))
 586                 return -EBUSY;
 587
 588         /* Will fail if probe is being used by ftrace or perf */
 589         if (unregister_kprobe_event(tk))
 590                 return -EBUSY;
 591
 592         __unregister_trace_kprobe(tk);
 593         list_del(&tk->list);
 594
 595         return 0;
 596 }
 597
 598 /* Register a trace_probe and probe_event */
 599 static int register_trace_kprobe(struct trace_kprobe *tk)
 600 {
 601         struct trace_kprobe *old_tk;
 602         int ret;
 603
 604         mutex_lock(&probe_lock);
 605
 606         /* Delete old (same name) event if exist */
 607         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 608                         tk->tp.call.class->system);
 609         if (old_tk) {
 610                 ret = unregister_trace_kprobe(old_tk);
 611                 if (ret < 0)
 612                         goto end;
 613                 free_trace_kprobe(old_tk);
 614         }
 615
 616         /* Register new event */
 617         ret = register_kprobe_event(tk);
 618         if (ret) {
 619                 pr_warn("Failed to register probe event(%d)\n", ret);
 620                 goto end;
 621         }
 622
 623         /* Register k*probe */
 624         ret = __register_trace_kprobe(tk);
 625         if (ret < 0)
 626                 unregister_kprobe_event(tk);
 627         else
 628                 list_add_tail(&tk->list, &probe_list);
 629
 630 end:
 631         mutex_unlock(&probe_lock);
 632         return ret;
 633 }
 634
 635 /* Module notifier call back, checking event on the module */
 636 static int trace_kprobe_module_callback(struct notifier_block *nb,
 637                                        unsigned long val, void *data)
 638 {
 639         struct module *mod = data;
 640         struct trace_kprobe *tk;
 641         int ret;
 642
 643         if (val != MODULE_STATE_COMING)
 644                 return NOTIFY_DONE;
 645
 646         /* Update probes on coming module */
 647         mutex_lock(&probe_lock);
 648         list_for_each_entry(tk, &probe_list, list) {
 649                 if (trace_kprobe_within_module(tk, mod)) {
 650                         /* Don't need to check busy - this should have gone. */
 651                         __unregister_trace_kprobe(tk);
 652                         ret = __register_trace_kprobe(tk);
 653                         if (ret)
 654                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 655                                         trace_event_name(&tk->tp.call),
 656                                         mod->name, ret);
 657                 }
 658         }
 659         mutex_unlock(&probe_lock);
 660
 661         return NOTIFY_DONE;
 662 }
 663
 664 static struct notifier_block trace_kprobe_module_nb = {
 665         .notifier_call = trace_kprobe_module_callback,
 666         .priority = 1   /* Invoked after kprobe module callback */
 667 };
 668
 669 /* Convert certain expected symbols into '_' when generating event names */
 670 static inline void sanitize_event_name(char *name)
 671 {
 672         while (*name++ != '\0')
 673                 if (*name == ':' || *name == '.')
 674                         *name = '_';
 675 }
 676
 677 static int create_trace_kprobe(int argc, char **argv)
 678 {
 679         /*
 680          * Argument syntax:
 681          *  - Add kprobe:
 682          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 683          *  - Add kretprobe:
 684          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 685          * Fetch args:
 686          *  $retval     : fetch return value
 687          *  $stack      : fetch stack address
 688          *  $stackN     : fetch Nth of stack (N:0-)
 689          *  $comm       : fetch current task comm
 690          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 691          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 692          *  %REG        : fetch register REG
 693          * Dereferencing memory fetch:
 694          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 695          * Alias name of args:
 696          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 697          * Type of args:
 698          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 699          */
 700         struct trace_kprobe *tk;
 701         int i, ret = 0;
 702         bool is_return = false, is_delete = false;
 703         char *symbol = NULL, *event = NULL, *group = NULL;
 704         int maxactive = 0;
 705         char *arg;
 706         long offset = 0;
 707         void *addr = NULL;
 708         char buf[MAX_EVENT_NAME_LEN];
 709
 710         /* argc must be >= 1 */
 711         if (argv[0][0] == 'p')
 712                 is_return = false;
 713         else if (argv[0][0] == 'r')
 714                 is_return = true;
 715         else if (argv[0][0] == '-')
 716                 is_delete = true;
 717         else {
 718                 pr_info("Probe definition must be started with 'p', 'r' or"
 719                         " '-'.\n");
 720                 return -EINVAL;
 721         }
 722
 723         event = strchr(&argv[0][1], ':');
 724         if (event) {
 725                 event[0] = '\0';
 726                 event++;
 727         }
 728         if (is_return && isdigit(argv[0][1])) {
 729                 ret = kstrtouint(&argv[0][1], 0, &maxactive);
 730                 if (ret) {
 731                         pr_info("Failed to parse maxactive.\n");
 732                         return ret;
 733                 }
 734                 /* kretprobes instances are iterated over via a list. The
 735                  * maximum should stay reasonable.
 736                  */
 737                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 738                         pr_info("Maxactive is too big (%d > %d).\n",
 739                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 740                         return -E2BIG;
 741                 }
 742         }
 743
 744         if (event) {
 745                 if (strchr(event, '/')) {
 746                         group = event;
 747                         event = strchr(group, '/') + 1;
 748                         event[-1] = '\0';
 749                         if (strlen(group) == 0) {
 750                                 pr_info("Group name is not specified\n");
 751                                 return -EINVAL;
 752                         }
 753                 }
 754                 if (strlen(event) == 0) {
 755                         pr_info("Event name is not specified\n");
 756                         return -EINVAL;
 757                 }
 758         }
 759         if (!group)
 760                 group = KPROBE_EVENT_SYSTEM;
 761
 762         if (is_delete) {
 763                 if (!event) {
 764                         pr_info("Delete command needs an event name.\n");
 765                         return -EINVAL;
 766                 }
 767                 mutex_lock(&probe_lock);
 768                 tk = find_trace_kprobe(event, group);
 769                 if (!tk) {
 770                         mutex_unlock(&probe_lock);
 771                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 772                         return -ENOENT;
 773                 }
 774                 /* delete an event */
 775                 ret = unregister_trace_kprobe(tk);
 776                 if (ret == 0)
 777                         free_trace_kprobe(tk);
 778                 mutex_unlock(&probe_lock);
 779                 return ret;
 780         }
 781
 782         if (argc < 2) {
 783                 pr_info("Probe point is not specified.\n");
 784                 return -EINVAL;
 785         }
 786
 787         /* try to parse an address. if that fails, try to read the
 788          * input as a symbol. */
 789         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 790                 /* a symbol specified */
 791                 symbol = argv[1];
 792                 /* TODO: support .init module functions */
 793                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 794                 if (ret || offset < 0 || offset > UINT_MAX) {
 795                         pr_info("Failed to parse either an address or a symbol.\n");
 796                         return ret;
 797                 }
 798                 if (offset && is_return &&
 799                     !kprobe_on_func_entry(NULL, symbol, offset)) {
 800                         pr_info("Given offset is not valid for return probe.\n");
 801                         return -EINVAL;
 802                 }
 803         }
 804         argc -= 2; argv += 2;
 805
 806         /* setup a probe */
 807         if (!event) {
 808                 /* Make a new event name */
 809                 if (symbol)
 810                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 811                                  is_return ? 'r' : 'p', symbol, offset);
 812                 else
 813                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 814                                  is_return ? 'r' : 'p', addr);
 815                 sanitize_event_name(buf);
 816                 event = buf;
 817         }
 818         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 819                                argc, is_return);
 820         if (IS_ERR(tk)) {
 821                 pr_info("Failed to allocate trace_probe.(%d)\n",
 822                         (int)PTR_ERR(tk));
 823                 return PTR_ERR(tk);
 824         }
 825
 826         /* parse arguments */
 827         ret = 0;
 828         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 829                 struct probe_arg *parg = &tk->tp.args[i];
 830
 831                 /* Increment count for freeing args in error case */
 832                 tk->tp.nr_args++;
 833
 834                 /* Parse argument name */
 835                 arg = strchr(argv[i], '=');
 836                 if (arg) {
 837                         *arg++ = '\0';
 838                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 839                 } else {
 840                         arg = argv[i];
 841                         /* If argument name is omitted, set "argN" */
 842                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 843                         parg->name = kstrdup(buf, GFP_KERNEL);
 844                 }
 845
 846                 if (!parg->name) {
 847                         pr_info("Failed to allocate argument[%d] name.\n", i);
 848                         ret = -ENOMEM;
 849                         goto error;
 850                 }
 851
 852                 if (!is_good_name(parg->name)) {
 853                         pr_info("Invalid argument[%d] name: %s\n",
 854                                 i, parg->name);
 855                         ret = -EINVAL;
 856                         goto error;
 857                 }
 858
 859                 if (traceprobe_conflict_field_name(parg->name,
 860                                                         tk->tp.args, i)) {
 861                         pr_info("Argument[%d] name '%s' conflicts with "
 862                                 "another field.\n", i, argv[i]);
 863                         ret = -EINVAL;
 864                         goto error;
 865                 }
 866
 867                 /* Parse fetch argument */
 868                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 869                                                 is_return, true,
 870                                                 kprobes_fetch_type_table);
 871                 if (ret) {
 872                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 873                         goto error;
 874                 }
 875         }
 876
 877         ret = register_trace_kprobe(tk);
 878         if (ret)
 879                 goto error;
 880         return 0;
 881
 882 error:
 883         free_trace_kprobe(tk);
 884         return ret;
 885 }
 886
 887 static int release_all_trace_kprobes(void)
 888 {
 889         struct trace_kprobe *tk;
 890         int ret = 0;
 891
 892         mutex_lock(&probe_lock);
 893         /* Ensure no probe is in use. */
 894         list_for_each_entry(tk, &probe_list, list)
 895                 if (trace_probe_is_enabled(&tk->tp)) {
 896                         ret = -EBUSY;
 897                         goto end;
 898                 }
 899         /* TODO: Use batch unregistration */
 900         while (!list_empty(&probe_list)) {
 901                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 902                 ret = unregister_trace_kprobe(tk);
 903                 if (ret)
 904                         goto end;
 905                 free_trace_kprobe(tk);
 906         }
 907
 908 end:
 909         mutex_unlock(&probe_lock);
 910
 911         return ret;
 912 }
 913
 914 /* Probes listing interfaces */
 915 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 916 {
 917         mutex_lock(&probe_lock);
 918         return seq_list_start(&probe_list, *pos);
 919 }
 920
 921 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 922 {
 923         return seq_list_next(v, &probe_list, pos);
 924 }
 925
 926 static void probes_seq_stop(struct seq_file *m, void *v)
 927 {
 928         mutex_unlock(&probe_lock);
 929 }
 930
 931 static int probes_seq_show(struct seq_file *m, void *v)
 932 {
 933         struct trace_kprobe *tk = v;
 934         int i;
 935
 936         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 937         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 938                         trace_event_name(&tk->tp.call));
 939
 940         if (!tk->symbol)
 941                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 942         else if (tk->rp.kp.offset)
 943                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 944                            tk->rp.kp.offset);
 945         else
 946                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 947
 948         for (i = 0; i < tk->tp.nr_args; i++)
 949                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 950         seq_putc(m, '\n');
 951
 952         return 0;
 953 }
 954
 955 static const struct seq_operations probes_seq_op = {
 956         .start  = probes_seq_start,
 957         .next   = probes_seq_next,
 958         .stop   = probes_seq_stop,
 959         .show   = probes_seq_show
 960 };
 961
 962 static int probes_open(struct inode *inode, struct file *file)
 963 {
 964         int ret;
 965
 966         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 967                 ret = release_all_trace_kprobes();
 968                 if (ret < 0)
 969                         return ret;
 970         }
 971
 972         return seq_open(file, &probes_seq_op);
 973 }
 974
 975 static ssize_t probes_write(struct file *file, const char __user *buffer,
 976                             size_t count, loff_t *ppos)
 977 {
 978         return trace_parse_run_command(file, buffer, count, ppos,
 979                                        create_trace_kprobe);
 980 }
 981
 982 static const struct file_operations kprobe_events_ops = {
 983         .owner          = THIS_MODULE,
 984         .open           = probes_open,
 985         .read           = seq_read,
 986         .llseek         = seq_lseek,
 987         .release        = seq_release,
 988         .write          = probes_write,
 989 };
 990
 991 /* Probes profiling interfaces */
 992 static int probes_profile_seq_show(struct seq_file *m, void *v)
 993 {
 994         struct trace_kprobe *tk = v;
 995
 996         seq_printf(m, "  %-44s %15lu %15lu\n",
 997                    trace_event_name(&tk->tp.call),
 998                    trace_kprobe_nhit(tk),
 999                    tk->rp.kp.nmissed);
1000
1001         return 0;
1002 }
1003
1004 static const struct seq_operations profile_seq_op = {
1005         .start  = probes_seq_start,
1006         .next   = probes_seq_next,
1007         .stop   = probes_seq_stop,
1008         .show   = probes_profile_seq_show
1009 };
1010
1011 static int profile_open(struct inode *inode, struct file *file)
1012 {
1013         return seq_open(file, &profile_seq_op);
1014 }
1015
1016 static const struct file_operations kprobe_profile_ops = {
1017         .owner          = THIS_MODULE,
1018         .open           = profile_open,
1019         .read           = seq_read,
1020         .llseek         = seq_lseek,
1021         .release        = seq_release,
1022 };
1023
1024 /* Kprobe handler */
1025 static nokprobe_inline void
1026 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1027                     struct trace_event_file *trace_file)
1028 {
1029         struct kprobe_trace_entry_head *entry;
1030         struct ring_buffer_event *event;
1031         struct ring_buffer *buffer;
1032         int size, dsize, pc;
1033         unsigned long irq_flags;
1034         struct trace_event_call *call = &tk->tp.call;
1035
1036         WARN_ON(call != trace_file->event_call);
1037
1038         if (trace_trigger_soft_disabled(trace_file))
1039                 return;
1040
1041         local_save_flags(irq_flags);
1042         pc = preempt_count();
1043
1044         dsize = __get_data_size(&tk->tp, regs);
1045         size = sizeof(*entry) + tk->tp.size + dsize;
1046
1047         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1048                                                 call->event.type,
1049                                                 size, irq_flags, pc);
1050         if (!event)
1051                 return;
1052
1053         entry = ring_buffer_event_data(event);
1054         entry->ip = (unsigned long)tk->rp.kp.addr;
1055         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1056
1057         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1058                                          entry, irq_flags, pc, regs);
1059 }
1060
1061 static void
1062 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1063 {
1064         struct event_file_link *link;
1065
1066         list_for_each_entry_rcu(link, &tk->tp.files, list)
1067                 __kprobe_trace_func(tk, regs, link->file);
1068 }
1069 NOKPROBE_SYMBOL(kprobe_trace_func);
1070
1071 /* Kretprobe handler */
1072 static nokprobe_inline void
1073 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1074                        struct pt_regs *regs,
1075                        struct trace_event_file *trace_file)
1076 {
1077         struct kretprobe_trace_entry_head *entry;
1078         struct ring_buffer_event *event;
1079         struct ring_buffer *buffer;
1080         int size, pc, dsize;
1081         unsigned long irq_flags;
1082         struct trace_event_call *call = &tk->tp.call;
1083
1084         WARN_ON(call != trace_file->event_call);
1085
1086         if (trace_trigger_soft_disabled(trace_file))
1087                 return;
1088
1089         local_save_flags(irq_flags);
1090         pc = preempt_count();
1091
1092         dsize = __get_data_size(&tk->tp, regs);
1093         size = sizeof(*entry) + tk->tp.size + dsize;
1094
1095         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1096                                                 call->event.type,
1097                                                 size, irq_flags, pc);
1098         if (!event)
1099                 return;
1100
1101         entry = ring_buffer_event_data(event);
1102         entry->func = (unsigned long)tk->rp.kp.addr;
1103         entry->ret_ip = (unsigned long)ri->ret_addr;
1104         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1105
1106         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1107                                          entry, irq_flags, pc, regs);
1108 }
1109
1110 static void
1111 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1112                      struct pt_regs *regs)
1113 {
1114         struct event_file_link *link;
1115
1116         list_for_each_entry_rcu(link, &tk->tp.files, list)
1117                 __kretprobe_trace_func(tk, ri, regs, link->file);
1118 }
1119 NOKPROBE_SYMBOL(kretprobe_trace_func);
1120
1121 /* Event entry printers */
1122 static enum print_line_t
1123 print_kprobe_event(struct trace_iterator *iter, int flags,
1124                    struct trace_event *event)
1125 {
1126         struct kprobe_trace_entry_head *field;
1127         struct trace_seq *s = &iter->seq;
1128         struct trace_probe *tp;
1129         u8 *data;
1130         int i;
1131
1132         field = (struct kprobe_trace_entry_head *)iter->ent;
1133         tp = container_of(event, struct trace_probe, call.event);
1134
1135         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1136
1137         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1138                 goto out;
1139
1140         trace_seq_putc(s, ')');
1141
1142         data = (u8 *)&field[1];
1143         for (i = 0; i < tp->nr_args; i++)
1144                 if (!tp->args[i].type->print(s, tp->args[i].name,
1145                                              data + tp->args[i].offset, field))
1146                         goto out;
1147
1148         trace_seq_putc(s, '\n');
1149  out:
1150         return trace_handle_return(s);
1151 }
1152
1153 static enum print_line_t
1154 print_kretprobe_event(struct trace_iterator *iter, int flags,
1155                       struct trace_event *event)
1156 {
1157         struct kretprobe_trace_entry_head *field;
1158         struct trace_seq *s = &iter->seq;
1159         struct trace_probe *tp;
1160         u8 *data;
1161         int i;
1162
1163         field = (struct kretprobe_trace_entry_head *)iter->ent;
1164         tp = container_of(event, struct trace_probe, call.event);
1165
1166         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1167
1168         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1169                 goto out;
1170
1171         trace_seq_puts(s, " <- ");
1172
1173         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1174                 goto out;
1175
1176         trace_seq_putc(s, ')');
1177
1178         data = (u8 *)&field[1];
1179         for (i = 0; i < tp->nr_args; i++)
1180                 if (!tp->args[i].type->print(s, tp->args[i].name,
1181                                              data + tp->args[i].offset, field))
1182                         goto out;
1183
1184         trace_seq_putc(s, '\n');
1185
1186  out:
1187         return trace_handle_return(s);
1188 }
1189
1190
1191 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1192 {
1193         int ret, i;
1194         struct kprobe_trace_entry_head field;
1195         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1196
1197         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1198         /* Set argument names as fields */
1199         for (i = 0; i < tk->tp.nr_args; i++) {
1200                 struct probe_arg *parg = &tk->tp.args[i];
1201
1202                 ret = trace_define_field(event_call, parg->type->fmttype,
1203                                          parg->name,
1204                                          sizeof(field) + parg->offset,
1205                                          parg->type->size,
1206                                          parg->type->is_signed,
1207                                          FILTER_OTHER);
1208                 if (ret)
1209                         return ret;
1210         }
1211         return 0;
1212 }
1213
1214 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1215 {
1216         int ret, i;
1217         struct kretprobe_trace_entry_head field;
1218         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1219
1220         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1221         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1222         /* Set argument names as fields */
1223         for (i = 0; i < tk->tp.nr_args; i++) {
1224                 struct probe_arg *parg = &tk->tp.args[i];
1225
1226                 ret = trace_define_field(event_call, parg->type->fmttype,
1227                                          parg->name,
1228                                          sizeof(field) + parg->offset,
1229                                          parg->type->size,
1230                                          parg->type->is_signed,
1231                                          FILTER_OTHER);
1232                 if (ret)
1233                         return ret;
1234         }
1235         return 0;
1236 }
1237
1238 #ifdef CONFIG_PERF_EVENTS
1239
1240 /* Kprobe profile handler */
1241 static int
1242 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1243 {
1244         struct trace_event_call *call = &tk->tp.call;
1245         struct kprobe_trace_entry_head *entry;
1246         struct hlist_head *head;
1247         int size, __size, dsize;
1248         int rctx;
1249
1250         if (bpf_prog_array_valid(call)) {
1251                 unsigned long orig_ip = instruction_pointer(regs);
1252                 int ret;
1253
1254                 ret = trace_call_bpf(call, regs);
1255
1256                 /*
1257                  * We need to check and see if we modified the pc of the
1258                  * pt_regs, and if so return 1 so that we don't do the
1259                  * single stepping.
1260                  */
1261                 if (orig_ip != instruction_pointer(regs))
1262                         return 1;
1263                 if (!ret)
1264                         return 0;
1265         }
1266
1267         head = this_cpu_ptr(call->perf_events);
1268         if (hlist_empty(head))
1269                 return 0;
1270
1271         dsize = __get_data_size(&tk->tp, regs);
1272         __size = sizeof(*entry) + tk->tp.size + dsize;
1273         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1274         size -= sizeof(u32);
1275
1276         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1277         if (!entry)
1278                 return 0;
1279
1280         entry->ip = (unsigned long)tk->rp.kp.addr;
1281         memset(&entry[1], 0, dsize);
1282         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1283         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1284                               head, NULL);
1285         return 0;
1286 }
1287 NOKPROBE_SYMBOL(kprobe_perf_func);
1288
1289 /* Kretprobe profile handler */
1290 static void
1291 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1292                     struct pt_regs *regs)
1293 {
1294         struct trace_event_call *call = &tk->tp.call;
1295         struct kretprobe_trace_entry_head *entry;
1296         struct hlist_head *head;
1297         int size, __size, dsize;
1298         int rctx;
1299
1300         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1301                 return;
1302
1303         head = this_cpu_ptr(call->perf_events);
1304         if (hlist_empty(head))
1305                 return;
1306
1307         dsize = __get_data_size(&tk->tp, regs);
1308         __size = sizeof(*entry) + tk->tp.size + dsize;
1309         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1310         size -= sizeof(u32);
1311
1312         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1313         if (!entry)
1314                 return;
1315
1316         entry->func = (unsigned long)tk->rp.kp.addr;
1317         entry->ret_ip = (unsigned long)ri->ret_addr;
1318         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1319         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1320                               head, NULL);
1321 }
1322 NOKPROBE_SYMBOL(kretprobe_perf_func);
1323
1324 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1325                         const char **symbol, u64 *probe_offset,
1326                         u64 *probe_addr, bool perf_type_tracepoint)
1327 {
1328         const char *pevent = trace_event_name(event->tp_event);
1329         const char *group = event->tp_event->class->system;
1330         struct trace_kprobe *tk;
1331
1332         if (perf_type_tracepoint)
1333                 tk = find_trace_kprobe(pevent, group);
1334         else
1335                 tk = event->tp_event->data;
1336         if (!tk)
1337                 return -EINVAL;
1338
1339         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1340                                               : BPF_FD_TYPE_KPROBE;
1341         if (tk->symbol) {
1342                 *symbol = tk->symbol;
1343                 *probe_offset = tk->rp.kp.offset;
1344                 *probe_addr = 0;
1345         } else {
1346                 *symbol = NULL;
1347                 *probe_offset = 0;
1348                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1349         }
1350         return 0;
1351 }
1352 #endif  /* CONFIG_PERF_EVENTS */
1353
1354 /*
1355  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1356  *
1357  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1358  * lockless, but we can't race with this __init function.
1359  */
1360 static int kprobe_register(struct trace_event_call *event,
1361                            enum trace_reg type, void *data)
1362 {
1363         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1364         struct trace_event_file *file = data;
1365
1366         switch (type) {
1367         case TRACE_REG_REGISTER:
1368                 return enable_trace_kprobe(tk, file);
1369         case TRACE_REG_UNREGISTER:
1370                 return disable_trace_kprobe(tk, file);
1371
1372 #ifdef CONFIG_PERF_EVENTS
1373         case TRACE_REG_PERF_REGISTER:
1374                 return enable_trace_kprobe(tk, NULL);
1375         case TRACE_REG_PERF_UNREGISTER:
1376                 return disable_trace_kprobe(tk, NULL);
1377         case TRACE_REG_PERF_OPEN:
1378         case TRACE_REG_PERF_CLOSE:
1379         case TRACE_REG_PERF_ADD:
1380         case TRACE_REG_PERF_DEL:
1381                 return 0;
1382 #endif
1383         }
1384         return 0;
1385 }
1386
1387 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1388 {
1389         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1390         int ret = 0;
1391
1392         raw_cpu_inc(*tk->nhit);
1393
1394         if (tk->tp.flags & TP_FLAG_TRACE)
1395                 kprobe_trace_func(tk, regs);
1396 #ifdef CONFIG_PERF_EVENTS
1397         if (tk->tp.flags & TP_FLAG_PROFILE)
1398                 ret = kprobe_perf_func(tk, regs);
1399 #endif
1400         return ret;
1401 }
1402 NOKPROBE_SYMBOL(kprobe_dispatcher);
1403
1404 static int
1405 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1406 {
1407         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1408
1409         raw_cpu_inc(*tk->nhit);
1410
1411         if (tk->tp.flags & TP_FLAG_TRACE)
1412                 kretprobe_trace_func(tk, ri, regs);
1413 #ifdef CONFIG_PERF_EVENTS
1414         if (tk->tp.flags & TP_FLAG_PROFILE)
1415                 kretprobe_perf_func(tk, ri, regs);
1416 #endif
1417         return 0;       /* We don't tweek kernel, so just return 0 */
1418 }
1419 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1420
1421 static struct trace_event_functions kretprobe_funcs = {
1422         .trace          = print_kretprobe_event
1423 };
1424
1425 static struct trace_event_functions kprobe_funcs = {
1426         .trace          = print_kprobe_event
1427 };
1428
1429 static inline void init_trace_event_call(struct trace_kprobe *tk,
1430                                          struct trace_event_call *call)
1431 {
1432         INIT_LIST_HEAD(&call->class->fields);
1433         if (trace_kprobe_is_return(tk)) {
1434                 call->event.funcs = &kretprobe_funcs;
1435                 call->class->define_fields = kretprobe_event_define_fields;
1436         } else {
1437                 call->event.funcs = &kprobe_funcs;
1438                 call->class->define_fields = kprobe_event_define_fields;
1439         }
1440
1441         call->flags = TRACE_EVENT_FL_KPROBE;
1442         call->class->reg = kprobe_register;
1443         call->data = tk;
1444 }
1445
1446 static int register_kprobe_event(struct trace_kprobe *tk)
1447 {
1448         struct trace_event_call *call = &tk->tp.call;
1449         int ret = 0;
1450
1451         init_trace_event_call(tk, call);
1452
1453         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1454                 return -ENOMEM;
1455         ret = register_trace_event(&call->event);
1456         if (!ret) {
1457                 kfree(call->print_fmt);
1458                 return -ENODEV;
1459         }
1460         ret = trace_add_event_call(call);
1461         if (ret) {
1462                 pr_info("Failed to register kprobe event: %s\n",
1463                         trace_event_name(call));
1464                 kfree(call->print_fmt);
1465                 unregister_trace_event(&call->event);
1466         }
1467         return ret;
1468 }
1469
1470 static int unregister_kprobe_event(struct trace_kprobe *tk)
1471 {
1472         int ret;
1473
1474         /* tp->event is unregistered in trace_remove_event_call() */
1475         ret = trace_remove_event_call(&tk->tp.call);
1476         if (!ret)
1477                 kfree(tk->tp.call.print_fmt);
1478         return ret;
1479 }
1480
1481 #ifdef CONFIG_PERF_EVENTS
1482 /* create a trace_kprobe, but don't add it to global lists */
1483 struct trace_event_call *
1484 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1485                           bool is_return)
1486 {
1487         struct trace_kprobe *tk;
1488         int ret;
1489         char *event;
1490
1491         /*
1492          * local trace_kprobes are not added to probe_list, so they are never
1493          * searched in find_trace_kprobe(). Therefore, there is no concern of
1494          * duplicated name here.
1495          */
1496         event = func ? func : "DUMMY_EVENT";
1497
1498         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1499                                 offs, 0 /* maxactive */, 0 /* nargs */,
1500                                 is_return);
1501
1502         if (IS_ERR(tk)) {
1503                 pr_info("Failed to allocate trace_probe.(%d)\n",
1504                         (int)PTR_ERR(tk));
1505                 return ERR_CAST(tk);
1506         }
1507
1508         init_trace_event_call(tk, &tk->tp.call);
1509
1510         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1511                 ret = -ENOMEM;
1512                 goto error;
1513         }
1514
1515         ret = __register_trace_kprobe(tk);
1516         if (ret < 0) {
1517                 kfree(tk->tp.call.print_fmt);
1518                 goto error;
1519         }
1520
1521         return &tk->tp.call;
1522 error:
1523         free_trace_kprobe(tk);
1524         return ERR_PTR(ret);
1525 }
1526
1527 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1528 {
1529         struct trace_kprobe *tk;
1530
1531         tk = container_of(event_call, struct trace_kprobe, tp.call);
1532
1533         if (trace_probe_is_enabled(&tk->tp)) {
1534                 WARN_ON(1);
1535                 return;
1536         }
1537
1538         __unregister_trace_kprobe(tk);
1539
1540         kfree(tk->tp.call.print_fmt);
1541         free_trace_kprobe(tk);
1542 }
1543 #endif /* CONFIG_PERF_EVENTS */
1544
1545 /* Make a tracefs interface for controlling probe points */
1546 static __init int init_kprobe_trace(void)
1547 {
1548         struct dentry *d_tracer;
1549         struct dentry *entry;
1550
1551         if (register_module_notifier(&trace_kprobe_module_nb))
1552                 return -EINVAL;
1553
1554         d_tracer = tracing_init_dentry();
1555         if (IS_ERR(d_tracer))
1556                 return 0;
1557
1558         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1559                                     NULL, &kprobe_events_ops);
1560
1561         /* Event list interface */
1562         if (!entry)
1563                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1564
1565         /* Profile interface */
1566         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1567                                     NULL, &kprobe_profile_ops);
1568
1569         if (!entry)
1570                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1571         return 0;
1572 }
1573 fs_initcall(init_kprobe_trace);
1574
1575
1576 #ifdef CONFIG_FTRACE_STARTUP_TEST
1577 static __init struct trace_event_file *
1578 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1579 {
1580         struct trace_event_file *file;
1581
1582         list_for_each_entry(file, &tr->events, list)
1583                 if (file->event_call == &tk->tp.call)
1584                         return file;
1585
1586         return NULL;
1587 }
1588
1589 /*
1590  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1591  * stage, we can do this lockless.
1592  */
1593 static __init int kprobe_trace_self_tests_init(void)
1594 {
1595         int ret, warn = 0;
1596         int (*target)(int, int, int, int, int, int);
1597         struct trace_kprobe *tk;
1598         struct trace_event_file *file;
1599
1600         if (tracing_is_disabled())
1601                 return -ENODEV;
1602
1603         target = kprobe_trace_selftest_target;
1604
1605         pr_info("Testing kprobe tracing: ");
1606
1607         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1608                                 "$stack $stack0 +0($stack)",
1609                                 create_trace_kprobe);
1610         if (WARN_ON_ONCE(ret)) {
1611                 pr_warn("error on probing function entry.\n");
1612                 warn++;
1613         } else {
1614                 /* Enable trace point */
1615                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1616                 if (WARN_ON_ONCE(tk == NULL)) {
1617                         pr_warn("error on getting new probe.\n");
1618                         warn++;
1619                 } else {
1620                         file = find_trace_probe_file(tk, top_trace_array());
1621                         if (WARN_ON_ONCE(file == NULL)) {
1622                                 pr_warn("error on getting probe file.\n");
1623                                 warn++;
1624                         } else
1625                                 enable_trace_kprobe(tk, file);
1626                 }
1627         }
1628
1629         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1630                                 "$retval", create_trace_kprobe);
1631         if (WARN_ON_ONCE(ret)) {
1632                 pr_warn("error on probing function return.\n");
1633                 warn++;
1634         } else {
1635                 /* Enable trace point */
1636                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1637                 if (WARN_ON_ONCE(tk == NULL)) {
1638                         pr_warn("error on getting 2nd new probe.\n");
1639                         warn++;
1640                 } else {
1641                         file = find_trace_probe_file(tk, top_trace_array());
1642                         if (WARN_ON_ONCE(file == NULL)) {
1643                                 pr_warn("error on getting probe file.\n");
1644                                 warn++;
1645                         } else
1646                                 enable_trace_kprobe(tk, file);
1647                 }
1648         }
1649
1650         if (warn)
1651                 goto end;
1652
1653         ret = target(1, 2, 3, 4, 5, 6);
1654
1655         /*
1656          * Not expecting an error here, the check is only to prevent the
1657          * optimizer from removing the call to target() as otherwise there
1658          * are no side-effects and the call is never performed.
1659          */
1660         if (ret != 21)
1661                 warn++;
1662
1663         /* Disable trace points before removing it */
1664         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1665         if (WARN_ON_ONCE(tk == NULL)) {
1666                 pr_warn("error on getting test probe.\n");
1667                 warn++;
1668         } else {
1669                 if (trace_kprobe_nhit(tk) != 1) {
1670                         pr_warn("incorrect number of testprobe hits\n");
1671                         warn++;
1672                 }
1673
1674                 file = find_trace_probe_file(tk, top_trace_array());
1675                 if (WARN_ON_ONCE(file == NULL)) {
1676                         pr_warn("error on getting probe file.\n");
1677                         warn++;
1678                 } else
1679                         disable_trace_kprobe(tk, file);
1680         }
1681
1682         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1683         if (WARN_ON_ONCE(tk == NULL)) {
1684                 pr_warn("error on getting 2nd test probe.\n");
1685                 warn++;
1686         } else {
1687                 if (trace_kprobe_nhit(tk) != 1) {
1688                         pr_warn("incorrect number of testprobe2 hits\n");
1689                         warn++;
1690                 }
1691
1692                 file = find_trace_probe_file(tk, top_trace_array());
1693                 if (WARN_ON_ONCE(file == NULL)) {
1694                         pr_warn("error on getting probe file.\n");
1695                         warn++;
1696                 } else
1697                         disable_trace_kprobe(tk, file);
1698         }
1699
1700         ret = trace_run_command("-:testprobe", create_trace_kprobe);
1701         if (WARN_ON_ONCE(ret)) {
1702                 pr_warn("error on deleting a probe.\n");
1703                 warn++;
1704         }
1705
1706         ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1707         if (WARN_ON_ONCE(ret)) {
1708                 pr_warn("error on deleting a probe.\n");
1709                 warn++;
1710         }
1711
1712 end:
1713         release_all_trace_kprobes();
1714         /*
1715          * Wait for the optimizer work to finish. Otherwise it might fiddle
1716          * with probes in already freed __init text.
1717          */
1718         wait_for_kprobe_optimizer();
1719         if (warn)
1720                 pr_cont("NG: Some tests are failed. Please check them.\n");
1721         else
1722                 pr_cont("OK\n");
1723         return 0;
1724 }
1725
1726 late_initcall(kprobe_trace_self_tests_init);
1727
1728 #endif