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 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 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 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 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 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         INIT_LIST_HEAD(&tu->tp.files);
 264         tu->consumer.handler = uprobe_dispatcher;
 265         if (is_ret)
 266                 tu->consumer.ret_handler = uretprobe_dispatcher;
 267         init_trace_uprobe_filter(&tu->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(ftrace_event_name(&tu->tp.call), 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(ftrace_event_name(&tu->tp.call),
 327                         tu->tp.call.class->system);
 328         if (old_tu) {
 329                 /* delete old event */
 330                 ret = unregister_trace_uprobe(old_tu);
 331                 if (ret)
 332                         goto end;
 333         }
 334
 335         ret = register_uprobe_event(tu);
 336         if (ret) {
 337                 pr_warning("Failed to register probe event(%d)\n", ret);
 338                 goto end;
 339         }
 340
 341         list_add_tail(&tu->list, &uprobe_list);
 342
 343 end:
 344         mutex_unlock(&uprobe_lock);
 345
 346         return ret;
 347 }
 348
 349 /*
 350  * Argument syntax:
 351  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 352  *
 353  *  - Remove uprobe: -:[GRP/]EVENT
 354  */
 355 static int create_trace_uprobe(int argc, char **argv)
 356 {
 357         struct trace_uprobe *tu;
 358         struct inode *inode;
 359         char *arg, *event, *group, *filename;
 360         char buf[MAX_EVENT_NAME_LEN];
 361         struct path path;
 362         unsigned long offset;
 363         bool is_delete, is_return;
 364         int i, ret;
 365
 366         inode = NULL;
 367         ret = 0;
 368         is_delete = false;
 369         is_return = false;
 370         event = NULL;
 371         group = NULL;
 372
 373         /* argc must be >= 1 */
 374         if (argv[0][0] == '-')
 375                 is_delete = true;
 376         else if (argv[0][0] == 'r')
 377                 is_return = true;
 378         else if (argv[0][0] != 'p') {
 379                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 380                 return -EINVAL;
 381         }
 382
 383         if (argv[0][1] == ':') {
 384                 event = &argv[0][2];
 385                 arg = strchr(event, '/');
 386
 387                 if (arg) {
 388                         group = event;
 389                         event = arg + 1;
 390                         event[-1] = '\0';
 391
 392                         if (strlen(group) == 0) {
 393                                 pr_info("Group name is not specified\n");
 394                                 return -EINVAL;
 395                         }
 396                 }
 397                 if (strlen(event) == 0) {
 398                         pr_info("Event name is not specified\n");
 399                         return -EINVAL;
 400                 }
 401         }
 402         if (!group)
 403                 group = UPROBE_EVENT_SYSTEM;
 404
 405         if (is_delete) {
 406                 int ret;
 407
 408                 if (!event) {
 409                         pr_info("Delete command needs an event name.\n");
 410                         return -EINVAL;
 411                 }
 412                 mutex_lock(&uprobe_lock);
 413                 tu = find_probe_event(event, group);
 414
 415                 if (!tu) {
 416                         mutex_unlock(&uprobe_lock);
 417                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 418                         return -ENOENT;
 419                 }
 420                 /* delete an event */
 421                 ret = unregister_trace_uprobe(tu);
 422                 mutex_unlock(&uprobe_lock);
 423                 return ret;
 424         }
 425
 426         if (argc < 2) {
 427                 pr_info("Probe point is not specified.\n");
 428                 return -EINVAL;
 429         }
 430         if (isdigit(argv[1][0])) {
 431                 pr_info("probe point must be have a filename.\n");
 432                 return -EINVAL;
 433         }
 434         arg = strchr(argv[1], ':');
 435         if (!arg) {
 436                 ret = -EINVAL;
 437                 goto fail_address_parse;
 438         }
 439
 440         *arg++ = '\0';
 441         filename = argv[1];
 442         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 443         if (ret)
 444                 goto fail_address_parse;
 445
 446         inode = igrab(path.dentry->d_inode);
 447         path_put(&path);
 448
 449         if (!inode || !S_ISREG(inode->i_mode)) {
 450                 ret = -EINVAL;
 451                 goto fail_address_parse;
 452         }
 453
 454         ret = kstrtoul(arg, 0, &offset);
 455         if (ret)
 456                 goto fail_address_parse;
 457
 458         argc -= 2;
 459         argv += 2;
 460
 461         /* setup a probe */
 462         if (!event) {
 463                 char *tail;
 464                 char *ptr;
 465
 466                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 467                 if (!tail) {
 468                         ret = -ENOMEM;
 469                         goto fail_address_parse;
 470                 }
 471
 472                 ptr = strpbrk(tail, ".-_");
 473                 if (ptr)
 474                         *ptr = '\0';
 475
 476                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 477                 event = buf;
 478                 kfree(tail);
 479         }
 480
 481         tu = alloc_trace_uprobe(group, event, argc, is_return);
 482         if (IS_ERR(tu)) {
 483                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 484                 ret = PTR_ERR(tu);
 485                 goto fail_address_parse;
 486         }
 487         tu->offset = offset;
 488         tu->inode = inode;
 489         tu->filename = kstrdup(filename, GFP_KERNEL);
 490
 491         if (!tu->filename) {
 492                 pr_info("Failed to allocate filename.\n");
 493                 ret = -ENOMEM;
 494                 goto error;
 495         }
 496
 497         /* parse arguments */
 498         ret = 0;
 499         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 500                 struct probe_arg *parg = &tu->tp.args[i];
 501
 502                 /* Increment count for freeing args in error case */
 503                 tu->tp.nr_args++;
 504
 505                 /* Parse argument name */
 506                 arg = strchr(argv[i], '=');
 507                 if (arg) {
 508                         *arg++ = '\0';
 509                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 510                 } else {
 511                         arg = argv[i];
 512                         /* If argument name is omitted, set "argN" */
 513                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 514                         parg->name = kstrdup(buf, GFP_KERNEL);
 515                 }
 516
 517                 if (!parg->name) {
 518                         pr_info("Failed to allocate argument[%d] name.\n", i);
 519                         ret = -ENOMEM;
 520                         goto error;
 521                 }
 522
 523                 if (!is_good_name(parg->name)) {
 524                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 525                         ret = -EINVAL;
 526                         goto error;
 527                 }
 528
 529                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 530                         pr_info("Argument[%d] name '%s' conflicts with "
 531                                 "another field.\n", i, argv[i]);
 532                         ret = -EINVAL;
 533                         goto error;
 534                 }
 535
 536                 /* Parse fetch argument */
 537                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 538                                                  is_return, false);
 539                 if (ret) {
 540                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 541                         goto error;
 542                 }
 543         }
 544
 545         ret = register_trace_uprobe(tu);
 546         if (ret)
 547                 goto error;
 548         return 0;
 549
 550 error:
 551         free_trace_uprobe(tu);
 552         return ret;
 553
 554 fail_address_parse:
 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,
 602                         ftrace_event_name(&tu->tp.call));
 603         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 604
 605         for (i = 0; i < tu->tp.nr_args; i++)
 606                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 607
 608         seq_putc(m, '\n');
 609         return 0;
 610 }
 611
 612 static const struct seq_operations probes_seq_op = {
 613         .start  = probes_seq_start,
 614         .next   = probes_seq_next,
 615         .stop   = probes_seq_stop,
 616         .show   = probes_seq_show
 617 };
 618
 619 static int probes_open(struct inode *inode, struct file *file)
 620 {
 621         int ret;
 622
 623         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 624                 ret = cleanup_all_probes();
 625                 if (ret)
 626                         return ret;
 627         }
 628
 629         return seq_open(file, &probes_seq_op);
 630 }
 631
 632 static ssize_t probes_write(struct file *file, const char __user *buffer,
 633                             size_t count, loff_t *ppos)
 634 {
 635         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 636 }
 637
 638 static const struct file_operations uprobe_events_ops = {
 639         .owner          = THIS_MODULE,
 640         .open           = probes_open,
 641         .read           = seq_read,
 642         .llseek         = seq_lseek,
 643         .release        = seq_release,
 644         .write          = probes_write,
 645 };
 646
 647 /* Probes profiling interfaces */
 648 static int probes_profile_seq_show(struct seq_file *m, void *v)
 649 {
 650         struct trace_uprobe *tu = v;
 651
 652         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 653                         ftrace_event_name(&tu->tp.call), tu->nhit);
 654         return 0;
 655 }
 656
 657 static const struct seq_operations profile_seq_op = {
 658         .start  = probes_seq_start,
 659         .next   = probes_seq_next,
 660         .stop   = probes_seq_stop,
 661         .show   = probes_profile_seq_show
 662 };
 663
 664 static int profile_open(struct inode *inode, struct file *file)
 665 {
 666         return seq_open(file, &profile_seq_op);
 667 }
 668
 669 static const struct file_operations uprobe_profile_ops = {
 670         .owner          = THIS_MODULE,
 671         .open           = profile_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = seq_release,
 675 };
 676
 677 struct uprobe_cpu_buffer {
 678         struct mutex mutex;
 679         void *buf;
 680 };
 681 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 682 static int uprobe_buffer_refcnt;
 683
 684 static int uprobe_buffer_init(void)
 685 {
 686         int cpu, err_cpu;
 687
 688         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 689         if (uprobe_cpu_buffer == NULL)
 690                 return -ENOMEM;
 691
 692         for_each_possible_cpu(cpu) {
 693                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 694                                                   GFP_KERNEL, 0);
 695                 if (p == NULL) {
 696                         err_cpu = cpu;
 697                         goto err;
 698                 }
 699                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 700                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 701         }
 702
 703         return 0;
 704
 705 err:
 706         for_each_possible_cpu(cpu) {
 707                 if (cpu == err_cpu)
 708                         break;
 709                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 710         }
 711
 712         free_percpu(uprobe_cpu_buffer);
 713         return -ENOMEM;
 714 }
 715
 716 static int uprobe_buffer_enable(void)
 717 {
 718         int ret = 0;
 719
 720         BUG_ON(!mutex_is_locked(&event_mutex));
 721
 722         if (uprobe_buffer_refcnt++ == 0) {
 723                 ret = uprobe_buffer_init();
 724                 if (ret < 0)
 725                         uprobe_buffer_refcnt--;
 726         }
 727
 728         return ret;
 729 }
 730
 731 static void uprobe_buffer_disable(void)
 732 {
 733         int cpu;
 734
 735         BUG_ON(!mutex_is_locked(&event_mutex));
 736
 737         if (--uprobe_buffer_refcnt == 0) {
 738                 for_each_possible_cpu(cpu)
 739                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 740                                                              cpu)->buf);
 741
 742                 free_percpu(uprobe_cpu_buffer);
 743                 uprobe_cpu_buffer = NULL;
 744         }
 745 }
 746
 747 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 748 {
 749         struct uprobe_cpu_buffer *ucb;
 750         int cpu;
 751
 752         cpu = raw_smp_processor_id();
 753         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 754
 755         /*
 756          * Use per-cpu buffers for fastest access, but we might migrate
 757          * so the mutex makes sure we have sole access to it.
 758          */
 759         mutex_lock(&ucb->mutex);
 760
 761         return ucb;
 762 }
 763
 764 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 765 {
 766         mutex_unlock(&ucb->mutex);
 767 }
 768
 769 static void __uprobe_trace_func(struct trace_uprobe *tu,
 770                                 unsigned long func, struct pt_regs *regs,
 771                                 struct uprobe_cpu_buffer *ucb, int dsize,
 772                                 struct ftrace_event_file *ftrace_file)
 773 {
 774         struct uprobe_trace_entry_head *entry;
 775         struct ring_buffer_event *event;
 776         struct ring_buffer *buffer;
 777         void *data;
 778         int size, esize;
 779         struct ftrace_event_call *call = &tu->tp.call;
 780
 781         WARN_ON(call != ftrace_file->event_call);
 782
 783         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 784                 return;
 785
 786         if (ftrace_trigger_soft_disabled(ftrace_file))
 787                 return;
 788
 789         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 790         size = esize + tu->tp.size + dsize;
 791         event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 792                                                 call->event.type, size, 0, 0);
 793         if (!event)
 794                 return;
 795
 796         entry = ring_buffer_event_data(event);
 797         if (is_ret_probe(tu)) {
 798                 entry->vaddr[0] = func;
 799                 entry->vaddr[1] = instruction_pointer(regs);
 800                 data = DATAOF_TRACE_ENTRY(entry, true);
 801         } else {
 802                 entry->vaddr[0] = instruction_pointer(regs);
 803                 data = DATAOF_TRACE_ENTRY(entry, false);
 804         }
 805
 806         memcpy(data, ucb->buf, tu->tp.size + dsize);
 807
 808         event_trigger_unlock_commit(ftrace_file, buffer, event, entry, 0, 0);
 809 }
 810
 811 /* uprobe handler */
 812 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 813                              struct uprobe_cpu_buffer *ucb, int dsize)
 814 {
 815         struct event_file_link *link;
 816
 817         if (is_ret_probe(tu))
 818                 return 0;
 819
 820         rcu_read_lock();
 821         list_for_each_entry_rcu(link, &tu->tp.files, list)
 822                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 823         rcu_read_unlock();
 824
 825         return 0;
 826 }
 827
 828 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 829                                  struct pt_regs *regs,
 830                                  struct uprobe_cpu_buffer *ucb, int dsize)
 831 {
 832         struct event_file_link *link;
 833
 834         rcu_read_lock();
 835         list_for_each_entry_rcu(link, &tu->tp.files, list)
 836                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 837         rcu_read_unlock();
 838 }
 839
 840 /* Event entry printers */
 841 static enum print_line_t
 842 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 843 {
 844         struct uprobe_trace_entry_head *entry;
 845         struct trace_seq *s = &iter->seq;
 846         struct trace_uprobe *tu;
 847         u8 *data;
 848         int i;
 849
 850         entry = (struct uprobe_trace_entry_head *)iter->ent;
 851         tu = container_of(event, struct trace_uprobe, tp.call.event);
 852
 853         if (is_ret_probe(tu)) {
 854                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 855                                  ftrace_event_name(&tu->tp.call),
 856                                  entry->vaddr[1], entry->vaddr[0]);
 857                 data = DATAOF_TRACE_ENTRY(entry, true);
 858         } else {
 859                 trace_seq_printf(s, "%s: (0x%lx)",
 860                                  ftrace_event_name(&tu->tp.call),
 861                                  entry->vaddr[0]);
 862                 data = DATAOF_TRACE_ENTRY(entry, false);
 863         }
 864
 865         for (i = 0; i < tu->tp.nr_args; i++) {
 866                 struct probe_arg *parg = &tu->tp.args[i];
 867
 868                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 869                         goto out;
 870         }
 871
 872         trace_seq_putc(s, '\n');
 873
 874  out:
 875         return trace_handle_return(s);
 876 }
 877
 878 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 879                                 enum uprobe_filter_ctx ctx,
 880                                 struct mm_struct *mm);
 881
 882 static int
 883 probe_event_enable(struct trace_uprobe *tu, struct ftrace_event_file *file,
 884                    filter_func_t filter)
 885 {
 886         bool enabled = trace_probe_is_enabled(&tu->tp);
 887         struct event_file_link *link = NULL;
 888         int ret;
 889
 890         if (file) {
 891                 if (tu->tp.flags & TP_FLAG_PROFILE)
 892                         return -EINTR;
 893
 894                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 895                 if (!link)
 896                         return -ENOMEM;
 897
 898                 link->file = file;
 899                 list_add_tail_rcu(&link->list, &tu->tp.files);
 900
 901                 tu->tp.flags |= TP_FLAG_TRACE;
 902         } else {
 903                 if (tu->tp.flags & TP_FLAG_TRACE)
 904                         return -EINTR;
 905
 906                 tu->tp.flags |= TP_FLAG_PROFILE;
 907         }
 908
 909         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 910
 911         if (enabled)
 912                 return 0;
 913
 914         ret = uprobe_buffer_enable();
 915         if (ret)
 916                 goto err_flags;
 917
 918         tu->consumer.filter = filter;
 919         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 920         if (ret)
 921                 goto err_buffer;
 922
 923         return 0;
 924
 925  err_buffer:
 926         uprobe_buffer_disable();
 927
 928  err_flags:
 929         if (file) {
 930                 list_del(&link->list);
 931                 kfree(link);
 932                 tu->tp.flags &= ~TP_FLAG_TRACE;
 933         } else {
 934                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 935         }
 936         return ret;
 937 }
 938
 939 static void
 940 probe_event_disable(struct trace_uprobe *tu, struct ftrace_event_file *file)
 941 {
 942         if (!trace_probe_is_enabled(&tu->tp))
 943                 return;
 944
 945         if (file) {
 946                 struct event_file_link *link;
 947
 948                 link = find_event_file_link(&tu->tp, file);
 949                 if (!link)
 950                         return;
 951
 952                 list_del_rcu(&link->list);
 953                 /* synchronize with u{,ret}probe_trace_func */
 954                 synchronize_sched();
 955                 kfree(link);
 956
 957                 if (!list_empty(&tu->tp.files))
 958                         return;
 959         }
 960
 961         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 962
 963         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 964         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 965
 966         uprobe_buffer_disable();
 967 }
 968
 969 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 970 {
 971         int ret, i, size;
 972         struct uprobe_trace_entry_head field;
 973         struct trace_uprobe *tu = event_call->data;
 974
 975         if (is_ret_probe(tu)) {
 976                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 977                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 978                 size = SIZEOF_TRACE_ENTRY(true);
 979         } else {
 980                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 981                 size = SIZEOF_TRACE_ENTRY(false);
 982         }
 983         /* Set argument names as fields */
 984         for (i = 0; i < tu->tp.nr_args; i++) {
 985                 struct probe_arg *parg = &tu->tp.args[i];
 986
 987                 ret = trace_define_field(event_call, parg->type->fmttype,
 988                                          parg->name, size + parg->offset,
 989                                          parg->type->size, parg->type->is_signed,
 990                                          FILTER_OTHER);
 991
 992                 if (ret)
 993                         return ret;
 994         }
 995         return 0;
 996 }
 997
 998 #ifdef CONFIG_PERF_EVENTS
 999 static bool
1000 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1001 {
1002         struct perf_event *event;
1003
1004         if (filter->nr_systemwide)
1005                 return true;
1006
1007         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1008                 if (event->hw.target->mm == mm)
1009                         return true;
1010         }
1011
1012         return false;
1013 }
1014
1015 static inline bool
1016 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1017 {
1018         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1019 }
1020
1021 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1022 {
1023         bool done;
1024
1025         write_lock(&tu->filter.rwlock);
1026         if (event->hw.target) {
1027                 list_del(&event->hw.tp_list);
1028                 done = tu->filter.nr_systemwide ||
1029                         (event->hw.target->flags & PF_EXITING) ||
1030                         uprobe_filter_event(tu, event);
1031         } else {
1032                 tu->filter.nr_systemwide--;
1033                 done = tu->filter.nr_systemwide;
1034         }
1035         write_unlock(&tu->filter.rwlock);
1036
1037         if (!done)
1038                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1039
1040         return 0;
1041 }
1042
1043 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1044 {
1045         bool done;
1046         int err;
1047
1048         write_lock(&tu->filter.rwlock);
1049         if (event->hw.target) {
1050                 /*
1051                  * event->parent != NULL means copy_process(), we can avoid
1052                  * uprobe_apply(). current->mm must be probed and we can rely
1053                  * on dup_mmap() which preserves the already installed bp's.
1054                  *
1055                  * attr.enable_on_exec means that exec/mmap will install the
1056                  * breakpoints we need.
1057                  */
1058                 done = tu->filter.nr_systemwide ||
1059                         event->parent || event->attr.enable_on_exec ||
1060                         uprobe_filter_event(tu, event);
1061                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1062         } else {
1063                 done = tu->filter.nr_systemwide;
1064                 tu->filter.nr_systemwide++;
1065         }
1066         write_unlock(&tu->filter.rwlock);
1067
1068         err = 0;
1069         if (!done) {
1070                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1071                 if (err)
1072                         uprobe_perf_close(tu, event);
1073         }
1074         return err;
1075 }
1076
1077 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1078                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1079 {
1080         struct trace_uprobe *tu;
1081         int ret;
1082
1083         tu = container_of(uc, struct trace_uprobe, consumer);
1084         read_lock(&tu->filter.rwlock);
1085         ret = __uprobe_perf_filter(&tu->filter, mm);
1086         read_unlock(&tu->filter.rwlock);
1087
1088         return ret;
1089 }
1090
1091 static void __uprobe_perf_func(struct trace_uprobe *tu,
1092                                unsigned long func, struct pt_regs *regs,
1093                                struct uprobe_cpu_buffer *ucb, int dsize)
1094 {
1095         struct ftrace_event_call *call = &tu->tp.call;
1096         struct uprobe_trace_entry_head *entry;
1097         struct hlist_head *head;
1098         void *data;
1099         int size, esize;
1100         int rctx;
1101
1102         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1103
1104         size = esize + tu->tp.size + dsize;
1105         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1106         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1107                 return;
1108
1109         preempt_disable();
1110         head = this_cpu_ptr(call->perf_events);
1111         if (hlist_empty(head))
1112                 goto out;
1113
1114         entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1115         if (!entry)
1116                 goto out;
1117
1118         if (is_ret_probe(tu)) {
1119                 entry->vaddr[0] = func;
1120                 entry->vaddr[1] = instruction_pointer(regs);
1121                 data = DATAOF_TRACE_ENTRY(entry, true);
1122         } else {
1123                 entry->vaddr[0] = instruction_pointer(regs);
1124                 data = DATAOF_TRACE_ENTRY(entry, false);
1125         }
1126
1127         memcpy(data, ucb->buf, tu->tp.size + dsize);
1128
1129         if (size - esize > tu->tp.size + dsize) {
1130                 int len = tu->tp.size + dsize;
1131
1132                 memset(data + len, 0, size - esize - len);
1133         }
1134
1135         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1136  out:
1137         preempt_enable();
1138 }
1139
1140 /* uprobe profile handler */
1141 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1142                             struct uprobe_cpu_buffer *ucb, int dsize)
1143 {
1144         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1145                 return UPROBE_HANDLER_REMOVE;
1146
1147         if (!is_ret_probe(tu))
1148                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1149         return 0;
1150 }
1151
1152 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1153                                 struct pt_regs *regs,
1154                                 struct uprobe_cpu_buffer *ucb, int dsize)
1155 {
1156         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1157 }
1158 #endif  /* CONFIG_PERF_EVENTS */
1159
1160 static int
1161 trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type,
1162                       void *data)
1163 {
1164         struct trace_uprobe *tu = event->data;
1165         struct ftrace_event_file *file = data;
1166
1167         switch (type) {
1168         case TRACE_REG_REGISTER:
1169                 return probe_event_enable(tu, file, NULL);
1170
1171         case TRACE_REG_UNREGISTER:
1172                 probe_event_disable(tu, file);
1173                 return 0;
1174
1175 #ifdef CONFIG_PERF_EVENTS
1176         case TRACE_REG_PERF_REGISTER:
1177                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1178
1179         case TRACE_REG_PERF_UNREGISTER:
1180                 probe_event_disable(tu, NULL);
1181                 return 0;
1182
1183         case TRACE_REG_PERF_OPEN:
1184                 return uprobe_perf_open(tu, data);
1185
1186         case TRACE_REG_PERF_CLOSE:
1187                 return uprobe_perf_close(tu, data);
1188
1189 #endif
1190         default:
1191                 return 0;
1192         }
1193         return 0;
1194 }
1195
1196 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1197 {
1198         struct trace_uprobe *tu;
1199         struct uprobe_dispatch_data udd;
1200         struct uprobe_cpu_buffer *ucb;
1201         int dsize, esize;
1202         int ret = 0;
1203
1204
1205         tu = container_of(con, struct trace_uprobe, consumer);
1206         tu->nhit++;
1207
1208         udd.tu = tu;
1209         udd.bp_addr = instruction_pointer(regs);
1210
1211         current->utask->vaddr = (unsigned long) &udd;
1212
1213         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1214                 return 0;
1215
1216         dsize = __get_data_size(&tu->tp, regs);
1217         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1218
1219         ucb = uprobe_buffer_get();
1220         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1221
1222         if (tu->tp.flags & TP_FLAG_TRACE)
1223                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1224
1225 #ifdef CONFIG_PERF_EVENTS
1226         if (tu->tp.flags & TP_FLAG_PROFILE)
1227                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1228 #endif
1229         uprobe_buffer_put(ucb);
1230         return ret;
1231 }
1232
1233 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1234                                 unsigned long func, struct pt_regs *regs)
1235 {
1236         struct trace_uprobe *tu;
1237         struct uprobe_dispatch_data udd;
1238         struct uprobe_cpu_buffer *ucb;
1239         int dsize, esize;
1240
1241         tu = container_of(con, struct trace_uprobe, consumer);
1242
1243         udd.tu = tu;
1244         udd.bp_addr = func;
1245
1246         current->utask->vaddr = (unsigned long) &udd;
1247
1248         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1249                 return 0;
1250
1251         dsize = __get_data_size(&tu->tp, regs);
1252         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1253
1254         ucb = uprobe_buffer_get();
1255         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1256
1257         if (tu->tp.flags & TP_FLAG_TRACE)
1258                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1259
1260 #ifdef CONFIG_PERF_EVENTS
1261         if (tu->tp.flags & TP_FLAG_PROFILE)
1262                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1263 #endif
1264         uprobe_buffer_put(ucb);
1265         return 0;
1266 }
1267
1268 static struct trace_event_functions uprobe_funcs = {
1269         .trace          = print_uprobe_event
1270 };
1271
1272 static int register_uprobe_event(struct trace_uprobe *tu)
1273 {
1274         struct ftrace_event_call *call = &tu->tp.call;
1275         int ret;
1276
1277         /* Initialize ftrace_event_call */
1278         INIT_LIST_HEAD(&call->class->fields);
1279         call->event.funcs = &uprobe_funcs;
1280         call->class->define_fields = uprobe_event_define_fields;
1281
1282         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1283                 return -ENOMEM;
1284
1285         ret = register_ftrace_event(&call->event);
1286         if (!ret) {
1287                 kfree(call->print_fmt);
1288                 return -ENODEV;
1289         }
1290
1291         call->class->reg = trace_uprobe_register;
1292         call->data = tu;
1293         ret = trace_add_event_call(call);
1294
1295         if (ret) {
1296                 pr_info("Failed to register uprobe event: %s\n",
1297                         ftrace_event_name(call));
1298                 kfree(call->print_fmt);
1299                 unregister_ftrace_event(&call->event);
1300         }
1301
1302         return ret;
1303 }
1304
1305 static int unregister_uprobe_event(struct trace_uprobe *tu)
1306 {
1307         int ret;
1308
1309         /* tu->event is unregistered in trace_remove_event_call() */
1310         ret = trace_remove_event_call(&tu->tp.call);
1311         if (ret)
1312                 return ret;
1313         kfree(tu->tp.call.print_fmt);
1314         tu->tp.call.print_fmt = NULL;
1315         return 0;
1316 }
1317
1318 /* Make a trace interface for controling probe points */
1319 static __init int init_uprobe_trace(void)
1320 {
1321         struct dentry *d_tracer;
1322
1323         d_tracer = tracing_init_dentry();
1324         if (IS_ERR(d_tracer))
1325                 return 0;
1326
1327         trace_create_file("uprobe_events", 0644, d_tracer,
1328                                     NULL, &uprobe_events_ops);
1329         /* Profile interface */
1330         trace_create_file("uprobe_profile", 0444, d_tracer,
1331                                     NULL, &uprobe_profile_ops);
1332         return 0;
1333 }
1334
1335 fs_initcall(init_uprobe_trace);