kernel/trace/trace_kprobe.c

   1 /*
   2  * Kprobes-based tracing events
   3  *
   4  * Created by Masami Hiramatsu <mhiramat@redhat.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19 #define pr_fmt(fmt)     "trace_kprobe: " fmt
  20
  21 #include <linux/module.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/rculist.h>
  24 #include <linux/error-injection.h>
  25
  26 #include "trace_probe.h"
  27
  28 #define KPROBE_EVENT_SYSTEM "kprobes"
  29 #define KRETPROBE_MAXACTIVE_MAX 4096
  30
  31 /**
  32  * Kprobe event core functions
  33  */
  34 struct trace_kprobe {
  35         struct list_head        list;
  36         struct kretprobe        rp;     /* Use rp.kp for kprobe use */
  37         unsigned long __percpu *nhit;
  38         const char              *symbol;        /* symbol name */
  39         struct trace_probe      tp;
  40 };
  41
  42 #define SIZEOF_TRACE_KPROBE(n)                          \
  43         (offsetof(struct trace_kprobe, tp.args) +       \
  44         (sizeof(struct probe_arg) * (n)))
  45
  46 static nokprobe_inline bool trace_kprobe_is_return(struct trace_kprobe *tk)
  47 {
  48         return tk->rp.handler != NULL;
  49 }
  50
  51 static nokprobe_inline const char *trace_kprobe_symbol(struct trace_kprobe *tk)
  52 {
  53         return tk->symbol ? tk->symbol : "unknown";
  54 }
  55
  56 static nokprobe_inline unsigned long trace_kprobe_offset(struct trace_kprobe *tk)
  57 {
  58         return tk->rp.kp.offset;
  59 }
  60
  61 static nokprobe_inline bool trace_kprobe_has_gone(struct trace_kprobe *tk)
  62 {
  63         return !!(kprobe_gone(&tk->rp.kp));
  64 }
  65
  66 static nokprobe_inline bool trace_kprobe_within_module(struct trace_kprobe *tk,
  67                                                  struct module *mod)
  68 {
  69         int len = strlen(mod->name);
  70         const char *name = trace_kprobe_symbol(tk);
  71         return strncmp(mod->name, name, len) == 0 && name[len] == ':';
  72 }
  73
  74 static nokprobe_inline bool trace_kprobe_is_on_module(struct trace_kprobe *tk)
  75 {
  76         return !!strchr(trace_kprobe_symbol(tk), ':');
  77 }
  78
  79 static nokprobe_inline unsigned long trace_kprobe_nhit(struct trace_kprobe *tk)
  80 {
  81         unsigned long nhit = 0;
  82         int cpu;
  83
  84         for_each_possible_cpu(cpu)
  85                 nhit += *per_cpu_ptr(tk->nhit, cpu);
  86
  87         return nhit;
  88 }
  89
  90 bool trace_kprobe_on_func_entry(struct trace_event_call *call)
  91 {
  92         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
  93
  94         return kprobe_on_func_entry(tk->rp.kp.addr,
  95                         tk->rp.kp.addr ? NULL : tk->rp.kp.symbol_name,
  96                         tk->rp.kp.addr ? 0 : tk->rp.kp.offset);
  97 }
  98
  99 bool trace_kprobe_error_injectable(struct trace_event_call *call)
 100 {
 101         struct trace_kprobe *tk = (struct trace_kprobe *)call->data;
 102         unsigned long addr;
 103
 104         if (tk->symbol) {
 105                 addr = (unsigned long)
 106                         kallsyms_lookup_name(trace_kprobe_symbol(tk));
 107                 addr += tk->rp.kp.offset;
 108         } else {
 109                 addr = (unsigned long)tk->rp.kp.addr;
 110         }
 111         return within_error_injection_list(addr);
 112 }
 113
 114 static int register_kprobe_event(struct trace_kprobe *tk);
 115 static int unregister_kprobe_event(struct trace_kprobe *tk);
 116
 117 static DEFINE_MUTEX(probe_lock);
 118 static LIST_HEAD(probe_list);
 119
 120 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs);
 121 static int kretprobe_dispatcher(struct kretprobe_instance *ri,
 122                                 struct pt_regs *regs);
 123
 124 /* Memory fetching by symbol */
 125 struct symbol_cache {
 126         char            *symbol;
 127         long            offset;
 128         unsigned long   addr;
 129 };
 130
 131 unsigned long update_symbol_cache(struct symbol_cache *sc)
 132 {
 133         sc->addr = (unsigned long)kallsyms_lookup_name(sc->symbol);
 134
 135         if (sc->addr)
 136                 sc->addr += sc->offset;
 137
 138         return sc->addr;
 139 }
 140
 141 void free_symbol_cache(struct symbol_cache *sc)
 142 {
 143         kfree(sc->symbol);
 144         kfree(sc);
 145 }
 146
 147 struct symbol_cache *alloc_symbol_cache(const char *sym, long offset)
 148 {
 149         struct symbol_cache *sc;
 150
 151         if (!sym || strlen(sym) == 0)
 152                 return NULL;
 153
 154         sc = kzalloc(sizeof(struct symbol_cache), GFP_KERNEL);
 155         if (!sc)
 156                 return NULL;
 157
 158         sc->symbol = kstrdup(sym, GFP_KERNEL);
 159         if (!sc->symbol) {
 160                 kfree(sc);
 161                 return NULL;
 162         }
 163         sc->offset = offset;
 164         update_symbol_cache(sc);
 165
 166         return sc;
 167 }
 168
 169 /*
 170  * Kprobes-specific fetch functions
 171  */
 172 #define DEFINE_FETCH_stack(type)                                        \
 173 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 174                                           void *offset, void *dest)     \
 175 {                                                                       \
 176         *(type *)dest = (type)regs_get_kernel_stack_nth(regs,           \
 177                                 (unsigned int)((unsigned long)offset)); \
 178 }                                                                       \
 179 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(stack, type));
 180
 181 DEFINE_BASIC_FETCH_FUNCS(stack)
 182 /* No string on the stack entry */
 183 #define fetch_stack_string      NULL
 184 #define fetch_stack_string_size NULL
 185
 186 #define DEFINE_FETCH_memory(type)                                       \
 187 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 188                                           void *addr, void *dest)       \
 189 {                                                                       \
 190         type retval;                                                    \
 191         if (probe_kernel_address(addr, retval))                         \
 192                 *(type *)dest = 0;                                      \
 193         else                                                            \
 194                 *(type *)dest = retval;                                 \
 195 }                                                                       \
 196 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, type));
 197
 198 DEFINE_BASIC_FETCH_FUNCS(memory)
 199 /*
 200  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 201  * length and relative data location.
 202  */
 203 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 204                                             void *addr, void *dest)
 205 {
 206         int maxlen = get_rloc_len(*(u32 *)dest);
 207         u8 *dst = get_rloc_data(dest);
 208         long ret;
 209
 210         if (!maxlen)
 211                 return;
 212
 213         /*
 214          * Try to get string again, since the string can be changed while
 215          * probing.
 216          */
 217         ret = strncpy_from_unsafe(dst, addr, maxlen);
 218
 219         if (ret < 0) {  /* Failed to fetch string */
 220                 dst[0] = '\0';
 221                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(*(u32 *)dest));
 222         } else {
 223                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(*(u32 *)dest));
 224         }
 225 }
 226 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string));
 227
 228 /* Return the length of string -- including null terminal byte */
 229 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 230                                                  void *addr, void *dest)
 231 {
 232         mm_segment_t old_fs;
 233         int ret, len = 0;
 234         u8 c;
 235
 236         old_fs = get_fs();
 237         set_fs(KERNEL_DS);
 238         pagefault_disable();
 239
 240         do {
 241                 ret = __copy_from_user_inatomic(&c, (u8 *)addr + len, 1);
 242                 len++;
 243         } while (c && ret == 0 && len < MAX_STRING_SIZE);
 244
 245         pagefault_enable();
 246         set_fs(old_fs);
 247
 248         if (ret < 0)    /* Failed to check the length */
 249                 *(u32 *)dest = 0;
 250         else
 251                 *(u32 *)dest = len;
 252 }
 253 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(memory, string_size));
 254
 255 #define DEFINE_FETCH_symbol(type)                                       \
 256 void FETCH_FUNC_NAME(symbol, type)(struct pt_regs *regs, void *data, void *dest)\
 257 {                                                                       \
 258         struct symbol_cache *sc = data;                                 \
 259         if (sc->addr)                                                   \
 260                 fetch_memory_##type(regs, (void *)sc->addr, dest);      \
 261         else                                                            \
 262                 *(type *)dest = 0;                                      \
 263 }                                                                       \
 264 NOKPROBE_SYMBOL(FETCH_FUNC_NAME(symbol, type));
 265
 266 DEFINE_BASIC_FETCH_FUNCS(symbol)
 267 DEFINE_FETCH_symbol(string)
 268 DEFINE_FETCH_symbol(string_size)
 269
 270 /* kprobes don't support file_offset fetch methods */
 271 #define fetch_file_offset_u8            NULL
 272 #define fetch_file_offset_u16           NULL
 273 #define fetch_file_offset_u32           NULL
 274 #define fetch_file_offset_u64           NULL
 275 #define fetch_file_offset_string        NULL
 276 #define fetch_file_offset_string_size   NULL
 277
 278 /* Fetch type information table */
 279 static const struct fetch_type kprobes_fetch_type_table[] = {
 280         /* Special types */
 281         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 282                                         sizeof(u32), 1, "__data_loc char[]"),
 283         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 284                                         string_size, sizeof(u32), 0, "u32"),
 285         /* Basic types */
 286         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 287         ASSIGN_FETCH_TYPE(u16, u16, 0),
 288         ASSIGN_FETCH_TYPE(u32, u32, 0),
 289         ASSIGN_FETCH_TYPE(u64, u64, 0),
 290         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 291         ASSIGN_FETCH_TYPE(s16, u16, 1),
 292         ASSIGN_FETCH_TYPE(s32, u32, 1),
 293         ASSIGN_FETCH_TYPE(s64, u64, 1),
 294         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 295         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 296         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 297         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 298
 299         ASSIGN_FETCH_TYPE_END
 300 };
 301
 302 /*
 303  * Allocate new trace_probe and initialize it (including kprobes).
 304  */
 305 static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 306                                              const char *event,
 307                                              void *addr,
 308                                              const char *symbol,
 309                                              unsigned long offs,
 310                                              int maxactive,
 311                                              int nargs, bool is_return)
 312 {
 313         struct trace_kprobe *tk;
 314         int ret = -ENOMEM;
 315
 316         tk = kzalloc(SIZEOF_TRACE_KPROBE(nargs), GFP_KERNEL);
 317         if (!tk)
 318                 return ERR_PTR(ret);
 319
 320         tk->nhit = alloc_percpu(unsigned long);
 321         if (!tk->nhit)
 322                 goto error;
 323
 324         if (symbol) {
 325                 tk->symbol = kstrdup(symbol, GFP_KERNEL);
 326                 if (!tk->symbol)
 327                         goto error;
 328                 tk->rp.kp.symbol_name = tk->symbol;
 329                 tk->rp.kp.offset = offs;
 330         } else
 331                 tk->rp.kp.addr = addr;
 332
 333         if (is_return)
 334                 tk->rp.handler = kretprobe_dispatcher;
 335         else
 336                 tk->rp.kp.pre_handler = kprobe_dispatcher;
 337
 338         tk->rp.maxactive = maxactive;
 339
 340         if (!event || !is_good_name(event)) {
 341                 ret = -EINVAL;
 342                 goto error;
 343         }
 344
 345         tk->tp.call.class = &tk->tp.class;
 346         tk->tp.call.name = kstrdup(event, GFP_KERNEL);
 347         if (!tk->tp.call.name)
 348                 goto error;
 349
 350         if (!group || !is_good_name(group)) {
 351                 ret = -EINVAL;
 352                 goto error;
 353         }
 354
 355         tk->tp.class.system = kstrdup(group, GFP_KERNEL);
 356         if (!tk->tp.class.system)
 357                 goto error;
 358
 359         INIT_LIST_HEAD(&tk->list);
 360         INIT_LIST_HEAD(&tk->tp.files);
 361         return tk;
 362 error:
 363         kfree(tk->tp.call.name);
 364         kfree(tk->symbol);
 365         free_percpu(tk->nhit);
 366         kfree(tk);
 367         return ERR_PTR(ret);
 368 }
 369
 370 static void free_trace_kprobe(struct trace_kprobe *tk)
 371 {
 372         int i;
 373
 374         for (i = 0; i < tk->tp.nr_args; i++)
 375                 traceprobe_free_probe_arg(&tk->tp.args[i]);
 376
 377         kfree(tk->tp.call.class->system);
 378         kfree(tk->tp.call.name);
 379         kfree(tk->symbol);
 380         free_percpu(tk->nhit);
 381         kfree(tk);
 382 }
 383
 384 static struct trace_kprobe *find_trace_kprobe(const char *event,
 385                                               const char *group)
 386 {
 387         struct trace_kprobe *tk;
 388
 389         list_for_each_entry(tk, &probe_list, list)
 390                 if (strcmp(trace_event_name(&tk->tp.call), event) == 0 &&
 391                     strcmp(tk->tp.call.class->system, group) == 0)
 392                         return tk;
 393         return NULL;
 394 }
 395
 396 /*
 397  * Enable trace_probe
 398  * if the file is NULL, enable "perf" handler, or enable "trace" handler.
 399  */
 400 static int
 401 enable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 402 {
 403         struct event_file_link *link = NULL;
 404         int ret = 0;
 405
 406         if (file) {
 407                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 408                 if (!link) {
 409                         ret = -ENOMEM;
 410                         goto out;
 411                 }
 412
 413                 link->file = file;
 414                 list_add_tail_rcu(&link->list, &tk->tp.files);
 415
 416                 tk->tp.flags |= TP_FLAG_TRACE;
 417         } else
 418                 tk->tp.flags |= TP_FLAG_PROFILE;
 419
 420         if (trace_probe_is_registered(&tk->tp) && !trace_kprobe_has_gone(tk)) {
 421                 if (trace_kprobe_is_return(tk))
 422                         ret = enable_kretprobe(&tk->rp);
 423                 else
 424                         ret = enable_kprobe(&tk->rp.kp);
 425         }
 426
 427         if (ret) {
 428                 if (file) {
 429                         /* Notice the if is true on not WARN() */
 430                         if (!WARN_ON_ONCE(!link))
 431                                 list_del_rcu(&link->list);
 432                         kfree(link);
 433                         tk->tp.flags &= ~TP_FLAG_TRACE;
 434                 } else {
 435                         tk->tp.flags &= ~TP_FLAG_PROFILE;
 436                 }
 437         }
 438  out:
 439         return ret;
 440 }
 441
 442 /*
 443  * Disable trace_probe
 444  * if the file is NULL, disable "perf" handler, or disable "trace" handler.
 445  */
 446 static int
 447 disable_trace_kprobe(struct trace_kprobe *tk, struct trace_event_file *file)
 448 {
 449         struct event_file_link *link = NULL;
 450         int wait = 0;
 451         int ret = 0;
 452
 453         if (file) {
 454                 link = find_event_file_link(&tk->tp, file);
 455                 if (!link) {
 456                         ret = -EINVAL;
 457                         goto out;
 458                 }
 459
 460                 list_del_rcu(&link->list);
 461                 wait = 1;
 462                 if (!list_empty(&tk->tp.files))
 463                         goto out;
 464
 465                 tk->tp.flags &= ~TP_FLAG_TRACE;
 466         } else
 467                 tk->tp.flags &= ~TP_FLAG_PROFILE;
 468
 469         if (!trace_probe_is_enabled(&tk->tp) && trace_probe_is_registered(&tk->tp)) {
 470                 if (trace_kprobe_is_return(tk))
 471                         disable_kretprobe(&tk->rp);
 472                 else
 473                         disable_kprobe(&tk->rp.kp);
 474                 wait = 1;
 475         }
 476
 477         /*
 478          * if tk is not added to any list, it must be a local trace_kprobe
 479          * created with perf_event_open. We don't need to wait for these
 480          * trace_kprobes
 481          */
 482         if (list_empty(&tk->list))
 483                 wait = 0;
 484  out:
 485         if (wait) {
 486                 /*
 487                  * Synchronize with kprobe_trace_func/kretprobe_trace_func
 488                  * to ensure disabled (all running handlers are finished).
 489                  * This is not only for kfree(), but also the caller,
 490                  * trace_remove_event_call() supposes it for releasing
 491                  * event_call related objects, which will be accessed in
 492                  * the kprobe_trace_func/kretprobe_trace_func.
 493                  */
 494                 synchronize_sched();
 495                 kfree(link);    /* Ignored if link == NULL */
 496         }
 497
 498         return ret;
 499 }
 500
 501 /* Internal register function - just handle k*probes and flags */
 502 static int __register_trace_kprobe(struct trace_kprobe *tk)
 503 {
 504         int i, ret;
 505
 506         if (trace_probe_is_registered(&tk->tp))
 507                 return -EINVAL;
 508
 509         for (i = 0; i < tk->tp.nr_args; i++)
 510                 traceprobe_update_arg(&tk->tp.args[i]);
 511
 512         /* Set/clear disabled flag according to tp->flag */
 513         if (trace_probe_is_enabled(&tk->tp))
 514                 tk->rp.kp.flags &= ~KPROBE_FLAG_DISABLED;
 515         else
 516                 tk->rp.kp.flags |= KPROBE_FLAG_DISABLED;
 517
 518         if (trace_kprobe_is_return(tk))
 519                 ret = register_kretprobe(&tk->rp);
 520         else
 521                 ret = register_kprobe(&tk->rp.kp);
 522
 523         if (ret == 0)
 524                 tk->tp.flags |= TP_FLAG_REGISTERED;
 525         else {
 526                 if (ret == -ENOENT && trace_kprobe_is_on_module(tk)) {
 527                         pr_warn("This probe might be able to register after target module is loaded. Continue.\n");
 528                         ret = 0;
 529                 } else if (ret == -EILSEQ) {
 530                         pr_warn("Probing address(0x%p) is not an instruction boundary.\n",
 531                                 tk->rp.kp.addr);
 532                         ret = -EINVAL;
 533                 }
 534         }
 535
 536         return ret;
 537 }
 538
 539 /* Internal unregister function - just handle k*probes and flags */
 540 static void __unregister_trace_kprobe(struct trace_kprobe *tk)
 541 {
 542         if (trace_probe_is_registered(&tk->tp)) {
 543                 if (trace_kprobe_is_return(tk))
 544                         unregister_kretprobe(&tk->rp);
 545                 else
 546                         unregister_kprobe(&tk->rp.kp);
 547                 tk->tp.flags &= ~TP_FLAG_REGISTERED;
 548                 /* Cleanup kprobe for reuse */
 549                 if (tk->rp.kp.symbol_name)
 550                         tk->rp.kp.addr = NULL;
 551         }
 552 }
 553
 554 /* Unregister a trace_probe and probe_event: call with locking probe_lock */
 555 static int unregister_trace_kprobe(struct trace_kprobe *tk)
 556 {
 557         /* Enabled event can not be unregistered */
 558         if (trace_probe_is_enabled(&tk->tp))
 559                 return -EBUSY;
 560
 561         /* Will fail if probe is being used by ftrace or perf */
 562         if (unregister_kprobe_event(tk))
 563                 return -EBUSY;
 564
 565         __unregister_trace_kprobe(tk);
 566         list_del(&tk->list);
 567
 568         return 0;
 569 }
 570
 571 /* Register a trace_probe and probe_event */
 572 static int register_trace_kprobe(struct trace_kprobe *tk)
 573 {
 574         struct trace_kprobe *old_tk;
 575         int ret;
 576
 577         mutex_lock(&probe_lock);
 578
 579         /* Delete old (same name) event if exist */
 580         old_tk = find_trace_kprobe(trace_event_name(&tk->tp.call),
 581                         tk->tp.call.class->system);
 582         if (old_tk) {
 583                 ret = unregister_trace_kprobe(old_tk);
 584                 if (ret < 0)
 585                         goto end;
 586                 free_trace_kprobe(old_tk);
 587         }
 588
 589         /* Register new event */
 590         ret = register_kprobe_event(tk);
 591         if (ret) {
 592                 pr_warn("Failed to register probe event(%d)\n", ret);
 593                 goto end;
 594         }
 595
 596         /* Register k*probe */
 597         ret = __register_trace_kprobe(tk);
 598         if (ret < 0)
 599                 unregister_kprobe_event(tk);
 600         else
 601                 list_add_tail(&tk->list, &probe_list);
 602
 603 end:
 604         mutex_unlock(&probe_lock);
 605         return ret;
 606 }
 607
 608 /* Module notifier call back, checking event on the module */
 609 static int trace_kprobe_module_callback(struct notifier_block *nb,
 610                                        unsigned long val, void *data)
 611 {
 612         struct module *mod = data;
 613         struct trace_kprobe *tk;
 614         int ret;
 615
 616         if (val != MODULE_STATE_COMING)
 617                 return NOTIFY_DONE;
 618
 619         /* Update probes on coming module */
 620         mutex_lock(&probe_lock);
 621         list_for_each_entry(tk, &probe_list, list) {
 622                 if (trace_kprobe_within_module(tk, mod)) {
 623                         /* Don't need to check busy - this should have gone. */
 624                         __unregister_trace_kprobe(tk);
 625                         ret = __register_trace_kprobe(tk);
 626                         if (ret)
 627                                 pr_warn("Failed to re-register probe %s on %s: %d\n",
 628                                         trace_event_name(&tk->tp.call),
 629                                         mod->name, ret);
 630                 }
 631         }
 632         mutex_unlock(&probe_lock);
 633
 634         return NOTIFY_DONE;
 635 }
 636
 637 static struct notifier_block trace_kprobe_module_nb = {
 638         .notifier_call = trace_kprobe_module_callback,
 639         .priority = 1   /* Invoked after kprobe module callback */
 640 };
 641
 642 /* Convert certain expected symbols into '_' when generating event names */
 643 static inline void sanitize_event_name(char *name)
 644 {
 645         while (*name++ != '\0')
 646                 if (*name == ':' || *name == '.')
 647                         *name = '_';
 648 }
 649
 650 static int create_trace_kprobe(int argc, char **argv)
 651 {
 652         /*
 653          * Argument syntax:
 654          *  - Add kprobe:
 655          *      p[:[GRP/]EVENT] [MOD:]KSYM[+OFFS]|KADDR [FETCHARGS]
 656          *  - Add kretprobe:
 657          *      r[MAXACTIVE][:[GRP/]EVENT] [MOD:]KSYM[+0] [FETCHARGS]
 658          * Fetch args:
 659          *  $retval     : fetch return value
 660          *  $stack      : fetch stack address
 661          *  $stackN     : fetch Nth of stack (N:0-)
 662          *  $comm       : fetch current task comm
 663          *  @ADDR       : fetch memory at ADDR (ADDR should be in kernel)
 664          *  @SYM[+|-offs] : fetch memory at SYM +|- offs (SYM is a data symbol)
 665          *  %REG        : fetch register REG
 666          * Dereferencing memory fetch:
 667          *  +|-offs(ARG) : fetch memory at ARG +|- offs address.
 668          * Alias name of args:
 669          *  NAME=FETCHARG : set NAME as alias of FETCHARG.
 670          * Type of args:
 671          *  FETCHARG:TYPE : use TYPE instead of unsigned long.
 672          */
 673         struct trace_kprobe *tk;
 674         int i, ret = 0;
 675         bool is_return = false, is_delete = false;
 676         char *symbol = NULL, *event = NULL, *group = NULL;
 677         int maxactive = 0;
 678         char *arg;
 679         long offset = 0;
 680         void *addr = NULL;
 681         char buf[MAX_EVENT_NAME_LEN];
 682
 683         /* argc must be >= 1 */
 684         if (argv[0][0] == 'p')
 685                 is_return = false;
 686         else if (argv[0][0] == 'r')
 687                 is_return = true;
 688         else if (argv[0][0] == '-')
 689                 is_delete = true;
 690         else {
 691                 pr_info("Probe definition must be started with 'p', 'r' or"
 692                         " '-'.\n");
 693                 return -EINVAL;
 694         }
 695
 696         event = strchr(&argv[0][1], ':');
 697         if (event) {
 698                 event[0] = '\0';
 699                 event++;
 700         }
 701         if (is_return && isdigit(argv[0][1])) {
 702                 ret = kstrtouint(&argv[0][1], 0, &maxactive);
 703                 if (ret) {
 704                         pr_info("Failed to parse maxactive.\n");
 705                         return ret;
 706                 }
 707                 /* kretprobes instances are iterated over via a list. The
 708                  * maximum should stay reasonable.
 709                  */
 710                 if (maxactive > KRETPROBE_MAXACTIVE_MAX) {
 711                         pr_info("Maxactive is too big (%d > %d).\n",
 712                                 maxactive, KRETPROBE_MAXACTIVE_MAX);
 713                         return -E2BIG;
 714                 }
 715         }
 716
 717         if (event) {
 718                 if (strchr(event, '/')) {
 719                         group = event;
 720                         event = strchr(group, '/') + 1;
 721                         event[-1] = '\0';
 722                         if (strlen(group) == 0) {
 723                                 pr_info("Group name is not specified\n");
 724                                 return -EINVAL;
 725                         }
 726                 }
 727                 if (strlen(event) == 0) {
 728                         pr_info("Event name is not specified\n");
 729                         return -EINVAL;
 730                 }
 731         }
 732         if (!group)
 733                 group = KPROBE_EVENT_SYSTEM;
 734
 735         if (is_delete) {
 736                 if (!event) {
 737                         pr_info("Delete command needs an event name.\n");
 738                         return -EINVAL;
 739                 }
 740                 mutex_lock(&probe_lock);
 741                 tk = find_trace_kprobe(event, group);
 742                 if (!tk) {
 743                         mutex_unlock(&probe_lock);
 744                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 745                         return -ENOENT;
 746                 }
 747                 /* delete an event */
 748                 ret = unregister_trace_kprobe(tk);
 749                 if (ret == 0)
 750                         free_trace_kprobe(tk);
 751                 mutex_unlock(&probe_lock);
 752                 return ret;
 753         }
 754
 755         if (argc < 2) {
 756                 pr_info("Probe point is not specified.\n");
 757                 return -EINVAL;
 758         }
 759
 760         /* try to parse an address. if that fails, try to read the
 761          * input as a symbol. */
 762         if (kstrtoul(argv[1], 0, (unsigned long *)&addr)) {
 763                 /* a symbol specified */
 764                 symbol = argv[1];
 765                 /* TODO: support .init module functions */
 766                 ret = traceprobe_split_symbol_offset(symbol, &offset);
 767                 if (ret || offset < 0 || offset > UINT_MAX) {
 768                         pr_info("Failed to parse either an address or a symbol.\n");
 769                         return ret;
 770                 }
 771                 if (offset && is_return &&
 772                     !kprobe_on_func_entry(NULL, symbol, offset)) {
 773                         pr_info("Given offset is not valid for return probe.\n");
 774                         return -EINVAL;
 775                 }
 776         }
 777         argc -= 2; argv += 2;
 778
 779         /* setup a probe */
 780         if (!event) {
 781                 /* Make a new event name */
 782                 if (symbol)
 783                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_%ld",
 784                                  is_return ? 'r' : 'p', symbol, offset);
 785                 else
 786                         snprintf(buf, MAX_EVENT_NAME_LEN, "%c_0x%p",
 787                                  is_return ? 'r' : 'p', addr);
 788                 sanitize_event_name(buf);
 789                 event = buf;
 790         }
 791         tk = alloc_trace_kprobe(group, event, addr, symbol, offset, maxactive,
 792                                argc, is_return);
 793         if (IS_ERR(tk)) {
 794                 pr_info("Failed to allocate trace_probe.(%d)\n",
 795                         (int)PTR_ERR(tk));
 796                 return PTR_ERR(tk);
 797         }
 798
 799         /* parse arguments */
 800         ret = 0;
 801         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 802                 struct probe_arg *parg = &tk->tp.args[i];
 803
 804                 /* Increment count for freeing args in error case */
 805                 tk->tp.nr_args++;
 806
 807                 /* Parse argument name */
 808                 arg = strchr(argv[i], '=');
 809                 if (arg) {
 810                         *arg++ = '\0';
 811                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 812                 } else {
 813                         arg = argv[i];
 814                         /* If argument name is omitted, set "argN" */
 815                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 816                         parg->name = kstrdup(buf, GFP_KERNEL);
 817                 }
 818
 819                 if (!parg->name) {
 820                         pr_info("Failed to allocate argument[%d] name.\n", i);
 821                         ret = -ENOMEM;
 822                         goto error;
 823                 }
 824
 825                 if (!is_good_name(parg->name)) {
 826                         pr_info("Invalid argument[%d] name: %s\n",
 827                                 i, parg->name);
 828                         ret = -EINVAL;
 829                         goto error;
 830                 }
 831
 832                 if (traceprobe_conflict_field_name(parg->name,
 833                                                         tk->tp.args, i)) {
 834                         pr_info("Argument[%d] name '%s' conflicts with "
 835                                 "another field.\n", i, argv[i]);
 836                         ret = -EINVAL;
 837                         goto error;
 838                 }
 839
 840                 /* Parse fetch argument */
 841                 ret = traceprobe_parse_probe_arg(arg, &tk->tp.size, parg,
 842                                                 is_return, true,
 843                                                 kprobes_fetch_type_table);
 844                 if (ret) {
 845                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 846                         goto error;
 847                 }
 848         }
 849
 850         ret = register_trace_kprobe(tk);
 851         if (ret)
 852                 goto error;
 853         return 0;
 854
 855 error:
 856         free_trace_kprobe(tk);
 857         return ret;
 858 }
 859
 860 static int release_all_trace_kprobes(void)
 861 {
 862         struct trace_kprobe *tk;
 863         int ret = 0;
 864
 865         mutex_lock(&probe_lock);
 866         /* Ensure no probe is in use. */
 867         list_for_each_entry(tk, &probe_list, list)
 868                 if (trace_probe_is_enabled(&tk->tp)) {
 869                         ret = -EBUSY;
 870                         goto end;
 871                 }
 872         /* TODO: Use batch unregistration */
 873         while (!list_empty(&probe_list)) {
 874                 tk = list_entry(probe_list.next, struct trace_kprobe, list);
 875                 ret = unregister_trace_kprobe(tk);
 876                 if (ret)
 877                         goto end;
 878                 free_trace_kprobe(tk);
 879         }
 880
 881 end:
 882         mutex_unlock(&probe_lock);
 883
 884         return ret;
 885 }
 886
 887 /* Probes listing interfaces */
 888 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 889 {
 890         mutex_lock(&probe_lock);
 891         return seq_list_start(&probe_list, *pos);
 892 }
 893
 894 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 895 {
 896         return seq_list_next(v, &probe_list, pos);
 897 }
 898
 899 static void probes_seq_stop(struct seq_file *m, void *v)
 900 {
 901         mutex_unlock(&probe_lock);
 902 }
 903
 904 static int probes_seq_show(struct seq_file *m, void *v)
 905 {
 906         struct trace_kprobe *tk = v;
 907         int i;
 908
 909         seq_putc(m, trace_kprobe_is_return(tk) ? 'r' : 'p');
 910         seq_printf(m, ":%s/%s", tk->tp.call.class->system,
 911                         trace_event_name(&tk->tp.call));
 912
 913         if (!tk->symbol)
 914                 seq_printf(m, " 0x%p", tk->rp.kp.addr);
 915         else if (tk->rp.kp.offset)
 916                 seq_printf(m, " %s+%u", trace_kprobe_symbol(tk),
 917                            tk->rp.kp.offset);
 918         else
 919                 seq_printf(m, " %s", trace_kprobe_symbol(tk));
 920
 921         for (i = 0; i < tk->tp.nr_args; i++)
 922                 seq_printf(m, " %s=%s", tk->tp.args[i].name, tk->tp.args[i].comm);
 923         seq_putc(m, '\n');
 924
 925         return 0;
 926 }
 927
 928 static const struct seq_operations probes_seq_op = {
 929         .start  = probes_seq_start,
 930         .next   = probes_seq_next,
 931         .stop   = probes_seq_stop,
 932         .show   = probes_seq_show
 933 };
 934
 935 static int probes_open(struct inode *inode, struct file *file)
 936 {
 937         int ret;
 938
 939         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 940                 ret = release_all_trace_kprobes();
 941                 if (ret < 0)
 942                         return ret;
 943         }
 944
 945         return seq_open(file, &probes_seq_op);
 946 }
 947
 948 static ssize_t probes_write(struct file *file, const char __user *buffer,
 949                             size_t count, loff_t *ppos)
 950 {
 951         return trace_parse_run_command(file, buffer, count, ppos,
 952                                        create_trace_kprobe);
 953 }
 954
 955 static const struct file_operations kprobe_events_ops = {
 956         .owner          = THIS_MODULE,
 957         .open           = probes_open,
 958         .read           = seq_read,
 959         .llseek         = seq_lseek,
 960         .release        = seq_release,
 961         .write          = probes_write,
 962 };
 963
 964 /* Probes profiling interfaces */
 965 static int probes_profile_seq_show(struct seq_file *m, void *v)
 966 {
 967         struct trace_kprobe *tk = v;
 968
 969         seq_printf(m, "  %-44s %15lu %15lu\n",
 970                    trace_event_name(&tk->tp.call),
 971                    trace_kprobe_nhit(tk),
 972                    tk->rp.kp.nmissed);
 973
 974         return 0;
 975 }
 976
 977 static const struct seq_operations profile_seq_op = {
 978         .start  = probes_seq_start,
 979         .next   = probes_seq_next,
 980         .stop   = probes_seq_stop,
 981         .show   = probes_profile_seq_show
 982 };
 983
 984 static int profile_open(struct inode *inode, struct file *file)
 985 {
 986         return seq_open(file, &profile_seq_op);
 987 }
 988
 989 static const struct file_operations kprobe_profile_ops = {
 990         .owner          = THIS_MODULE,
 991         .open           = profile_open,
 992         .read           = seq_read,
 993         .llseek         = seq_lseek,
 994         .release        = seq_release,
 995 };
 996
 997 /* Kprobe handler */
 998 static nokprobe_inline void
 999 __kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs,
1000                     struct trace_event_file *trace_file)
1001 {
1002         struct kprobe_trace_entry_head *entry;
1003         struct ring_buffer_event *event;
1004         struct ring_buffer *buffer;
1005         int size, dsize, pc;
1006         unsigned long irq_flags;
1007         struct trace_event_call *call = &tk->tp.call;
1008
1009         WARN_ON(call != trace_file->event_call);
1010
1011         if (trace_trigger_soft_disabled(trace_file))
1012                 return;
1013
1014         local_save_flags(irq_flags);
1015         pc = preempt_count();
1016
1017         dsize = __get_data_size(&tk->tp, regs);
1018         size = sizeof(*entry) + tk->tp.size + dsize;
1019
1020         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1021                                                 call->event.type,
1022                                                 size, irq_flags, pc);
1023         if (!event)
1024                 return;
1025
1026         entry = ring_buffer_event_data(event);
1027         entry->ip = (unsigned long)tk->rp.kp.addr;
1028         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1029
1030         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1031                                          entry, irq_flags, pc, regs);
1032 }
1033
1034 static void
1035 kprobe_trace_func(struct trace_kprobe *tk, struct pt_regs *regs)
1036 {
1037         struct event_file_link *link;
1038
1039         list_for_each_entry_rcu(link, &tk->tp.files, list)
1040                 __kprobe_trace_func(tk, regs, link->file);
1041 }
1042 NOKPROBE_SYMBOL(kprobe_trace_func);
1043
1044 /* Kretprobe handler */
1045 static nokprobe_inline void
1046 __kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1047                        struct pt_regs *regs,
1048                        struct trace_event_file *trace_file)
1049 {
1050         struct kretprobe_trace_entry_head *entry;
1051         struct ring_buffer_event *event;
1052         struct ring_buffer *buffer;
1053         int size, pc, dsize;
1054         unsigned long irq_flags;
1055         struct trace_event_call *call = &tk->tp.call;
1056
1057         WARN_ON(call != trace_file->event_call);
1058
1059         if (trace_trigger_soft_disabled(trace_file))
1060                 return;
1061
1062         local_save_flags(irq_flags);
1063         pc = preempt_count();
1064
1065         dsize = __get_data_size(&tk->tp, regs);
1066         size = sizeof(*entry) + tk->tp.size + dsize;
1067
1068         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
1069                                                 call->event.type,
1070                                                 size, irq_flags, pc);
1071         if (!event)
1072                 return;
1073
1074         entry = ring_buffer_event_data(event);
1075         entry->func = (unsigned long)tk->rp.kp.addr;
1076         entry->ret_ip = (unsigned long)ri->ret_addr;
1077         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1078
1079         event_trigger_unlock_commit_regs(trace_file, buffer, event,
1080                                          entry, irq_flags, pc, regs);
1081 }
1082
1083 static void
1084 kretprobe_trace_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1085                      struct pt_regs *regs)
1086 {
1087         struct event_file_link *link;
1088
1089         list_for_each_entry_rcu(link, &tk->tp.files, list)
1090                 __kretprobe_trace_func(tk, ri, regs, link->file);
1091 }
1092 NOKPROBE_SYMBOL(kretprobe_trace_func);
1093
1094 /* Event entry printers */
1095 static enum print_line_t
1096 print_kprobe_event(struct trace_iterator *iter, int flags,
1097                    struct trace_event *event)
1098 {
1099         struct kprobe_trace_entry_head *field;
1100         struct trace_seq *s = &iter->seq;
1101         struct trace_probe *tp;
1102         u8 *data;
1103         int i;
1104
1105         field = (struct kprobe_trace_entry_head *)iter->ent;
1106         tp = container_of(event, struct trace_probe, call.event);
1107
1108         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1109
1110         if (!seq_print_ip_sym(s, field->ip, flags | TRACE_ITER_SYM_OFFSET))
1111                 goto out;
1112
1113         trace_seq_putc(s, ')');
1114
1115         data = (u8 *)&field[1];
1116         for (i = 0; i < tp->nr_args; i++)
1117                 if (!tp->args[i].type->print(s, tp->args[i].name,
1118                                              data + tp->args[i].offset, field))
1119                         goto out;
1120
1121         trace_seq_putc(s, '\n');
1122  out:
1123         return trace_handle_return(s);
1124 }
1125
1126 static enum print_line_t
1127 print_kretprobe_event(struct trace_iterator *iter, int flags,
1128                       struct trace_event *event)
1129 {
1130         struct kretprobe_trace_entry_head *field;
1131         struct trace_seq *s = &iter->seq;
1132         struct trace_probe *tp;
1133         u8 *data;
1134         int i;
1135
1136         field = (struct kretprobe_trace_entry_head *)iter->ent;
1137         tp = container_of(event, struct trace_probe, call.event);
1138
1139         trace_seq_printf(s, "%s: (", trace_event_name(&tp->call));
1140
1141         if (!seq_print_ip_sym(s, field->ret_ip, flags | TRACE_ITER_SYM_OFFSET))
1142                 goto out;
1143
1144         trace_seq_puts(s, " <- ");
1145
1146         if (!seq_print_ip_sym(s, field->func, flags & ~TRACE_ITER_SYM_OFFSET))
1147                 goto out;
1148
1149         trace_seq_putc(s, ')');
1150
1151         data = (u8 *)&field[1];
1152         for (i = 0; i < tp->nr_args; i++)
1153                 if (!tp->args[i].type->print(s, tp->args[i].name,
1154                                              data + tp->args[i].offset, field))
1155                         goto out;
1156
1157         trace_seq_putc(s, '\n');
1158
1159  out:
1160         return trace_handle_return(s);
1161 }
1162
1163
1164 static int kprobe_event_define_fields(struct trace_event_call *event_call)
1165 {
1166         int ret, i;
1167         struct kprobe_trace_entry_head field;
1168         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1169
1170         DEFINE_FIELD(unsigned long, ip, FIELD_STRING_IP, 0);
1171         /* Set argument names as fields */
1172         for (i = 0; i < tk->tp.nr_args; i++) {
1173                 struct probe_arg *parg = &tk->tp.args[i];
1174
1175                 ret = trace_define_field(event_call, parg->type->fmttype,
1176                                          parg->name,
1177                                          sizeof(field) + parg->offset,
1178                                          parg->type->size,
1179                                          parg->type->is_signed,
1180                                          FILTER_OTHER);
1181                 if (ret)
1182                         return ret;
1183         }
1184         return 0;
1185 }
1186
1187 static int kretprobe_event_define_fields(struct trace_event_call *event_call)
1188 {
1189         int ret, i;
1190         struct kretprobe_trace_entry_head field;
1191         struct trace_kprobe *tk = (struct trace_kprobe *)event_call->data;
1192
1193         DEFINE_FIELD(unsigned long, func, FIELD_STRING_FUNC, 0);
1194         DEFINE_FIELD(unsigned long, ret_ip, FIELD_STRING_RETIP, 0);
1195         /* Set argument names as fields */
1196         for (i = 0; i < tk->tp.nr_args; i++) {
1197                 struct probe_arg *parg = &tk->tp.args[i];
1198
1199                 ret = trace_define_field(event_call, parg->type->fmttype,
1200                                          parg->name,
1201                                          sizeof(field) + parg->offset,
1202                                          parg->type->size,
1203                                          parg->type->is_signed,
1204                                          FILTER_OTHER);
1205                 if (ret)
1206                         return ret;
1207         }
1208         return 0;
1209 }
1210
1211 #ifdef CONFIG_PERF_EVENTS
1212
1213 /* Kprobe profile handler */
1214 static int
1215 kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
1216 {
1217         struct trace_event_call *call = &tk->tp.call;
1218         struct kprobe_trace_entry_head *entry;
1219         struct hlist_head *head;
1220         int size, __size, dsize;
1221         int rctx;
1222
1223         if (bpf_prog_array_valid(call)) {
1224                 unsigned long orig_ip = instruction_pointer(regs);
1225                 int ret;
1226
1227                 ret = trace_call_bpf(call, regs);
1228
1229                 /*
1230                  * We need to check and see if we modified the pc of the
1231                  * pt_regs, and if so return 1 so that we don't do the
1232                  * single stepping.
1233                  */
1234                 if (orig_ip != instruction_pointer(regs))
1235                         return 1;
1236                 if (!ret)
1237                         return 0;
1238         }
1239
1240         head = this_cpu_ptr(call->perf_events);
1241         if (hlist_empty(head))
1242                 return 0;
1243
1244         dsize = __get_data_size(&tk->tp, regs);
1245         __size = sizeof(*entry) + tk->tp.size + dsize;
1246         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1247         size -= sizeof(u32);
1248
1249         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1250         if (!entry)
1251                 return 0;
1252
1253         entry->ip = (unsigned long)tk->rp.kp.addr;
1254         memset(&entry[1], 0, dsize);
1255         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1256         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1257                               head, NULL);
1258         return 0;
1259 }
1260 NOKPROBE_SYMBOL(kprobe_perf_func);
1261
1262 /* Kretprobe profile handler */
1263 static void
1264 kretprobe_perf_func(struct trace_kprobe *tk, struct kretprobe_instance *ri,
1265                     struct pt_regs *regs)
1266 {
1267         struct trace_event_call *call = &tk->tp.call;
1268         struct kretprobe_trace_entry_head *entry;
1269         struct hlist_head *head;
1270         int size, __size, dsize;
1271         int rctx;
1272
1273         if (bpf_prog_array_valid(call) && !trace_call_bpf(call, regs))
1274                 return;
1275
1276         head = this_cpu_ptr(call->perf_events);
1277         if (hlist_empty(head))
1278                 return;
1279
1280         dsize = __get_data_size(&tk->tp, regs);
1281         __size = sizeof(*entry) + tk->tp.size + dsize;
1282         size = ALIGN(__size + sizeof(u32), sizeof(u64));
1283         size -= sizeof(u32);
1284
1285         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1286         if (!entry)
1287                 return;
1288
1289         entry->func = (unsigned long)tk->rp.kp.addr;
1290         entry->ret_ip = (unsigned long)ri->ret_addr;
1291         store_trace_args(sizeof(*entry), &tk->tp, regs, (u8 *)&entry[1], dsize);
1292         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1293                               head, NULL);
1294 }
1295 NOKPROBE_SYMBOL(kretprobe_perf_func);
1296
1297 int bpf_get_kprobe_info(const struct perf_event *event, u32 *fd_type,
1298                         const char **symbol, u64 *probe_offset,
1299                         u64 *probe_addr, bool perf_type_tracepoint)
1300 {
1301         const char *pevent = trace_event_name(event->tp_event);
1302         const char *group = event->tp_event->class->system;
1303         struct trace_kprobe *tk;
1304
1305         if (perf_type_tracepoint)
1306                 tk = find_trace_kprobe(pevent, group);
1307         else
1308                 tk = event->tp_event->data;
1309         if (!tk)
1310                 return -EINVAL;
1311
1312         *fd_type = trace_kprobe_is_return(tk) ? BPF_FD_TYPE_KRETPROBE
1313                                               : BPF_FD_TYPE_KPROBE;
1314         if (tk->symbol) {
1315                 *symbol = tk->symbol;
1316                 *probe_offset = tk->rp.kp.offset;
1317                 *probe_addr = 0;
1318         } else {
1319                 *symbol = NULL;
1320                 *probe_offset = 0;
1321                 *probe_addr = (unsigned long)tk->rp.kp.addr;
1322         }
1323         return 0;
1324 }
1325 #endif  /* CONFIG_PERF_EVENTS */
1326
1327 /*
1328  * called by perf_trace_init() or __ftrace_set_clr_event() under event_mutex.
1329  *
1330  * kprobe_trace_self_tests_init() does enable_trace_probe/disable_trace_probe
1331  * lockless, but we can't race with this __init function.
1332  */
1333 static int kprobe_register(struct trace_event_call *event,
1334                            enum trace_reg type, void *data)
1335 {
1336         struct trace_kprobe *tk = (struct trace_kprobe *)event->data;
1337         struct trace_event_file *file = data;
1338
1339         switch (type) {
1340         case TRACE_REG_REGISTER:
1341                 return enable_trace_kprobe(tk, file);
1342         case TRACE_REG_UNREGISTER:
1343                 return disable_trace_kprobe(tk, file);
1344
1345 #ifdef CONFIG_PERF_EVENTS
1346         case TRACE_REG_PERF_REGISTER:
1347                 return enable_trace_kprobe(tk, NULL);
1348         case TRACE_REG_PERF_UNREGISTER:
1349                 return disable_trace_kprobe(tk, NULL);
1350         case TRACE_REG_PERF_OPEN:
1351         case TRACE_REG_PERF_CLOSE:
1352         case TRACE_REG_PERF_ADD:
1353         case TRACE_REG_PERF_DEL:
1354                 return 0;
1355 #endif
1356         }
1357         return 0;
1358 }
1359
1360 static int kprobe_dispatcher(struct kprobe *kp, struct pt_regs *regs)
1361 {
1362         struct trace_kprobe *tk = container_of(kp, struct trace_kprobe, rp.kp);
1363         int ret = 0;
1364
1365         raw_cpu_inc(*tk->nhit);
1366
1367         if (tk->tp.flags & TP_FLAG_TRACE)
1368                 kprobe_trace_func(tk, regs);
1369 #ifdef CONFIG_PERF_EVENTS
1370         if (tk->tp.flags & TP_FLAG_PROFILE)
1371                 ret = kprobe_perf_func(tk, regs);
1372 #endif
1373         return ret;
1374 }
1375 NOKPROBE_SYMBOL(kprobe_dispatcher);
1376
1377 static int
1378 kretprobe_dispatcher(struct kretprobe_instance *ri, struct pt_regs *regs)
1379 {
1380         struct trace_kprobe *tk = container_of(ri->rp, struct trace_kprobe, rp);
1381
1382         raw_cpu_inc(*tk->nhit);
1383
1384         if (tk->tp.flags & TP_FLAG_TRACE)
1385                 kretprobe_trace_func(tk, ri, regs);
1386 #ifdef CONFIG_PERF_EVENTS
1387         if (tk->tp.flags & TP_FLAG_PROFILE)
1388                 kretprobe_perf_func(tk, ri, regs);
1389 #endif
1390         return 0;       /* We don't tweek kernel, so just return 0 */
1391 }
1392 NOKPROBE_SYMBOL(kretprobe_dispatcher);
1393
1394 static struct trace_event_functions kretprobe_funcs = {
1395         .trace          = print_kretprobe_event
1396 };
1397
1398 static struct trace_event_functions kprobe_funcs = {
1399         .trace          = print_kprobe_event
1400 };
1401
1402 static inline void init_trace_event_call(struct trace_kprobe *tk,
1403                                          struct trace_event_call *call)
1404 {
1405         INIT_LIST_HEAD(&call->class->fields);
1406         if (trace_kprobe_is_return(tk)) {
1407                 call->event.funcs = &kretprobe_funcs;
1408                 call->class->define_fields = kretprobe_event_define_fields;
1409         } else {
1410                 call->event.funcs = &kprobe_funcs;
1411                 call->class->define_fields = kprobe_event_define_fields;
1412         }
1413
1414         call->flags = TRACE_EVENT_FL_KPROBE;
1415         call->class->reg = kprobe_register;
1416         call->data = tk;
1417 }
1418
1419 static int register_kprobe_event(struct trace_kprobe *tk)
1420 {
1421         struct trace_event_call *call = &tk->tp.call;
1422         int ret = 0;
1423
1424         init_trace_event_call(tk, call);
1425
1426         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0)
1427                 return -ENOMEM;
1428         ret = register_trace_event(&call->event);
1429         if (!ret) {
1430                 kfree(call->print_fmt);
1431                 return -ENODEV;
1432         }
1433         ret = trace_add_event_call(call);
1434         if (ret) {
1435                 pr_info("Failed to register kprobe event: %s\n",
1436                         trace_event_name(call));
1437                 kfree(call->print_fmt);
1438                 unregister_trace_event(&call->event);
1439         }
1440         return ret;
1441 }
1442
1443 static int unregister_kprobe_event(struct trace_kprobe *tk)
1444 {
1445         int ret;
1446
1447         /* tp->event is unregistered in trace_remove_event_call() */
1448         ret = trace_remove_event_call(&tk->tp.call);
1449         if (!ret)
1450                 kfree(tk->tp.call.print_fmt);
1451         return ret;
1452 }
1453
1454 #ifdef CONFIG_PERF_EVENTS
1455 /* create a trace_kprobe, but don't add it to global lists */
1456 struct trace_event_call *
1457 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
1458                           bool is_return)
1459 {
1460         struct trace_kprobe *tk;
1461         int ret;
1462         char *event;
1463
1464         /*
1465          * local trace_kprobes are not added to probe_list, so they are never
1466          * searched in find_trace_kprobe(). Therefore, there is no concern of
1467          * duplicated name here.
1468          */
1469         event = func ? func : "DUMMY_EVENT";
1470
1471         tk = alloc_trace_kprobe(KPROBE_EVENT_SYSTEM, event, (void *)addr, func,
1472                                 offs, 0 /* maxactive */, 0 /* nargs */,
1473                                 is_return);
1474
1475         if (IS_ERR(tk)) {
1476                 pr_info("Failed to allocate trace_probe.(%d)\n",
1477                         (int)PTR_ERR(tk));
1478                 return ERR_CAST(tk);
1479         }
1480
1481         init_trace_event_call(tk, &tk->tp.call);
1482
1483         if (set_print_fmt(&tk->tp, trace_kprobe_is_return(tk)) < 0) {
1484                 ret = -ENOMEM;
1485                 goto error;
1486         }
1487
1488         ret = __register_trace_kprobe(tk);
1489         if (ret < 0) {
1490                 kfree(tk->tp.call.print_fmt);
1491                 goto error;
1492         }
1493
1494         return &tk->tp.call;
1495 error:
1496         free_trace_kprobe(tk);
1497         return ERR_PTR(ret);
1498 }
1499
1500 void destroy_local_trace_kprobe(struct trace_event_call *event_call)
1501 {
1502         struct trace_kprobe *tk;
1503
1504         tk = container_of(event_call, struct trace_kprobe, tp.call);
1505
1506         if (trace_probe_is_enabled(&tk->tp)) {
1507                 WARN_ON(1);
1508                 return;
1509         }
1510
1511         __unregister_trace_kprobe(tk);
1512
1513         kfree(tk->tp.call.print_fmt);
1514         free_trace_kprobe(tk);
1515 }
1516 #endif /* CONFIG_PERF_EVENTS */
1517
1518 /* Make a tracefs interface for controlling probe points */
1519 static __init int init_kprobe_trace(void)
1520 {
1521         struct dentry *d_tracer;
1522         struct dentry *entry;
1523
1524         if (register_module_notifier(&trace_kprobe_module_nb))
1525                 return -EINVAL;
1526
1527         d_tracer = tracing_init_dentry();
1528         if (IS_ERR(d_tracer))
1529                 return 0;
1530
1531         entry = tracefs_create_file("kprobe_events", 0644, d_tracer,
1532                                     NULL, &kprobe_events_ops);
1533
1534         /* Event list interface */
1535         if (!entry)
1536                 pr_warn("Could not create tracefs 'kprobe_events' entry\n");
1537
1538         /* Profile interface */
1539         entry = tracefs_create_file("kprobe_profile", 0444, d_tracer,
1540                                     NULL, &kprobe_profile_ops);
1541
1542         if (!entry)
1543                 pr_warn("Could not create tracefs 'kprobe_profile' entry\n");
1544         return 0;
1545 }
1546 fs_initcall(init_kprobe_trace);
1547
1548
1549 #ifdef CONFIG_FTRACE_STARTUP_TEST
1550 /*
1551  * The "__used" keeps gcc from removing the function symbol
1552  * from the kallsyms table. 'noinline' makes sure that there
1553  * isn't an inlined version used by the test method below
1554  */
1555 static __used __init noinline int
1556 kprobe_trace_selftest_target(int a1, int a2, int a3, int a4, int a5, int a6)
1557 {
1558         return a1 + a2 + a3 + a4 + a5 + a6;
1559 }
1560
1561 static __init struct trace_event_file *
1562 find_trace_probe_file(struct trace_kprobe *tk, struct trace_array *tr)
1563 {
1564         struct trace_event_file *file;
1565
1566         list_for_each_entry(file, &tr->events, list)
1567                 if (file->event_call == &tk->tp.call)
1568                         return file;
1569
1570         return NULL;
1571 }
1572
1573 /*
1574  * Nobody but us can call enable_trace_kprobe/disable_trace_kprobe at this
1575  * stage, we can do this lockless.
1576  */
1577 static __init int kprobe_trace_self_tests_init(void)
1578 {
1579         int ret, warn = 0;
1580         int (*target)(int, int, int, int, int, int);
1581         struct trace_kprobe *tk;
1582         struct trace_event_file *file;
1583
1584         if (tracing_is_disabled())
1585                 return -ENODEV;
1586
1587         target = kprobe_trace_selftest_target;
1588
1589         pr_info("Testing kprobe tracing: ");
1590
1591         ret = trace_run_command("p:testprobe kprobe_trace_selftest_target "
1592                                 "$stack $stack0 +0($stack)",
1593                                 create_trace_kprobe);
1594         if (WARN_ON_ONCE(ret)) {
1595                 pr_warn("error on probing function entry.\n");
1596                 warn++;
1597         } else {
1598                 /* Enable trace point */
1599                 tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1600                 if (WARN_ON_ONCE(tk == NULL)) {
1601                         pr_warn("error on getting new probe.\n");
1602                         warn++;
1603                 } else {
1604                         file = find_trace_probe_file(tk, top_trace_array());
1605                         if (WARN_ON_ONCE(file == NULL)) {
1606                                 pr_warn("error on getting probe file.\n");
1607                                 warn++;
1608                         } else
1609                                 enable_trace_kprobe(tk, file);
1610                 }
1611         }
1612
1613         ret = trace_run_command("r:testprobe2 kprobe_trace_selftest_target "
1614                                 "$retval", create_trace_kprobe);
1615         if (WARN_ON_ONCE(ret)) {
1616                 pr_warn("error on probing function return.\n");
1617                 warn++;
1618         } else {
1619                 /* Enable trace point */
1620                 tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1621                 if (WARN_ON_ONCE(tk == NULL)) {
1622                         pr_warn("error on getting 2nd new probe.\n");
1623                         warn++;
1624                 } else {
1625                         file = find_trace_probe_file(tk, top_trace_array());
1626                         if (WARN_ON_ONCE(file == NULL)) {
1627                                 pr_warn("error on getting probe file.\n");
1628                                 warn++;
1629                         } else
1630                                 enable_trace_kprobe(tk, file);
1631                 }
1632         }
1633
1634         if (warn)
1635                 goto end;
1636
1637         ret = target(1, 2, 3, 4, 5, 6);
1638
1639         /*
1640          * Not expecting an error here, the check is only to prevent the
1641          * optimizer from removing the call to target() as otherwise there
1642          * are no side-effects and the call is never performed.
1643          */
1644         if (ret != 21)
1645                 warn++;
1646
1647         /* Disable trace points before removing it */
1648         tk = find_trace_kprobe("testprobe", KPROBE_EVENT_SYSTEM);
1649         if (WARN_ON_ONCE(tk == NULL)) {
1650                 pr_warn("error on getting test probe.\n");
1651                 warn++;
1652         } else {
1653                 if (trace_kprobe_nhit(tk) != 1) {
1654                         pr_warn("incorrect number of testprobe hits\n");
1655                         warn++;
1656                 }
1657
1658                 file = find_trace_probe_file(tk, top_trace_array());
1659                 if (WARN_ON_ONCE(file == NULL)) {
1660                         pr_warn("error on getting probe file.\n");
1661                         warn++;
1662                 } else
1663                         disable_trace_kprobe(tk, file);
1664         }
1665
1666         tk = find_trace_kprobe("testprobe2", KPROBE_EVENT_SYSTEM);
1667         if (WARN_ON_ONCE(tk == NULL)) {
1668                 pr_warn("error on getting 2nd test probe.\n");
1669                 warn++;
1670         } else {
1671                 if (trace_kprobe_nhit(tk) != 1) {
1672                         pr_warn("incorrect number of testprobe2 hits\n");
1673                         warn++;
1674                 }
1675
1676                 file = find_trace_probe_file(tk, top_trace_array());
1677                 if (WARN_ON_ONCE(file == NULL)) {
1678                         pr_warn("error on getting probe file.\n");
1679                         warn++;
1680                 } else
1681                         disable_trace_kprobe(tk, file);
1682         }
1683
1684         ret = trace_run_command("-:testprobe", create_trace_kprobe);
1685         if (WARN_ON_ONCE(ret)) {
1686                 pr_warn("error on deleting a probe.\n");
1687                 warn++;
1688         }
1689
1690         ret = trace_run_command("-:testprobe2", create_trace_kprobe);
1691         if (WARN_ON_ONCE(ret)) {
1692                 pr_warn("error on deleting a probe.\n");
1693                 warn++;
1694         }
1695
1696 end:
1697         release_all_trace_kprobes();
1698         /*
1699          * Wait for the optimizer work to finish. Otherwise it might fiddle
1700          * with probes in already freed __init text.
1701          */
1702         wait_for_kprobe_optimizer();
1703         if (warn)
1704                 pr_cont("NG: Some tests are failed. Please check them.\n");
1705         else
1706                 pr_cont("OK\n");
1707         return 0;
1708 }
1709
1710 late_initcall(kprobe_trace_self_tests_init);
1711
1712 #endif