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         tu->tp.call.flags |= TRACE_EVENT_FL_USE_CALL_FILTER;
 269         return tu;
 270
 271 error:
 272         kfree(tu->tp.call.name);
 273         kfree(tu);
 274
 275         return ERR_PTR(-ENOMEM);
 276 }
 277
 278 static void free_trace_uprobe(struct trace_uprobe *tu)
 279 {
 280         int i;
 281
 282         for (i = 0; i < tu->tp.nr_args; i++)
 283                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 284
 285         iput(tu->inode);
 286         kfree(tu->tp.call.class->system);
 287         kfree(tu->tp.call.name);
 288         kfree(tu->filename);
 289         kfree(tu);
 290 }
 291
 292 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 293 {
 294         struct trace_uprobe *tu;
 295
 296         list_for_each_entry(tu, &uprobe_list, list)
 297                 if (strcmp(ftrace_event_name(&tu->tp.call), event) == 0 &&
 298                     strcmp(tu->tp.call.class->system, group) == 0)
 299                         return tu;
 300
 301         return NULL;
 302 }
 303
 304 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 305 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 306 {
 307         int ret;
 308
 309         ret = unregister_uprobe_event(tu);
 310         if (ret)
 311                 return ret;
 312
 313         list_del(&tu->list);
 314         free_trace_uprobe(tu);
 315         return 0;
 316 }
 317
 318 /* Register a trace_uprobe and probe_event */
 319 static int register_trace_uprobe(struct trace_uprobe *tu)
 320 {
 321         struct trace_uprobe *old_tu;
 322         int ret;
 323
 324         mutex_lock(&uprobe_lock);
 325
 326         /* register as an event */
 327         old_tu = find_probe_event(ftrace_event_name(&tu->tp.call),
 328                         tu->tp.call.class->system);
 329         if (old_tu) {
 330                 /* delete old event */
 331                 ret = unregister_trace_uprobe(old_tu);
 332                 if (ret)
 333                         goto end;
 334         }
 335
 336         ret = register_uprobe_event(tu);
 337         if (ret) {
 338                 pr_warning("Failed to register probe event(%d)\n", ret);
 339                 goto end;
 340         }
 341
 342         list_add_tail(&tu->list, &uprobe_list);
 343
 344 end:
 345         mutex_unlock(&uprobe_lock);
 346
 347         return ret;
 348 }
 349
 350 /*
 351  * Argument syntax:
 352  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 353  *
 354  *  - Remove uprobe: -:[GRP/]EVENT
 355  */
 356 static int create_trace_uprobe(int argc, char **argv)
 357 {
 358         struct trace_uprobe *tu;
 359         struct inode *inode;
 360         char *arg, *event, *group, *filename;
 361         char buf[MAX_EVENT_NAME_LEN];
 362         struct path path;
 363         unsigned long offset;
 364         bool is_delete, is_return;
 365         int i, ret;
 366
 367         inode = NULL;
 368         ret = 0;
 369         is_delete = false;
 370         is_return = false;
 371         event = NULL;
 372         group = NULL;
 373
 374         /* argc must be >= 1 */
 375         if (argv[0][0] == '-')
 376                 is_delete = true;
 377         else if (argv[0][0] == 'r')
 378                 is_return = true;
 379         else if (argv[0][0] != 'p') {
 380                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 381                 return -EINVAL;
 382         }
 383
 384         if (argv[0][1] == ':') {
 385                 event = &argv[0][2];
 386                 arg = strchr(event, '/');
 387
 388                 if (arg) {
 389                         group = event;
 390                         event = arg + 1;
 391                         event[-1] = '\0';
 392
 393                         if (strlen(group) == 0) {
 394                                 pr_info("Group name is not specified\n");
 395                                 return -EINVAL;
 396                         }
 397                 }
 398                 if (strlen(event) == 0) {
 399                         pr_info("Event name is not specified\n");
 400                         return -EINVAL;
 401                 }
 402         }
 403         if (!group)
 404                 group = UPROBE_EVENT_SYSTEM;
 405
 406         if (is_delete) {
 407                 int ret;
 408
 409                 if (!event) {
 410                         pr_info("Delete command needs an event name.\n");
 411                         return -EINVAL;
 412                 }
 413                 mutex_lock(&uprobe_lock);
 414                 tu = find_probe_event(event, group);
 415
 416                 if (!tu) {
 417                         mutex_unlock(&uprobe_lock);
 418                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 419                         return -ENOENT;
 420                 }
 421                 /* delete an event */
 422                 ret = unregister_trace_uprobe(tu);
 423                 mutex_unlock(&uprobe_lock);
 424                 return ret;
 425         }
 426
 427         if (argc < 2) {
 428                 pr_info("Probe point is not specified.\n");
 429                 return -EINVAL;
 430         }
 431         if (isdigit(argv[1][0])) {
 432                 pr_info("probe point must be have a filename.\n");
 433                 return -EINVAL;
 434         }
 435         arg = strchr(argv[1], ':');
 436         if (!arg) {
 437                 ret = -EINVAL;
 438                 goto fail_address_parse;
 439         }
 440
 441         *arg++ = '\0';
 442         filename = argv[1];
 443         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 444         if (ret)
 445                 goto fail_address_parse;
 446
 447         inode = igrab(path.dentry->d_inode);
 448         path_put(&path);
 449
 450         if (!inode || !S_ISREG(inode->i_mode)) {
 451                 ret = -EINVAL;
 452                 goto fail_address_parse;
 453         }
 454
 455         ret = kstrtoul(arg, 0, &offset);
 456         if (ret)
 457                 goto fail_address_parse;
 458
 459         argc -= 2;
 460         argv += 2;
 461
 462         /* setup a probe */
 463         if (!event) {
 464                 char *tail;
 465                 char *ptr;
 466
 467                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 468                 if (!tail) {
 469                         ret = -ENOMEM;
 470                         goto fail_address_parse;
 471                 }
 472
 473                 ptr = strpbrk(tail, ".-_");
 474                 if (ptr)
 475                         *ptr = '\0';
 476
 477                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 478                 event = buf;
 479                 kfree(tail);
 480         }
 481
 482         tu = alloc_trace_uprobe(group, event, argc, is_return);
 483         if (IS_ERR(tu)) {
 484                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 485                 ret = PTR_ERR(tu);
 486                 goto fail_address_parse;
 487         }
 488         tu->offset = offset;
 489         tu->inode = inode;
 490         tu->filename = kstrdup(filename, GFP_KERNEL);
 491
 492         if (!tu->filename) {
 493                 pr_info("Failed to allocate filename.\n");
 494                 ret = -ENOMEM;
 495                 goto error;
 496         }
 497
 498         /* parse arguments */
 499         ret = 0;
 500         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 501                 struct probe_arg *parg = &tu->tp.args[i];
 502
 503                 /* Increment count for freeing args in error case */
 504                 tu->tp.nr_args++;
 505
 506                 /* Parse argument name */
 507                 arg = strchr(argv[i], '=');
 508                 if (arg) {
 509                         *arg++ = '\0';
 510                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 511                 } else {
 512                         arg = argv[i];
 513                         /* If argument name is omitted, set "argN" */
 514                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 515                         parg->name = kstrdup(buf, GFP_KERNEL);
 516                 }
 517
 518                 if (!parg->name) {
 519                         pr_info("Failed to allocate argument[%d] name.\n", i);
 520                         ret = -ENOMEM;
 521                         goto error;
 522                 }
 523
 524                 if (!is_good_name(parg->name)) {
 525                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 526                         ret = -EINVAL;
 527                         goto error;
 528                 }
 529
 530                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 531                         pr_info("Argument[%d] name '%s' conflicts with "
 532                                 "another field.\n", i, argv[i]);
 533                         ret = -EINVAL;
 534                         goto error;
 535                 }
 536
 537                 /* Parse fetch argument */
 538                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 539                                                  is_return, false);
 540                 if (ret) {
 541                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 542                         goto error;
 543                 }
 544         }
 545
 546         ret = register_trace_uprobe(tu);
 547         if (ret)
 548                 goto error;
 549         return 0;
 550
 551 error:
 552         free_trace_uprobe(tu);
 553         return ret;
 554
 555 fail_address_parse:
 556         if (inode)
 557                 iput(inode);
 558
 559         pr_info("Failed to parse address or file.\n");
 560
 561         return ret;
 562 }
 563
 564 static int cleanup_all_probes(void)
 565 {
 566         struct trace_uprobe *tu;
 567         int ret = 0;
 568
 569         mutex_lock(&uprobe_lock);
 570         while (!list_empty(&uprobe_list)) {
 571                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 572                 ret = unregister_trace_uprobe(tu);
 573                 if (ret)
 574                         break;
 575         }
 576         mutex_unlock(&uprobe_lock);
 577         return ret;
 578 }
 579
 580 /* Probes listing interfaces */
 581 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 582 {
 583         mutex_lock(&uprobe_lock);
 584         return seq_list_start(&uprobe_list, *pos);
 585 }
 586
 587 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 588 {
 589         return seq_list_next(v, &uprobe_list, pos);
 590 }
 591
 592 static void probes_seq_stop(struct seq_file *m, void *v)
 593 {
 594         mutex_unlock(&uprobe_lock);
 595 }
 596
 597 static int probes_seq_show(struct seq_file *m, void *v)
 598 {
 599         struct trace_uprobe *tu = v;
 600         char c = is_ret_probe(tu) ? 'r' : 'p';
 601         int i;
 602
 603         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 604                         ftrace_event_name(&tu->tp.call));
 605         seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
 606
 607         for (i = 0; i < tu->tp.nr_args; i++)
 608                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 609
 610         seq_printf(m, "\n");
 611         return 0;
 612 }
 613
 614 static const struct seq_operations probes_seq_op = {
 615         .start  = probes_seq_start,
 616         .next   = probes_seq_next,
 617         .stop   = probes_seq_stop,
 618         .show   = probes_seq_show
 619 };
 620
 621 static int probes_open(struct inode *inode, struct file *file)
 622 {
 623         int ret;
 624
 625         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 626                 ret = cleanup_all_probes();
 627                 if (ret)
 628                         return ret;
 629         }
 630
 631         return seq_open(file, &probes_seq_op);
 632 }
 633
 634 static ssize_t probes_write(struct file *file, const char __user *buffer,
 635                             size_t count, loff_t *ppos)
 636 {
 637         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 638 }
 639
 640 static const struct file_operations uprobe_events_ops = {
 641         .owner          = THIS_MODULE,
 642         .open           = probes_open,
 643         .read           = seq_read,
 644         .llseek         = seq_lseek,
 645         .release        = seq_release,
 646         .write          = probes_write,
 647 };
 648
 649 /* Probes profiling interfaces */
 650 static int probes_profile_seq_show(struct seq_file *m, void *v)
 651 {
 652         struct trace_uprobe *tu = v;
 653
 654         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 655                         ftrace_event_name(&tu->tp.call), tu->nhit);
 656         return 0;
 657 }
 658
 659 static const struct seq_operations profile_seq_op = {
 660         .start  = probes_seq_start,
 661         .next   = probes_seq_next,
 662         .stop   = probes_seq_stop,
 663         .show   = probes_profile_seq_show
 664 };
 665
 666 static int profile_open(struct inode *inode, struct file *file)
 667 {
 668         return seq_open(file, &profile_seq_op);
 669 }
 670
 671 static const struct file_operations uprobe_profile_ops = {
 672         .owner          = THIS_MODULE,
 673         .open           = profile_open,
 674         .read           = seq_read,
 675         .llseek         = seq_lseek,
 676         .release        = seq_release,
 677 };
 678
 679 struct uprobe_cpu_buffer {
 680         struct mutex mutex;
 681         void *buf;
 682 };
 683 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 684 static int uprobe_buffer_refcnt;
 685
 686 static int uprobe_buffer_init(void)
 687 {
 688         int cpu, err_cpu;
 689
 690         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 691         if (uprobe_cpu_buffer == NULL)
 692                 return -ENOMEM;
 693
 694         for_each_possible_cpu(cpu) {
 695                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 696                                                   GFP_KERNEL, 0);
 697                 if (p == NULL) {
 698                         err_cpu = cpu;
 699                         goto err;
 700                 }
 701                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 702                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 703         }
 704
 705         return 0;
 706
 707 err:
 708         for_each_possible_cpu(cpu) {
 709                 if (cpu == err_cpu)
 710                         break;
 711                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 712         }
 713
 714         free_percpu(uprobe_cpu_buffer);
 715         return -ENOMEM;
 716 }
 717
 718 static int uprobe_buffer_enable(void)
 719 {
 720         int ret = 0;
 721
 722         BUG_ON(!mutex_is_locked(&event_mutex));
 723
 724         if (uprobe_buffer_refcnt++ == 0) {
 725                 ret = uprobe_buffer_init();
 726                 if (ret < 0)
 727                         uprobe_buffer_refcnt--;
 728         }
 729
 730         return ret;
 731 }
 732
 733 static void uprobe_buffer_disable(void)
 734 {
 735         int cpu;
 736
 737         BUG_ON(!mutex_is_locked(&event_mutex));
 738
 739         if (--uprobe_buffer_refcnt == 0) {
 740                 for_each_possible_cpu(cpu)
 741                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 742                                                              cpu)->buf);
 743
 744                 free_percpu(uprobe_cpu_buffer);
 745                 uprobe_cpu_buffer = NULL;
 746         }
 747 }
 748
 749 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 750 {
 751         struct uprobe_cpu_buffer *ucb;
 752         int cpu;
 753
 754         cpu = raw_smp_processor_id();
 755         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 756
 757         /*
 758          * Use per-cpu buffers for fastest access, but we might migrate
 759          * so the mutex makes sure we have sole access to it.
 760          */
 761         mutex_lock(&ucb->mutex);
 762
 763         return ucb;
 764 }
 765
 766 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 767 {
 768         mutex_unlock(&ucb->mutex);
 769 }
 770
 771 static void __uprobe_trace_func(struct trace_uprobe *tu,
 772                                 unsigned long func, struct pt_regs *regs,
 773                                 struct uprobe_cpu_buffer *ucb, int dsize,
 774                                 struct ftrace_event_file *ftrace_file)
 775 {
 776         struct uprobe_trace_entry_head *entry;
 777         struct ring_buffer_event *event;
 778         struct ring_buffer *buffer;
 779         void *data;
 780         int size, esize;
 781         struct ftrace_event_call *call = &tu->tp.call;
 782
 783         WARN_ON(call != ftrace_file->event_call);
 784
 785         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 786                 return;
 787
 788         if (ftrace_trigger_soft_disabled(ftrace_file))
 789                 return;
 790
 791         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 792         size = esize + tu->tp.size + dsize;
 793         event = trace_event_buffer_lock_reserve(&buffer, ftrace_file,
 794                                                 call->event.type, size, 0, 0);
 795         if (!event)
 796                 return;
 797
 798         entry = ring_buffer_event_data(event);
 799         if (is_ret_probe(tu)) {
 800                 entry->vaddr[0] = func;
 801                 entry->vaddr[1] = instruction_pointer(regs);
 802                 data = DATAOF_TRACE_ENTRY(entry, true);
 803         } else {
 804                 entry->vaddr[0] = instruction_pointer(regs);
 805                 data = DATAOF_TRACE_ENTRY(entry, false);
 806         }
 807
 808         memcpy(data, ucb->buf, tu->tp.size + dsize);
 809
 810         event_trigger_unlock_commit(ftrace_file, buffer, event, entry, 0, 0);
 811 }
 812
 813 /* uprobe handler */
 814 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 815                              struct uprobe_cpu_buffer *ucb, int dsize)
 816 {
 817         struct event_file_link *link;
 818
 819         if (is_ret_probe(tu))
 820                 return 0;
 821
 822         rcu_read_lock();
 823         list_for_each_entry_rcu(link, &tu->tp.files, list)
 824                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 825         rcu_read_unlock();
 826
 827         return 0;
 828 }
 829
 830 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 831                                  struct pt_regs *regs,
 832                                  struct uprobe_cpu_buffer *ucb, int dsize)
 833 {
 834         struct event_file_link *link;
 835
 836         rcu_read_lock();
 837         list_for_each_entry_rcu(link, &tu->tp.files, list)
 838                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 839         rcu_read_unlock();
 840 }
 841
 842 /* Event entry printers */
 843 static enum print_line_t
 844 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 845 {
 846         struct uprobe_trace_entry_head *entry;
 847         struct trace_seq *s = &iter->seq;
 848         struct trace_uprobe *tu;
 849         u8 *data;
 850         int i;
 851
 852         entry = (struct uprobe_trace_entry_head *)iter->ent;
 853         tu = container_of(event, struct trace_uprobe, tp.call.event);
 854
 855         if (is_ret_probe(tu)) {
 856                 if (!trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 857                                         ftrace_event_name(&tu->tp.call),
 858                                         entry->vaddr[1], entry->vaddr[0]))
 859                         goto partial;
 860                 data = DATAOF_TRACE_ENTRY(entry, true);
 861         } else {
 862                 if (!trace_seq_printf(s, "%s: (0x%lx)",
 863                                         ftrace_event_name(&tu->tp.call),
 864                                         entry->vaddr[0]))
 865                         goto partial;
 866                 data = DATAOF_TRACE_ENTRY(entry, false);
 867         }
 868
 869         for (i = 0; i < tu->tp.nr_args; i++) {
 870                 struct probe_arg *parg = &tu->tp.args[i];
 871
 872                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 873                         goto partial;
 874         }
 875
 876         if (trace_seq_puts(s, "\n"))
 877                 return TRACE_TYPE_HANDLED;
 878
 879 partial:
 880         return TRACE_TYPE_PARTIAL_LINE;
 881 }
 882
 883 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 884                                 enum uprobe_filter_ctx ctx,
 885                                 struct mm_struct *mm);
 886
 887 static int
 888 probe_event_enable(struct trace_uprobe *tu, struct ftrace_event_file *file,
 889                    filter_func_t filter)
 890 {
 891         bool enabled = trace_probe_is_enabled(&tu->tp);
 892         struct event_file_link *link = NULL;
 893         int ret;
 894
 895         if (file) {
 896                 if (tu->tp.flags & TP_FLAG_PROFILE)
 897                         return -EINTR;
 898
 899                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 900                 if (!link)
 901                         return -ENOMEM;
 902
 903                 link->file = file;
 904                 list_add_tail_rcu(&link->list, &tu->tp.files);
 905
 906                 tu->tp.flags |= TP_FLAG_TRACE;
 907         } else {
 908                 if (tu->tp.flags & TP_FLAG_TRACE)
 909                         return -EINTR;
 910
 911                 tu->tp.flags |= TP_FLAG_PROFILE;
 912         }
 913
 914         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 915
 916         if (enabled)
 917                 return 0;
 918
 919         ret = uprobe_buffer_enable();
 920         if (ret)
 921                 goto err_flags;
 922
 923         tu->consumer.filter = filter;
 924         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 925         if (ret)
 926                 goto err_buffer;
 927
 928         return 0;
 929
 930  err_buffer:
 931         uprobe_buffer_disable();
 932
 933  err_flags:
 934         if (file) {
 935                 list_del(&link->list);
 936                 kfree(link);
 937                 tu->tp.flags &= ~TP_FLAG_TRACE;
 938         } else {
 939                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 940         }
 941         return ret;
 942 }
 943
 944 static void
 945 probe_event_disable(struct trace_uprobe *tu, struct ftrace_event_file *file)
 946 {
 947         if (!trace_probe_is_enabled(&tu->tp))
 948                 return;
 949
 950         if (file) {
 951                 struct event_file_link *link;
 952
 953                 link = find_event_file_link(&tu->tp, file);
 954                 if (!link)
 955                         return;
 956
 957                 list_del_rcu(&link->list);
 958                 /* synchronize with u{,ret}probe_trace_func */
 959                 synchronize_sched();
 960                 kfree(link);
 961
 962                 if (!list_empty(&tu->tp.files))
 963                         return;
 964         }
 965
 966         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 967
 968         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 969         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 970
 971         uprobe_buffer_disable();
 972 }
 973
 974 static int uprobe_event_define_fields(struct ftrace_event_call *event_call)
 975 {
 976         int ret, i, size;
 977         struct uprobe_trace_entry_head field;
 978         struct trace_uprobe *tu = event_call->data;
 979
 980         if (is_ret_probe(tu)) {
 981                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 982                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 983                 size = SIZEOF_TRACE_ENTRY(true);
 984         } else {
 985                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 986                 size = SIZEOF_TRACE_ENTRY(false);
 987         }
 988         /* Set argument names as fields */
 989         for (i = 0; i < tu->tp.nr_args; i++) {
 990                 struct probe_arg *parg = &tu->tp.args[i];
 991
 992                 ret = trace_define_field(event_call, parg->type->fmttype,
 993                                          parg->name, size + parg->offset,
 994                                          parg->type->size, parg->type->is_signed,
 995                                          FILTER_OTHER);
 996
 997                 if (ret)
 998                         return ret;
 999         }
1000         return 0;
1001 }
1002
1003 #ifdef CONFIG_PERF_EVENTS
1004 static bool
1005 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1006 {
1007         struct perf_event *event;
1008
1009         if (filter->nr_systemwide)
1010                 return true;
1011
1012         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1013                 if (event->hw.tp_target->mm == mm)
1014                         return true;
1015         }
1016
1017         return false;
1018 }
1019
1020 static inline bool
1021 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1022 {
1023         return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
1024 }
1025
1026 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1027 {
1028         bool done;
1029
1030         write_lock(&tu->filter.rwlock);
1031         if (event->hw.tp_target) {
1032                 list_del(&event->hw.tp_list);
1033                 done = tu->filter.nr_systemwide ||
1034                         (event->hw.tp_target->flags & PF_EXITING) ||
1035                         uprobe_filter_event(tu, event);
1036         } else {
1037                 tu->filter.nr_systemwide--;
1038                 done = tu->filter.nr_systemwide;
1039         }
1040         write_unlock(&tu->filter.rwlock);
1041
1042         if (!done)
1043                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1044
1045         return 0;
1046 }
1047
1048 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1049 {
1050         bool done;
1051         int err;
1052
1053         write_lock(&tu->filter.rwlock);
1054         if (event->hw.tp_target) {
1055                 /*
1056                  * event->parent != NULL means copy_process(), we can avoid
1057                  * uprobe_apply(). current->mm must be probed and we can rely
1058                  * on dup_mmap() which preserves the already installed bp's.
1059                  *
1060                  * attr.enable_on_exec means that exec/mmap will install the
1061                  * breakpoints we need.
1062                  */
1063                 done = tu->filter.nr_systemwide ||
1064                         event->parent || event->attr.enable_on_exec ||
1065                         uprobe_filter_event(tu, event);
1066                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1067         } else {
1068                 done = tu->filter.nr_systemwide;
1069                 tu->filter.nr_systemwide++;
1070         }
1071         write_unlock(&tu->filter.rwlock);
1072
1073         err = 0;
1074         if (!done) {
1075                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1076                 if (err)
1077                         uprobe_perf_close(tu, event);
1078         }
1079         return err;
1080 }
1081
1082 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1083                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1084 {
1085         struct trace_uprobe *tu;
1086         int ret;
1087
1088         tu = container_of(uc, struct trace_uprobe, consumer);
1089         read_lock(&tu->filter.rwlock);
1090         ret = __uprobe_perf_filter(&tu->filter, mm);
1091         read_unlock(&tu->filter.rwlock);
1092
1093         return ret;
1094 }
1095
1096 static void __uprobe_perf_func(struct trace_uprobe *tu,
1097                                unsigned long func, struct pt_regs *regs,
1098                                struct uprobe_cpu_buffer *ucb, int dsize)
1099 {
1100         struct ftrace_event_call *call = &tu->tp.call;
1101         struct uprobe_trace_entry_head *entry;
1102         struct hlist_head *head;
1103         void *data;
1104         int size, esize;
1105         int rctx;
1106
1107         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1108
1109         size = esize + tu->tp.size + dsize;
1110         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1111         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1112                 return;
1113
1114         preempt_disable();
1115         head = this_cpu_ptr(call->perf_events);
1116         if (hlist_empty(head))
1117                 goto out;
1118
1119         entry = perf_trace_buf_prepare(size, call->event.type, regs, &rctx);
1120         if (!entry)
1121                 goto out;
1122
1123         if (is_ret_probe(tu)) {
1124                 entry->vaddr[0] = func;
1125                 entry->vaddr[1] = instruction_pointer(regs);
1126                 data = DATAOF_TRACE_ENTRY(entry, true);
1127         } else {
1128                 entry->vaddr[0] = instruction_pointer(regs);
1129                 data = DATAOF_TRACE_ENTRY(entry, false);
1130         }
1131
1132         memcpy(data, ucb->buf, tu->tp.size + dsize);
1133
1134         if (size - esize > tu->tp.size + dsize) {
1135                 int len = tu->tp.size + dsize;
1136
1137                 memset(data + len, 0, size - esize - len);
1138         }
1139
1140         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1141  out:
1142         preempt_enable();
1143 }
1144
1145 /* uprobe profile handler */
1146 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1147                             struct uprobe_cpu_buffer *ucb, int dsize)
1148 {
1149         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1150                 return UPROBE_HANDLER_REMOVE;
1151
1152         if (!is_ret_probe(tu))
1153                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1154         return 0;
1155 }
1156
1157 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1158                                 struct pt_regs *regs,
1159                                 struct uprobe_cpu_buffer *ucb, int dsize)
1160 {
1161         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1162 }
1163 #endif  /* CONFIG_PERF_EVENTS */
1164
1165 static int
1166 trace_uprobe_register(struct ftrace_event_call *event, enum trace_reg type,
1167                       void *data)
1168 {
1169         struct trace_uprobe *tu = event->data;
1170         struct ftrace_event_file *file = data;
1171
1172         switch (type) {
1173         case TRACE_REG_REGISTER:
1174                 return probe_event_enable(tu, file, NULL);
1175
1176         case TRACE_REG_UNREGISTER:
1177                 probe_event_disable(tu, file);
1178                 return 0;
1179
1180 #ifdef CONFIG_PERF_EVENTS
1181         case TRACE_REG_PERF_REGISTER:
1182                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1183
1184         case TRACE_REG_PERF_UNREGISTER:
1185                 probe_event_disable(tu, NULL);
1186                 return 0;
1187
1188         case TRACE_REG_PERF_OPEN:
1189                 return uprobe_perf_open(tu, data);
1190
1191         case TRACE_REG_PERF_CLOSE:
1192                 return uprobe_perf_close(tu, data);
1193
1194 #endif
1195         default:
1196                 return 0;
1197         }
1198         return 0;
1199 }
1200
1201 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1202 {
1203         struct trace_uprobe *tu;
1204         struct uprobe_dispatch_data udd;
1205         struct uprobe_cpu_buffer *ucb;
1206         int dsize, esize;
1207         int ret = 0;
1208
1209
1210         tu = container_of(con, struct trace_uprobe, consumer);
1211         tu->nhit++;
1212
1213         udd.tu = tu;
1214         udd.bp_addr = instruction_pointer(regs);
1215
1216         current->utask->vaddr = (unsigned long) &udd;
1217
1218         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1219                 return 0;
1220
1221         dsize = __get_data_size(&tu->tp, regs);
1222         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1223
1224         ucb = uprobe_buffer_get();
1225         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1226
1227         if (tu->tp.flags & TP_FLAG_TRACE)
1228                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1229
1230 #ifdef CONFIG_PERF_EVENTS
1231         if (tu->tp.flags & TP_FLAG_PROFILE)
1232                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1233 #endif
1234         uprobe_buffer_put(ucb);
1235         return ret;
1236 }
1237
1238 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1239                                 unsigned long func, struct pt_regs *regs)
1240 {
1241         struct trace_uprobe *tu;
1242         struct uprobe_dispatch_data udd;
1243         struct uprobe_cpu_buffer *ucb;
1244         int dsize, esize;
1245
1246         tu = container_of(con, struct trace_uprobe, consumer);
1247
1248         udd.tu = tu;
1249         udd.bp_addr = func;
1250
1251         current->utask->vaddr = (unsigned long) &udd;
1252
1253         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1254                 return 0;
1255
1256         dsize = __get_data_size(&tu->tp, regs);
1257         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1258
1259         ucb = uprobe_buffer_get();
1260         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1261
1262         if (tu->tp.flags & TP_FLAG_TRACE)
1263                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1264
1265 #ifdef CONFIG_PERF_EVENTS
1266         if (tu->tp.flags & TP_FLAG_PROFILE)
1267                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1268 #endif
1269         uprobe_buffer_put(ucb);
1270         return 0;
1271 }
1272
1273 static struct trace_event_functions uprobe_funcs = {
1274         .trace          = print_uprobe_event
1275 };
1276
1277 static int register_uprobe_event(struct trace_uprobe *tu)
1278 {
1279         struct ftrace_event_call *call = &tu->tp.call;
1280         int ret;
1281
1282         /* Initialize ftrace_event_call */
1283         INIT_LIST_HEAD(&call->class->fields);
1284         call->event.funcs = &uprobe_funcs;
1285         call->class->define_fields = uprobe_event_define_fields;
1286
1287         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1288                 return -ENOMEM;
1289
1290         ret = register_ftrace_event(&call->event);
1291         if (!ret) {
1292                 kfree(call->print_fmt);
1293                 return -ENODEV;
1294         }
1295         call->flags = 0;
1296         call->class->reg = trace_uprobe_register;
1297         call->data = tu;
1298         ret = trace_add_event_call(call);
1299
1300         if (ret) {
1301                 pr_info("Failed to register uprobe event: %s\n",
1302                         ftrace_event_name(call));
1303                 kfree(call->print_fmt);
1304                 unregister_ftrace_event(&call->event);
1305         }
1306
1307         return ret;
1308 }
1309
1310 static int unregister_uprobe_event(struct trace_uprobe *tu)
1311 {
1312         int ret;
1313
1314         /* tu->event is unregistered in trace_remove_event_call() */
1315         ret = trace_remove_event_call(&tu->tp.call);
1316         if (ret)
1317                 return ret;
1318         kfree(tu->tp.call.print_fmt);
1319         tu->tp.call.print_fmt = NULL;
1320         return 0;
1321 }
1322
1323 /* Make a trace interface for controling probe points */
1324 static __init int init_uprobe_trace(void)
1325 {
1326         struct dentry *d_tracer;
1327
1328         d_tracer = tracing_init_dentry();
1329         if (!d_tracer)
1330                 return 0;
1331
1332         trace_create_file("uprobe_events", 0644, d_tracer,
1333                                     NULL, &uprobe_events_ops);
1334         /* Profile interface */
1335         trace_create_file("uprobe_profile", 0444, d_tracer,
1336                                     NULL, &uprobe_profile_ops);
1337         return 0;
1338 }
1339
1340 fs_initcall(init_uprobe_trace);