kernel/trace/trace_uprobe.c

   1 /*
   2  * uprobes-based tracing events
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  16  *
  17  * Copyright (C) IBM Corporation, 2010-2012
  18  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/module.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/uprobes.h>
  24 #include <linux/namei.h>
  25 #include <linux/string.h>
  26
  27 #include "trace_probe.h"
  28
  29 #define UPROBE_EVENT_SYSTEM     "uprobes"
  30
  31 struct uprobe_trace_entry_head {
  32         struct trace_entry      ent;
  33         unsigned long           vaddr[];
  34 };
  35
  36 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  37         (sizeof(struct uprobe_trace_entry_head) +       \
  38          sizeof(unsigned long) * (is_return ? 2 : 1))
  39
  40 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  41         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  42
  43 struct trace_uprobe_filter {
  44         rwlock_t                rwlock;
  45         int                     nr_systemwide;
  46         struct list_head        perf_events;
  47 };
  48
  49 /*
  50  * uprobe event core functions
  51  */
  52 struct trace_uprobe {
  53         struct list_head                list;
  54         struct trace_uprobe_filter      filter;
  55         struct uprobe_consumer          consumer;
  56         struct inode                    *inode;
  57         char                            *filename;
  58         unsigned long                   offset;
  59         unsigned long                   nhit;
  60         struct trace_probe              tp;
  61 };
  62
  63 #define SIZEOF_TRACE_UPROBE(n)                          \
  64         (offsetof(struct trace_uprobe, tp.args) +       \
  65         (sizeof(struct probe_arg) * (n)))
  66
  67 static int register_uprobe_event(struct trace_uprobe *tu);
  68 static int unregister_uprobe_event(struct trace_uprobe *tu);
  69
  70 static DEFINE_MUTEX(uprobe_lock);
  71 static LIST_HEAD(uprobe_list);
  72
  73 struct uprobe_dispatch_data {
  74         struct trace_uprobe     *tu;
  75         unsigned long           bp_addr;
  76 };
  77
  78 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  79 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  80                                 unsigned long func, struct pt_regs *regs);
  81
  82 #ifdef CONFIG_STACK_GROWSUP
  83 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  84 {
  85         return addr - (n * sizeof(long));
  86 }
  87 #else
  88 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  89 {
  90         return addr + (n * sizeof(long));
  91 }
  92 #endif
  93
  94 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  95 {
  96         unsigned long ret;
  97         unsigned long addr = user_stack_pointer(regs);
  98
  99         addr = adjust_stack_addr(addr, n);
 100
 101         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 102                 return 0;
 103
 104         return ret;
 105 }
 106
 107 /*
 108  * Uprobes-specific fetch functions
 109  */
 110 #define DEFINE_FETCH_stack(type)                                        \
 111 static __kprobes void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,\
 112                                           void *offset, void *dest)     \
 113 {                                                                       \
 114         *(type *)dest = (type)get_user_stack_nth(regs,                  \
 115                                               ((unsigned long)offset)); \
 116 }
 117 DEFINE_BASIC_FETCH_FUNCS(stack)
 118 /* No string on the stack entry */
 119 #define fetch_stack_string      NULL
 120 #define fetch_stack_string_size NULL
 121
 122 #define DEFINE_FETCH_memory(type)                                       \
 123 static __kprobes void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,\
 124                                                 void *addr, void *dest) \
 125 {                                                                       \
 126         type retval;                                                    \
 127         void __user *vaddr = (void __force __user *) addr;              \
 128                                                                         \
 129         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
 130                 *(type *)dest = 0;                                      \
 131         else                                                            \
 132                 *(type *) dest = retval;                                \
 133 }
 134 DEFINE_BASIC_FETCH_FUNCS(memory)
 135 /*
 136  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 137  * length and relative data location.
 138  */
 139 static __kprobes void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 140                                                       void *addr, void *dest)
 141 {
 142         long ret;
 143         u32 rloc = *(u32 *)dest;
 144         int maxlen  = get_rloc_len(rloc);
 145         u8 *dst = get_rloc_data(dest);
 146         void __user *src = (void __force __user *) addr;
 147
 148         if (!maxlen)
 149                 return;
 150
 151         ret = strncpy_from_user(dst, src, maxlen);
 152
 153         if (ret < 0) {  /* Failed to fetch string */
 154                 ((u8 *)get_rloc_data(dest))[0] = '\0';
 155                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 156         } else {
 157                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 158         }
 159 }
 160
 161 static __kprobes void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 162                                                       void *addr, void *dest)
 163 {
 164         int len;
 165         void __user *vaddr = (void __force __user *) addr;
 166
 167         len = strnlen_user(vaddr, MAX_STRING_SIZE);
 168
 169         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 170                 *(u32 *)dest = 0;
 171         else
 172                 *(u32 *)dest = len;
 173 }
 174
 175 static unsigned long translate_user_vaddr(void *file_offset)
 176 {
 177         unsigned long base_addr;
 178         struct uprobe_dispatch_data *udd;
 179
 180         udd = (void *) current->utask->vaddr;
 181
 182         base_addr = udd->bp_addr - udd->tu->offset;
 183         return base_addr + (unsigned long)file_offset;
 184 }
 185
 186 #define DEFINE_FETCH_file_offset(type)                                  \
 187 static __kprobes void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,\
 188                                         void *offset, void *dest)       \
 189 {                                                                       \
 190         void *vaddr = (void *)translate_user_vaddr(offset);             \
 191                                                                         \
 192         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
 193 }
 194 DEFINE_BASIC_FETCH_FUNCS(file_offset)
 195 DEFINE_FETCH_file_offset(string)
 196 DEFINE_FETCH_file_offset(string_size)
 197
 198 /* Fetch type information table */
 199 const struct fetch_type uprobes_fetch_type_table[] = {
 200         /* Special types */
 201         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 202                                         sizeof(u32), 1, "__data_loc char[]"),
 203         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 204                                         string_size, sizeof(u32), 0, "u32"),
 205         /* Basic types */
 206         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 207         ASSIGN_FETCH_TYPE(u16, u16, 0),
 208         ASSIGN_FETCH_TYPE(u32, u32, 0),
 209         ASSIGN_FETCH_TYPE(u64, u64, 0),
 210         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 211         ASSIGN_FETCH_TYPE(s16, u16, 1),
 212         ASSIGN_FETCH_TYPE(s32, u32, 1),
 213         ASSIGN_FETCH_TYPE(s64, u64, 1),
 214
 215         ASSIGN_FETCH_TYPE_END
 216 };
 217
 218 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 219 {
 220         rwlock_init(&filter->rwlock);
 221         filter->nr_systemwide = 0;
 222         INIT_LIST_HEAD(&filter->perf_events);
 223 }
 224
 225 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 226 {
 227         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 228 }
 229
 230 static inline bool is_ret_probe(struct trace_uprobe *tu)
 231 {
 232         return tu->consumer.ret_handler != NULL;
 233 }
 234
 235 /*
 236  * Allocate new trace_uprobe and initialize it (including uprobes).
 237  */
 238 static struct trace_uprobe *
 239 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 240 {
 241         struct trace_uprobe *tu;
 242
 243         if (!event || !is_good_name(event))
 244                 return ERR_PTR(-EINVAL);
 245
 246         if (!group || !is_good_name(group))
 247                 return ERR_PTR(-EINVAL);
 248
 249         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 250         if (!tu)
 251                 return ERR_PTR(-ENOMEM);
 252
 253         tu->tp.call.class = &tu->tp.class;
 254         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 255         if (!tu->tp.call.name)
 256                 goto error;
 257
 258         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 259         if (!tu->tp.class.system)
 260                 goto error;
 261
 262         INIT_LIST_HEAD(&tu->list);
 263         tu->consumer.handler = uprobe_dispatcher;
 264         if (is_ret)
 265                 tu->consumer.ret_handler = uretprobe_dispatcher;
 266         init_trace_uprobe_filter(&tu->filter);
 267         tu->tp.call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
 268         return tu;
 269
 270 error:
 271         kfree(tu->tp.call.name);
 272         kfree(tu);
 273
 274         return ERR_PTR(-ENOMEM);
 275 }
 276
 277 static void free_trace_uprobe(struct trace_uprobe *tu)
 278 {
 279         int i;
 280
 281         for (i = 0; i < tu->tp.nr_args; i++)
 282                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 283
 284         iput(tu->inode);
 285         kfree(tu->tp.call.class->system);
 286         kfree(tu->tp.call.name);
 287         kfree(tu->filename);
 288         kfree(tu);
 289 }
 290
 291 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 292 {
 293         struct trace_uprobe *tu;
 294
 295         list_for_each_entry(tu, &uprobe_list, list)
 296                 if (strcmp(tu->tp.call.name, event) == 0 &&
 297                     strcmp(tu->tp.call.class->system, group) == 0)
 298                         return tu;
 299
 300         return NULL;
 301 }
 302
 303 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 304 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 305 {
 306         int ret;
 307
 308         ret = unregister_uprobe_event(tu);
 309         if (ret)
 310                 return ret;
 311
 312         list_del(&tu->list);
 313         free_trace_uprobe(tu);
 314         return 0;
 315 }
 316
 317 /* Register a trace_uprobe and probe_event */
 318 static int register_trace_uprobe(struct trace_uprobe *tu)
 319 {
 320         struct trace_uprobe *old_tu;
 321         int ret;
 322
 323         mutex_lock(&uprobe_lock);
 324
 325         /* register as an event */
 326         old_tu = find_probe_event(tu->tp.call.name, tu->tp.call.class->system);
 327         if (old_tu) {
 328                 /* delete old event */
 329                 ret = unregister_trace_uprobe(old_tu);
 330                 if (ret)
 331                         goto end;
 332         }
 333
 334         ret = register_uprobe_event(tu);
 335         if (ret) {
 336                 pr_warning("Failed to register probe event(%d)\n", ret);
 337                 goto end;
 338         }
 339
 340         list_add_tail(&tu->list, &uprobe_list);
 341
 342 end:
 343         mutex_unlock(&uprobe_lock);
 344
 345         return ret;
 346 }
 347
 348 /*
 349  * Argument syntax:
 350  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 351  *
 352  *  - Remove uprobe: -:[GRP/]EVENT
 353  */
 354 static int create_trace_uprobe(int argc, char **argv)
 355 {
 356         struct trace_uprobe *tu;
 357         struct inode *inode;
 358         char *arg, *event, *group, *filename;
 359         char buf[MAX_EVENT_NAME_LEN];
 360         struct path path;
 361         unsigned long offset;
 362         bool is_delete, is_return;
 363         int i, ret;
 364
 365         inode = NULL;
 366         ret = 0;
 367         is_delete = false;
 368         is_return = false;
 369         event = NULL;
 370         group = NULL;
 371
 372         /* argc must be >= 1 */
 373         if (argv[0][0] == '-')
 374                 is_delete = true;
 375         else if (argv[0][0] == 'r')
 376                 is_return = true;
 377         else if (argv[0][0] != 'p') {
 378                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 379                 return -EINVAL;
 380         }
 381
 382         if (argv[0][1] == ':') {
 383                 event = &argv[0][2];
 384                 arg = strchr(event, '/');
 385
 386                 if (arg) {
 387                         group = event;
 388                         event = arg + 1;
 389                         event[-1] = '\0';
 390
 391                         if (strlen(group) == 0) {
 392                                 pr_info("Group name is not specified\n");
 393                                 return -EINVAL;
 394                         }
 395                 }
 396                 if (strlen(event) == 0) {
 397                         pr_info("Event name is not specified\n");
 398                         return -EINVAL;
 399                 }
 400         }
 401         if (!group)
 402                 group = UPROBE_EVENT_SYSTEM;
 403
 404         if (is_delete) {
 405                 int ret;
 406
 407                 if (!event) {
 408                         pr_info("Delete command needs an event name.\n");
 409                         return -EINVAL;
 410                 }
 411                 mutex_lock(&uprobe_lock);
 412                 tu = find_probe_event(event, group);
 413
 414                 if (!tu) {
 415                         mutex_unlock(&uprobe_lock);
 416                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 417                         return -ENOENT;
 418                 }
 419                 /* delete an event */
 420                 ret = unregister_trace_uprobe(tu);
 421                 mutex_unlock(&uprobe_lock);
 422                 return ret;
 423         }
 424
 425         if (argc < 2) {
 426                 pr_info("Probe point is not specified.\n");
 427                 return -EINVAL;
 428         }
 429         if (isdigit(argv[1][0])) {
 430                 pr_info("probe point must be have a filename.\n");
 431                 return -EINVAL;
 432         }
 433         arg = strchr(argv[1], ':');
 434         if (!arg) {
 435                 ret = -EINVAL;
 436                 goto fail_address_parse;
 437         }
 438
 439         *arg++ = '\0';
 440         filename = argv[1];
 441         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 442         if (ret)
 443                 goto fail_address_parse;
 444
 445         inode = igrab(path.dentry->d_inode);
 446         path_put(&path);
 447
 448         if (!inode || !S_ISREG(inode->i_mode)) {
 449                 ret = -EINVAL;
 450                 goto fail_address_parse;
 451         }
 452
 453         ret = kstrtoul(arg, 0, &offset);
 454         if (ret)
 455                 goto fail_address_parse;
 456
 457         argc -= 2;
 458         argv += 2;
 459
 460         /* setup a probe */
 461         if (!event) {
 462                 char *tail;
 463                 char *ptr;
 464
 465                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 466                 if (!tail) {
 467                         ret = -ENOMEM;
 468                         goto fail_address_parse;
 469                 }
 470
 471                 ptr = strpbrk(tail, ".-_");
 472                 if (ptr)
 473                         *ptr = '\0';
 474
 475                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 476                 event = buf;
 477                 kfree(tail);
 478         }
 479
 480         tu = alloc_trace_uprobe(group, event, argc, is_return);
 481         if (IS_ERR(tu)) {
 482                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 483                 ret = PTR_ERR(tu);
 484                 goto fail_address_parse;
 485         }
 486         tu->offset = offset;
 487         tu->inode = inode;
 488         tu->filename = kstrdup(filename, GFP_KERNEL);
 489
 490         if (!tu->filename) {
 491                 pr_info("Failed to allocate filename.\n");
 492                 ret = -ENOMEM;
 493                 goto error;
 494         }
 495
 496         /* parse arguments */
 497         ret = 0;
 498         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 499                 struct probe_arg *parg = &tu->tp.args[i];
 500
 501                 /* Increment count for freeing args in error case */
 502                 tu->tp.nr_args++;
 503
 504                 /* Parse argument name */
 505                 arg = strchr(argv[i], '=');
 506                 if (arg) {
 507                         *arg++ = '\0';
 508                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 509                 } else {
 510                         arg = argv[i];
 511                         /* If argument name is omitted, set "argN" */
 512                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 513                         parg->name = kstrdup(buf, GFP_KERNEL);
 514                 }
 515
 516                 if (!parg->name) {
 517                         pr_info("Failed to allocate argument[%d] name.\n", i);
 518                         ret = -ENOMEM;
 519                         goto error;
 520                 }
 521
 522                 if (!is_good_name(parg->name)) {
 523                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 524                         ret = -EINVAL;
 525                         goto error;
 526                 }
 527
 528                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 529                         pr_info("Argument[%d] name '%s' conflicts with "
 530                                 "another field.\n", i, argv[i]);
 531                         ret = -EINVAL;
 532                         goto error;
 533                 }
 534
 535                 /* Parse fetch argument */
 536                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 537                                                  is_return, false);
 538                 if (ret) {
 539                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 540                         goto error;
 541                 }
 542         }
 543
 544         ret = register_trace_uprobe(tu);
 545         if (ret)
 546                 goto error;
 547         return 0;
 548
 549 error:
 550         free_trace_uprobe(tu);
 551         return ret;
 552
 553 fail_address_parse:
 554         if (inode)
 555                 iput(inode);
 556
 557         pr_info("Failed to parse address or file.\n");
 558
 559         return ret;
 560 }
 561
 562 static int cleanup_all_probes(void)
 563 {
 564         struct trace_uprobe *tu;
 565         int ret = 0;
 566
 567         mutex_lock(&uprobe_lock);
 568         while (!list_empty(&uprobe_list)) {
 569                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 570                 ret = unregister_trace_uprobe(tu);
 571                 if (ret)
 572                         break;
 573         }
 574         mutex_unlock(&uprobe_lock);
 575         return ret;
 576 }
 577
 578 /* Probes listing interfaces */
 579 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 580 {
 581         mutex_lock(&uprobe_lock);
 582         return seq_list_start(&uprobe_list, *pos);
 583 }
 584
 585 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 586 {
 587         return seq_list_next(v, &uprobe_list, pos);
 588 }
 589
 590 static void probes_seq_stop(struct seq_file *m, void *v)
 591 {
 592         mutex_unlock(&uprobe_lock);
 593 }
 594
 595 static int probes_seq_show(struct seq_file *m, void *v)
 596 {
 597         struct trace_uprobe *tu = v;
 598         char c = is_ret_probe(tu) ? 'r' : 'p';
 599         int i;
 600
 601         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system, tu->tp.call.name);
 602         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 603
 604         for (i = 0; i < tu->tp.nr_args; i++)
 605                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 606
 607         seq_printf(m, "\n");
 608         return 0;
 609 }
 610
 611 static const struct seq_operations probes_seq_op = {
 612         .start  = probes_seq_start,
 613         .next   = probes_seq_next,
 614         .stop   = probes_seq_stop,
 615         .show   = probes_seq_show
 616 };
 617
 618 static int probes_open(struct inode *inode, struct file *file)
 619 {
 620         int ret;
 621
 622         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 623                 ret = cleanup_all_probes();
 624                 if (ret)
 625                         return ret;
 626         }
 627
 628         return seq_open(file, &probes_seq_op);
 629 }
 630
 631 static ssize_t probes_write(struct file *file, const char __user *buffer,
 632                             size_t count, loff_t *ppos)
 633 {
 634         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 635 }
 636
 637 static const struct file_operations uprobe_events_ops = {
 638         .owner          = THIS_MODULE,
 639         .open           = probes_open,
 640         .read           = seq_read,
 641         .llseek         = seq_lseek,
 642         .release        = seq_release,
 643         .write          = probes_write,
 644 };
 645
 646 /* Probes profiling interfaces */
 647 static int probes_profile_seq_show(struct seq_file *m, void *v)
 648 {
 649         struct trace_uprobe *tu = v;
 650
 651         seq_printf(m, "  %s %-44s %15lu\n", tu->filename, tu->tp.call.name, tu->nhit);
 652         return 0;
 653 }
 654
 655 static const struct seq_operations profile_seq_op = {
 656         .start  = probes_seq_start,
 657         .next   = probes_seq_next,
 658         .stop   = probes_seq_stop,
 659         .show   = probes_profile_seq_show
 660 };
 661
 662 static int profile_open(struct inode *inode, struct file *file)
 663 {
 664         return seq_open(file, &profile_seq_op);
 665 }
 666
 667 static const struct file_operations uprobe_profile_ops = {
 668         .owner          = THIS_MODULE,
 669         .open           = profile_open,
 670         .read           = seq_read,
 671         .llseek         = seq_lseek,
 672         .release        = seq_release,
 673 };
 674
 675 struct uprobe_cpu_buffer {
 676         struct mutex mutex;
 677         void *buf;
 678 };
 679 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 680 static int uprobe_buffer_refcnt;
 681
 682 static int uprobe_buffer_init(void)
 683 {
 684         int cpu, err_cpu;
 685
 686         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 687         if (uprobe_cpu_buffer == NULL)
 688                 return -ENOMEM;
 689
 690         for_each_possible_cpu(cpu) {
 691                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 692                                                   GFP_KERNEL, 0);
 693                 if (p == NULL) {
 694                         err_cpu = cpu;
 695                         goto err;
 696                 }
 697                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 698                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 699         }
 700
 701         return 0;
 702
 703 err:
 704         for_each_possible_cpu(cpu) {
 705                 if (cpu == err_cpu)
 706                         break;
 707                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 708         }
 709
 710         free_percpu(uprobe_cpu_buffer);
 711         return -ENOMEM;
 712 }
 713
 714 static int uprobe_buffer_enable(void)
 715 {
 716         int ret = 0;
 717
 718         BUG_ON(!mutex_is_locked(&event_mutex));
 719
 720         if (uprobe_buffer_refcnt++ == 0) {
 721                 ret = uprobe_buffer_init();
 722                 if (ret < 0)
 723                         uprobe_buffer_refcnt--;
 724         }
 725
 726         return ret;
 727 }
 728
 729 static void uprobe_buffer_disable(void)
 730 {
 731         BUG_ON(!mutex_is_locked(&event_mutex));
 732
 733         if (--uprobe_buffer_refcnt == 0) {
 734                 free_percpu(uprobe_cpu_buffer);
 735                 uprobe_cpu_buffer = NULL;
 736         }
 737 }
 738
 739 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 740 {
 741         struct uprobe_cpu_buffer *ucb;
 742         int cpu;
 743
 744         cpu = raw_smp_processor_id();
 745         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 746
 747         /*
 748          * Use per-cpu buffers for fastest access, but we might migrate
 749          * so the mutex makes sure we have sole access to it.
 750          */
 751         mutex_lock(&ucb->mutex);
 752
 753         return ucb;
 754 }
 755
 756 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 757 {
 758         mutex_unlock(&ucb->mutex);
 759 }
 760
 761 static void uprobe_trace_print(struct trace_uprobe *tu,
 762                                 unsigned long func, struct pt_regs *regs)
 763 {
 764         struct uprobe_trace_entry_head *entry;
 765         struct ring_buffer_event *event;
 766         struct ring_buffer *buffer;
 767         struct uprobe_cpu_buffer *ucb;
 768         void *data;
 769         int size, dsize, esize;
 770         struct ftrace_event_call *call = &tu->tp.call;
 771
 772         dsize = __get_data_size(&tu->tp, regs);
 773         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 774
 775         if (WARN_ON_ONCE(!uprobe_cpu_buffer || tu->tp.size + dsize > PAGE_SIZE))
 776                 return;
 777
 778         ucb = uprobe_buffer_get();
 779         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
 780
 781         size = esize + tu->tp.size + dsize;
 782         event = trace_current_buffer_lock_reserve(&buffer, call->event.type,
 783                                                   size, 0, 0);
 784         if (!event)
 785                 goto out;
 786
 787         entry = ring_buffer_event_data(event);
 788         if (is_ret_probe(tu)) {
 789                 entry->vaddr[0] = func;
 790                 entry->vaddr[1] = instruction_pointer(regs);
 791                 data = DATAOF_TRACE_ENTRY(entry, true);
 792         } else {
 793                 entry->vaddr[0] = instruction_pointer(regs);
 794                 data = DATAOF_TRACE_ENTRY(entry, false);
 795         }
 796
 797         memcpy(data, ucb->buf, tu->tp.size + dsize);
 798
 799         if (!call_filter_check_discard(call, entry, buffer, event))
 800                 trace_buffer_unlock_commit(buffer, event, 0, 0);
 801
 802 out:
 803         uprobe_buffer_put(ucb);
 804 }
 805
 806 /* uprobe handler */
 807 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs)
 808 {
 809         if (!is_ret_probe(tu))
 810                 uprobe_trace_print(tu, 0, regs);
 811         return 0;
 812 }
 813
 814 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 815                                 struct pt_regs *regs)
 816 {
 817         uprobe_trace_print(tu, func, regs);
 818 }
 819
 820 /* Event entry printers */
 821 static enum print_line_t
 822 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 823 {
 824         struct uprobe_trace_entry_head *entry;
 825         struct trace_seq *s = &iter->seq;
 826         struct trace_uprobe *tu;
 827         u8 *data;
 828         int i;
 829
 830         entry = (struct uprobe_trace_entry_head *)iter->ent;
 831         tu = container_of(event, struct trace_uprobe, tp.call.event);
 832
 833         if (is_ret_probe(tu)) {
 834                 if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)", tu->tp.call.name,
 835                                         entry->vaddr[1], entry->vaddr[0]))
 836                         goto partial;
 837                 data = DATAOF_TRACE_ENTRY(entry, true);
 838         } else {
 839                 if (!trace_seq_printf(s, "%s: (0x%lx)", tu->tp.call.name,
 840                                         entry->vaddr[0]))
 841                         goto partial;
 842                 data = DATAOF_TRACE_ENTRY(entry, false);
 843         }
 844
 845         for (i = 0; i < tu->tp.nr_args; i++) {
 846                 struct probe_arg *parg = &tu->tp.args[i];
 847
 848                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 849                         goto partial;
 850         }
 851
 852         if (trace_seq_puts(s, "\n"))
 853                 return TRACE_TYPE_HANDLED;
 854
 855 partial:
 856         return TRACE_TYPE_PARTIAL_LINE;
 857 }
 858
 859 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 860                                 enum uprobe_filter_ctx ctx,
 861                                 struct mm_struct *mm);
 862
 863 static int
 864 probe_event_enable(struct trace_uprobe *tu, int flag, filter_func_t filter)
 865 {
 866         int ret = 0;
 867
 868         if (trace_probe_is_enabled(&tu->tp))
 869                 return -EINTR;
 870
 871         ret = uprobe_buffer_enable();
 872         if (ret < 0)
 873                 return ret;
 874
 875         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 876
 877         tu->tp.flags |= flag;
 878         tu->consumer.filter = filter;
 879         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 880         if (ret)
 881                 tu->tp.flags &= ~flag;
 882
 883         return ret;
 884 }
 885
 886 static void probe_event_disable(struct trace_uprobe *tu, int flag)
 887 {
 888         if (!trace_probe_is_enabled(&tu->tp))
 889                 return;
 890
 891         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 892
 893         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 894         tu->tp.flags &= ~flag;
 895
 896         uprobe_buffer_disable();
 897 }
 898
 899 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 900 {
 901         int ret, i, size;
 902         struct uprobe_trace_entry_head field;
 903         struct trace_uprobe *tu = event_call->data;
 904
 905         if (is_ret_probe(tu)) {
 906                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 907                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 908                 size = SIZEOF_TRACE_ENTRY(true);
 909         } else {
 910                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 911                 size = SIZEOF_TRACE_ENTRY(false);
 912         }
 913         /* Set argument names as fields */
 914         for (i = 0; i < tu->tp.nr_args; i++) {
 915                 struct probe_arg *parg = &tu->tp.args[i];
 916
 917                 ret = trace_define_field(event_call, parg->type->fmttype,
 918                                          parg->name, size + parg->offset,
 919                                          parg->type->size, parg->type->is_signed,
 920                                          FILTER_OTHER);
 921
 922                 if (ret)
 923                         return ret;
 924         }
 925         return 0;
 926 }
 927
 928 #ifdef CONFIG_PERF_EVENTS
 929 static bool
 930 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 931 {
 932         struct perf_event *event;
 933
 934         if (filter->nr_systemwide)
 935                 return true;
 936
 937         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
 938                 if (event->hw.tp_target->mm == mm)
 939                         return true;
 940         }
 941
 942         return false;
 943 }
 944
 945 static inline bool
 946 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 947 {
 948         return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 949 }
 950
 951 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 952 {
 953         bool done;
 954
 955         write_lock(&tu->filter.rwlock);
 956         if (event->hw.tp_target) {
 957                 /*
 958                  * event->parent != NULL means copy_process(), we can avoid
 959                  * uprobe_apply(). current->mm must be probed and we can rely
 960                  * on dup_mmap() which preserves the already installed bp's.
 961                  *
 962                  * attr.enable_on_exec means that exec/mmap will install the
 963                  * breakpoints we need.
 964                  */
 965                 done = tu->filter.nr_systemwide ||
 966                         event->parent || event->attr.enable_on_exec ||
 967                         uprobe_filter_event(tu, event);
 968                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
 969         } else {
 970                 done = tu->filter.nr_systemwide;
 971                 tu->filter.nr_systemwide++;
 972         }
 973         write_unlock(&tu->filter.rwlock);
 974
 975         if (!done)
 976                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
 977
 978         return 0;
 979 }
 980
 981 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 982 {
 983         bool done;
 984
 985         write_lock(&tu->filter.rwlock);
 986         if (event->hw.tp_target) {
 987                 list_del(&event->hw.tp_list);
 988                 done = tu->filter.nr_systemwide ||
 989                         (event->hw.tp_target->flags & PF_EXITING) ||
 990                         uprobe_filter_event(tu, event);
 991         } else {
 992                 tu->filter.nr_systemwide--;
 993                 done = tu->filter.nr_systemwide;
 994         }
 995         write_unlock(&tu->filter.rwlock);
 996
 997         if (!done)
 998                 uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 999
1000         return 0;
1001 }
1002
1003 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1004                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1005 {
1006         struct trace_uprobe *tu;
1007         int ret;
1008
1009         tu = container_of(uc, struct trace_uprobe, consumer);
1010         read_lock(&tu->filter.rwlock);
1011         ret = __uprobe_perf_filter(&tu->filter, mm);
1012         read_unlock(&tu->filter.rwlock);
1013
1014         return ret;
1015 }
1016
1017 static void uprobe_perf_print(struct trace_uprobe *tu,
1018                                 unsigned long func, struct pt_regs *regs)
1019 {
1020         struct ftrace_event_call *call = &tu->tp.call;
1021         struct uprobe_trace_entry_head *entry;
1022         struct hlist_head *head;
1023         struct uprobe_cpu_buffer *ucb;
1024         void *data;
1025         int size, dsize, esize;
1026         int rctx;
1027
1028         dsize = __get_data_size(&tu->tp, regs);
1029         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1030
1031         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1032                 return;
1033
1034         size = esize + tu->tp.size + dsize;
1035         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1036         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1037                 return;
1038
1039         ucb = uprobe_buffer_get();
1040         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1041
1042         preempt_disable();
1043         head = this_cpu_ptr(call->perf_events);
1044         if (hlist_empty(head))
1045                 goto out;
1046
1047         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1048         if (!entry)
1049                 goto out;
1050
1051         if (is_ret_probe(tu)) {
1052                 entry->vaddr[0] = func;
1053                 entry->vaddr[1] = instruction_pointer(regs);
1054                 data = DATAOF_TRACE_ENTRY(entry, true);
1055         } else {
1056                 entry->vaddr[0] = instruction_pointer(regs);
1057                 data = DATAOF_TRACE_ENTRY(entry, false);
1058         }
1059
1060         memcpy(data, ucb->buf, tu->tp.size + dsize);
1061
1062         if (size - esize > tu->tp.size + dsize) {
1063                 int len = tu->tp.size + dsize;
1064
1065                 memset(data + len, 0, size - esize - len);
1066         }
1067
1068         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1069  out:
1070         preempt_enable();
1071         uprobe_buffer_put(ucb);
1072 }
1073
1074 /* uprobe profile handler */
1075 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs)
1076 {
1077         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1078                 return UPROBE_HANDLER_REMOVE;
1079
1080         if (!is_ret_probe(tu))
1081                 uprobe_perf_print(tu, 0, regs);
1082         return 0;
1083 }
1084
1085 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1086                                 struct pt_regs *regs)
1087 {
1088         uprobe_perf_print(tu, func, regs);
1089 }
1090 #endif  /* CONFIG_PERF_EVENTS */
1091
1092 static
1093 int trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type, void *data)
1094 {
1095         struct trace_uprobe *tu = event->data;
1096
1097         switch (type) {
1098         case TRACE_REG_REGISTER:
1099                 return probe_event_enable(tu, TP_FLAG_TRACE, NULL);
1100
1101         case TRACE_REG_UNREGISTER:
1102                 probe_event_disable(tu, TP_FLAG_TRACE);
1103                 return 0;
1104
1105 #ifdef CONFIG_PERF_EVENTS
1106         case TRACE_REG_PERF_REGISTER:
1107                 return probe_event_enable(tu, TP_FLAG_PROFILE, uprobe_perf_filter);
1108
1109         case TRACE_REG_PERF_UNREGISTER:
1110                 probe_event_disable(tu, TP_FLAG_PROFILE);
1111                 return 0;
1112
1113         case TRACE_REG_PERF_OPEN:
1114                 return uprobe_perf_open(tu, data);
1115
1116         case TRACE_REG_PERF_CLOSE:
1117                 return uprobe_perf_close(tu, data);
1118
1119 #endif
1120         default:
1121                 return 0;
1122         }
1123         return 0;
1124 }
1125
1126 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1127 {
1128         struct trace_uprobe *tu;
1129         struct uprobe_dispatch_data udd;
1130         int ret = 0;
1131
1132         tu = container_of(con, struct trace_uprobe, consumer);
1133         tu->nhit++;
1134
1135         udd.tu = tu;
1136         udd.bp_addr = instruction_pointer(regs);
1137
1138         current->utask->vaddr = (unsigned long) &udd;
1139
1140         if (tu->tp.flags & TP_FLAG_TRACE)
1141                 ret |= uprobe_trace_func(tu, regs);
1142
1143 #ifdef CONFIG_PERF_EVENTS
1144         if (tu->tp.flags & TP_FLAG_PROFILE)
1145                 ret |= uprobe_perf_func(tu, regs);
1146 #endif
1147         return ret;
1148 }
1149
1150 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1151                                 unsigned long func, struct pt_regs *regs)
1152 {
1153         struct trace_uprobe *tu;
1154         struct uprobe_dispatch_data udd;
1155
1156         tu = container_of(con, struct trace_uprobe, consumer);
1157
1158         udd.tu = tu;
1159         udd.bp_addr = func;
1160
1161         current->utask->vaddr = (unsigned long) &udd;
1162
1163         if (tu->tp.flags & TP_FLAG_TRACE)
1164                 uretprobe_trace_func(tu, func, regs);
1165
1166 #ifdef CONFIG_PERF_EVENTS
1167         if (tu->tp.flags & TP_FLAG_PROFILE)
1168                 uretprobe_perf_func(tu, func, regs);
1169 #endif
1170         return 0;
1171 }
1172
1173 static struct trace_event_functions uprobe_funcs = {
1174         .trace          = print_uprobe_event
1175 };
1176
1177 static int register_uprobe_event(struct trace_uprobe *tu)
1178 {
1179         struct ftrace_event_call *call = &tu->tp.call;
1180         int ret;
1181
1182         /* Initialize ftrace_event_call */
1183         INIT_LIST_HEAD(&call->class->fields);
1184         call->event.funcs = &uprobe_funcs;
1185         call->class->define_fields = uprobe_event_define_fields;
1186
1187         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1188                 return -ENOMEM;
1189
1190         ret = register_ftrace_event(&call->event);
1191         if (!ret) {
1192                 kfree(call->print_fmt);
1193                 return -ENODEV;
1194         }
1195         call->flags = 0;
1196         call->class->reg = trace_uprobe_register;
1197         call->data = tu;
1198         ret = trace_add_event_call(call);
1199
1200         if (ret) {
1201                 pr_info("Failed to register uprobe event: %s\n", call->name);
1202                 kfree(call->print_fmt);
1203                 unregister_ftrace_event(&call->event);
1204         }
1205
1206         return ret;
1207 }
1208
1209 static int unregister_uprobe_event(struct trace_uprobe *tu)
1210 {
1211         int ret;
1212
1213         /* tu->event is unregistered in trace_remove_event_call() */
1214         ret = trace_remove_event_call(&tu->tp.call);
1215         if (ret)
1216                 return ret;
1217         kfree(tu->tp.call.print_fmt);
1218         tu->tp.call.print_fmt = NULL;
1219         return 0;
1220 }
1221
1222 /* Make a trace interface for controling probe points */
1223 static __init int init_uprobe_trace(void)
1224 {
1225         struct dentry *d_tracer;
1226
1227         d_tracer = tracing_init_dentry();
1228         if (!d_tracer)
1229                 return 0;
1230
1231         trace_create_file("uprobe_events", 0644, d_tracer,
1232                                     NULL, &uprobe_events_ops);
1233         /* Profile interface */
1234         trace_create_file("uprobe_profile", 0444, d_tracer,
1235                                     NULL, &uprobe_profile_ops);
1236         return 0;
1237 }
1238
1239 fs_initcall(init_uprobe_trace);