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 #define pr_fmt(fmt)     "trace_kprobe: " fmt
  21
  22 #include <linux/module.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/uprobes.h>
  25 #include <linux/namei.h>
  26 #include <linux/string.h>
  27
  28 #include "trace_probe.h"
  29
  30 #define UPROBE_EVENT_SYSTEM     "uprobes"
  31
  32 struct uprobe_trace_entry_head {
  33         struct trace_entry      ent;
  34         unsigned long           vaddr[];
  35 };
  36
  37 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  38         (sizeof(struct uprobe_trace_entry_head) +       \
  39          sizeof(unsigned long) * (is_return ? 2 : 1))
  40
  41 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  42         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  43
  44 struct trace_uprobe_filter {
  45         rwlock_t                rwlock;
  46         int                     nr_systemwide;
  47         struct list_head        perf_events;
  48 };
  49
  50 /*
  51  * uprobe event core functions
  52  */
  53 struct trace_uprobe {
  54         struct list_head                list;
  55         struct trace_uprobe_filter      filter;
  56         struct uprobe_consumer          consumer;
  57         struct inode                    *inode;
  58         char                            *filename;
  59         unsigned long                   offset;
  60         unsigned long                   nhit;
  61         struct trace_probe              tp;
  62 };
  63
  64 #define SIZEOF_TRACE_UPROBE(n)                          \
  65         (offsetof(struct trace_uprobe, tp.args) +       \
  66         (sizeof(struct probe_arg) * (n)))
  67
  68 static int register_uprobe_event(struct trace_uprobe *tu);
  69 static int unregister_uprobe_event(struct trace_uprobe *tu);
  70
  71 static DEFINE_MUTEX(uprobe_lock);
  72 static LIST_HEAD(uprobe_list);
  73
  74 struct uprobe_dispatch_data {
  75         struct trace_uprobe     *tu;
  76         unsigned long           bp_addr;
  77 };
  78
  79 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  80 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  81                                 unsigned long func, struct pt_regs *regs);
  82
  83 #ifdef CONFIG_STACK_GROWSUP
  84 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  85 {
  86         return addr - (n * sizeof(long));
  87 }
  88 #else
  89 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  90 {
  91         return addr + (n * sizeof(long));
  92 }
  93 #endif
  94
  95 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  96 {
  97         unsigned long ret;
  98         unsigned long addr = user_stack_pointer(regs);
  99
 100         addr = adjust_stack_addr(addr, n);
 101
 102         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 103                 return 0;
 104
 105         return ret;
 106 }
 107
 108 /*
 109  * Uprobes-specific fetch functions
 110  */
 111 #define DEFINE_FETCH_stack(type)                                        \
 112 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 113                                          void *offset, void *dest)      \
 114 {                                                                       \
 115         *(type *)dest = (type)get_user_stack_nth(regs,                  \
 116                                               ((unsigned long)offset)); \
 117 }
 118 DEFINE_BASIC_FETCH_FUNCS(stack)
 119 /* No string on the stack entry */
 120 #define fetch_stack_string      NULL
 121 #define fetch_stack_string_size NULL
 122
 123 #define DEFINE_FETCH_memory(type)                                       \
 124 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 125                                           void *addr, void *dest)       \
 126 {                                                                       \
 127         type retval;                                                    \
 128         void __user *vaddr = (void __force __user *) addr;              \
 129                                                                         \
 130         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
 131                 *(type *)dest = 0;                                      \
 132         else                                                            \
 133                 *(type *) dest = retval;                                \
 134 }
 135 DEFINE_BASIC_FETCH_FUNCS(memory)
 136 /*
 137  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 138  * length and relative data location.
 139  */
 140 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 141                                             void *addr, void *dest)
 142 {
 143         long ret;
 144         u32 rloc = *(u32 *)dest;
 145         int maxlen  = get_rloc_len(rloc);
 146         u8 *dst = get_rloc_data(dest);
 147         void __user *src = (void __force __user *) addr;
 148
 149         if (!maxlen)
 150                 return;
 151
 152         ret = strncpy_from_user(dst, src, maxlen);
 153
 154         if (ret < 0) {  /* Failed to fetch string */
 155                 ((u8 *)get_rloc_data(dest))[0] = '\0';
 156                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 157         } else {
 158                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 159         }
 160 }
 161
 162 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 163                                                  void *addr, void *dest)
 164 {
 165         int len;
 166         void __user *vaddr = (void __force __user *) addr;
 167
 168         len = strnlen_user(vaddr, MAX_STRING_SIZE);
 169
 170         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 171                 *(u32 *)dest = 0;
 172         else
 173                 *(u32 *)dest = len;
 174 }
 175
 176 static unsigned long translate_user_vaddr(void *file_offset)
 177 {
 178         unsigned long base_addr;
 179         struct uprobe_dispatch_data *udd;
 180
 181         udd = (void *) current->utask->vaddr;
 182
 183         base_addr = udd->bp_addr - udd->tu->offset;
 184         return base_addr + (unsigned long)file_offset;
 185 }
 186
 187 #define DEFINE_FETCH_file_offset(type)                                  \
 188 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,    \
 189                                                void *offset, void *dest)\
 190 {                                                                       \
 191         void *vaddr = (void *)translate_user_vaddr(offset);             \
 192                                                                         \
 193         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
 194 }
 195 DEFINE_BASIC_FETCH_FUNCS(file_offset)
 196 DEFINE_FETCH_file_offset(string)
 197 DEFINE_FETCH_file_offset(string_size)
 198
 199 /* Fetch type information table */
 200 static const struct fetch_type uprobes_fetch_type_table[] = {
 201         /* Special types */
 202         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 203                                         sizeof(u32), 1, "__data_loc char[]"),
 204         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 205                                         string_size, sizeof(u32), 0, "u32"),
 206         /* Basic types */
 207         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 208         ASSIGN_FETCH_TYPE(u16, u16, 0),
 209         ASSIGN_FETCH_TYPE(u32, u32, 0),
 210         ASSIGN_FETCH_TYPE(u64, u64, 0),
 211         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 212         ASSIGN_FETCH_TYPE(s16, u16, 1),
 213         ASSIGN_FETCH_TYPE(s32, u32, 1),
 214         ASSIGN_FETCH_TYPE(s64, u64, 1),
 215         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 216         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 217         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 218         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 219
 220         ASSIGN_FETCH_TYPE_END
 221 };
 222
 223 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 224 {
 225         rwlock_init(&filter->rwlock);
 226         filter->nr_systemwide = 0;
 227         INIT_LIST_HEAD(&filter->perf_events);
 228 }
 229
 230 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 231 {
 232         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 233 }
 234
 235 static inline bool is_ret_probe(struct trace_uprobe *tu)
 236 {
 237         return tu->consumer.ret_handler != NULL;
 238 }
 239
 240 /*
 241  * Allocate new trace_uprobe and initialize it (including uprobes).
 242  */
 243 static struct trace_uprobe *
 244 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 245 {
 246         struct trace_uprobe *tu;
 247
 248         if (!event || !is_good_name(event))
 249                 return ERR_PTR(-EINVAL);
 250
 251         if (!group || !is_good_name(group))
 252                 return ERR_PTR(-EINVAL);
 253
 254         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 255         if (!tu)
 256                 return ERR_PTR(-ENOMEM);
 257
 258         tu->tp.call.class = &tu->tp.class;
 259         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 260         if (!tu->tp.call.name)
 261                 goto error;
 262
 263         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 264         if (!tu->tp.class.system)
 265                 goto error;
 266
 267         INIT_LIST_HEAD(&tu->list);
 268         INIT_LIST_HEAD(&tu->tp.files);
 269         tu->consumer.handler = uprobe_dispatcher;
 270         if (is_ret)
 271                 tu->consumer.ret_handler = uretprobe_dispatcher;
 272         init_trace_uprobe_filter(&tu->filter);
 273         return tu;
 274
 275 error:
 276         kfree(tu->tp.call.name);
 277         kfree(tu);
 278
 279         return ERR_PTR(-ENOMEM);
 280 }
 281
 282 static void free_trace_uprobe(struct trace_uprobe *tu)
 283 {
 284         int i;
 285
 286         for (i = 0; i < tu->tp.nr_args; i++)
 287                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 288
 289         iput(tu->inode);
 290         kfree(tu->tp.call.class->system);
 291         kfree(tu->tp.call.name);
 292         kfree(tu->filename);
 293         kfree(tu);
 294 }
 295
 296 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 297 {
 298         struct trace_uprobe *tu;
 299
 300         list_for_each_entry(tu, &uprobe_list, list)
 301                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 302                     strcmp(tu->tp.call.class->system, group) == 0)
 303                         return tu;
 304
 305         return NULL;
 306 }
 307
 308 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 309 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 310 {
 311         int ret;
 312
 313         ret = unregister_uprobe_event(tu);
 314         if (ret)
 315                 return ret;
 316
 317         list_del(&tu->list);
 318         free_trace_uprobe(tu);
 319         return 0;
 320 }
 321
 322 /* Register a trace_uprobe and probe_event */
 323 static int register_trace_uprobe(struct trace_uprobe *tu)
 324 {
 325         struct trace_uprobe *old_tu;
 326         int ret;
 327
 328         mutex_lock(&uprobe_lock);
 329
 330         /* register as an event */
 331         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
 332                         tu->tp.call.class->system);
 333         if (old_tu) {
 334                 /* delete old event */
 335                 ret = unregister_trace_uprobe(old_tu);
 336                 if (ret)
 337                         goto end;
 338         }
 339
 340         ret = register_uprobe_event(tu);
 341         if (ret) {
 342                 pr_warn("Failed to register probe event(%d)\n", ret);
 343                 goto end;
 344         }
 345
 346         list_add_tail(&tu->list, &uprobe_list);
 347
 348 end:
 349         mutex_unlock(&uprobe_lock);
 350
 351         return ret;
 352 }
 353
 354 /*
 355  * Argument syntax:
 356  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 357  *
 358  *  - Remove uprobe: -:[GRP/]EVENT
 359  */
 360 static int create_trace_uprobe(int argc, char **argv)
 361 {
 362         struct trace_uprobe *tu;
 363         struct inode *inode;
 364         char *arg, *event, *group, *filename;
 365         char buf[MAX_EVENT_NAME_LEN];
 366         struct path path;
 367         unsigned long offset;
 368         bool is_delete, is_return;
 369         int i, ret;
 370
 371         inode = NULL;
 372         ret = 0;
 373         is_delete = false;
 374         is_return = false;
 375         event = NULL;
 376         group = NULL;
 377
 378         /* argc must be >= 1 */
 379         if (argv[0][0] == '-')
 380                 is_delete = true;
 381         else if (argv[0][0] == 'r')
 382                 is_return = true;
 383         else if (argv[0][0] != 'p') {
 384                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 385                 return -EINVAL;
 386         }
 387
 388         if (argv[0][1] == ':') {
 389                 event = &argv[0][2];
 390                 arg = strchr(event, '/');
 391
 392                 if (arg) {
 393                         group = event;
 394                         event = arg + 1;
 395                         event[-1] = '\0';
 396
 397                         if (strlen(group) == 0) {
 398                                 pr_info("Group name is not specified\n");
 399                                 return -EINVAL;
 400                         }
 401                 }
 402                 if (strlen(event) == 0) {
 403                         pr_info("Event name is not specified\n");
 404                         return -EINVAL;
 405                 }
 406         }
 407         if (!group)
 408                 group = UPROBE_EVENT_SYSTEM;
 409
 410         if (is_delete) {
 411                 int ret;
 412
 413                 if (!event) {
 414                         pr_info("Delete command needs an event name.\n");
 415                         return -EINVAL;
 416                 }
 417                 mutex_lock(&uprobe_lock);
 418                 tu = find_probe_event(event, group);
 419
 420                 if (!tu) {
 421                         mutex_unlock(&uprobe_lock);
 422                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 423                         return -ENOENT;
 424                 }
 425                 /* delete an event */
 426                 ret = unregister_trace_uprobe(tu);
 427                 mutex_unlock(&uprobe_lock);
 428                 return ret;
 429         }
 430
 431         if (argc < 2) {
 432                 pr_info("Probe point is not specified.\n");
 433                 return -EINVAL;
 434         }
 435         /* Find the last occurrence, in case the path contains ':' too. */
 436         arg = strrchr(argv[1], ':');
 437         if (!arg) {
 438                 ret = -EINVAL;
 439                 goto fail_address_parse;
 440         }
 441
 442         *arg++ = '\0';
 443         filename = argv[1];
 444         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 445         if (ret)
 446                 goto fail_address_parse;
 447
 448         inode = igrab(d_inode(path.dentry));
 449         path_put(&path);
 450
 451         if (!inode || !S_ISREG(inode->i_mode)) {
 452                 ret = -EINVAL;
 453                 goto fail_address_parse;
 454         }
 455
 456         ret = kstrtoul(arg, 0, &offset);
 457         if (ret)
 458                 goto fail_address_parse;
 459
 460         argc -= 2;
 461         argv += 2;
 462
 463         /* setup a probe */
 464         if (!event) {
 465                 char *tail;
 466                 char *ptr;
 467
 468                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 469                 if (!tail) {
 470                         ret = -ENOMEM;
 471                         goto fail_address_parse;
 472                 }
 473
 474                 ptr = strpbrk(tail, ".-_");
 475                 if (ptr)
 476                         *ptr = '\0';
 477
 478                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 479                 event = buf;
 480                 kfree(tail);
 481         }
 482
 483         tu = alloc_trace_uprobe(group, event, argc, is_return);
 484         if (IS_ERR(tu)) {
 485                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 486                 ret = PTR_ERR(tu);
 487                 goto fail_address_parse;
 488         }
 489         tu->offset = offset;
 490         tu->inode = inode;
 491         tu->filename = kstrdup(filename, GFP_KERNEL);
 492
 493         if (!tu->filename) {
 494                 pr_info("Failed to allocate filename.\n");
 495                 ret = -ENOMEM;
 496                 goto error;
 497         }
 498
 499         /* parse arguments */
 500         ret = 0;
 501         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 502                 struct probe_arg *parg = &tu->tp.args[i];
 503
 504                 /* Increment count for freeing args in error case */
 505                 tu->tp.nr_args++;
 506
 507                 /* Parse argument name */
 508                 arg = strchr(argv[i], '=');
 509                 if (arg) {
 510                         *arg++ = '\0';
 511                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 512                 } else {
 513                         arg = argv[i];
 514                         /* If argument name is omitted, set "argN" */
 515                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 516                         parg->name = kstrdup(buf, GFP_KERNEL);
 517                 }
 518
 519                 if (!parg->name) {
 520                         pr_info("Failed to allocate argument[%d] name.\n", i);
 521                         ret = -ENOMEM;
 522                         goto error;
 523                 }
 524
 525                 if (!is_good_name(parg->name)) {
 526                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 527                         ret = -EINVAL;
 528                         goto error;
 529                 }
 530
 531                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 532                         pr_info("Argument[%d] name '%s' conflicts with "
 533                                 "another field.\n", i, argv[i]);
 534                         ret = -EINVAL;
 535                         goto error;
 536                 }
 537
 538                 /* Parse fetch argument */
 539                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 540                                                  is_return, false,
 541                                                  uprobes_fetch_type_table);
 542                 if (ret) {
 543                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 544                         goto error;
 545                 }
 546         }
 547
 548         ret = register_trace_uprobe(tu);
 549         if (ret)
 550                 goto error;
 551         return 0;
 552
 553 error:
 554         free_trace_uprobe(tu);
 555         return ret;
 556
 557 fail_address_parse:
 558         iput(inode);
 559
 560         pr_info("Failed to parse address or file.\n");
 561
 562         return ret;
 563 }
 564
 565 static int cleanup_all_probes(void)
 566 {
 567         struct trace_uprobe *tu;
 568         int ret = 0;
 569
 570         mutex_lock(&uprobe_lock);
 571         while (!list_empty(&uprobe_list)) {
 572                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 573                 ret = unregister_trace_uprobe(tu);
 574                 if (ret)
 575                         break;
 576         }
 577         mutex_unlock(&uprobe_lock);
 578         return ret;
 579 }
 580
 581 /* Probes listing interfaces */
 582 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 583 {
 584         mutex_lock(&uprobe_lock);
 585         return seq_list_start(&uprobe_list, *pos);
 586 }
 587
 588 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 589 {
 590         return seq_list_next(v, &uprobe_list, pos);
 591 }
 592
 593 static void probes_seq_stop(struct seq_file *m, void *v)
 594 {
 595         mutex_unlock(&uprobe_lock);
 596 }
 597
 598 static int probes_seq_show(struct seq_file *m, void *v)
 599 {
 600         struct trace_uprobe *tu = v;
 601         char c = is_ret_probe(tu) ? 'r' : 'p';
 602         int i;
 603
 604         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 605                         trace_event_name(&tu->tp.call));
 606         seq_printf(m, " %s:", tu->filename);
 607
 608         /* Don't print "0x  (null)" when offset is 0 */
 609         if (tu->offset) {
 610                 seq_printf(m, "0x%p", (void *)tu->offset);
 611         } else {
 612                 switch (sizeof(void *)) {
 613                 case 4:
 614                         seq_printf(m, "0x00000000");
 615                         break;
 616                 case 8:
 617                 default:
 618                         seq_printf(m, "0x0000000000000000");
 619                         break;
 620                 }
 621         }
 622
 623         for (i = 0; i < tu->tp.nr_args; i++)
 624                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 625
 626         seq_putc(m, '\n');
 627         return 0;
 628 }
 629
 630 static const struct seq_operations probes_seq_op = {
 631         .start  = probes_seq_start,
 632         .next   = probes_seq_next,
 633         .stop   = probes_seq_stop,
 634         .show   = probes_seq_show
 635 };
 636
 637 static int probes_open(struct inode *inode, struct file *file)
 638 {
 639         int ret;
 640
 641         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 642                 ret = cleanup_all_probes();
 643                 if (ret)
 644                         return ret;
 645         }
 646
 647         return seq_open(file, &probes_seq_op);
 648 }
 649
 650 static ssize_t probes_write(struct file *file, const char __user *buffer,
 651                             size_t count, loff_t *ppos)
 652 {
 653         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 654 }
 655
 656 static const struct file_operations uprobe_events_ops = {
 657         .owner          = THIS_MODULE,
 658         .open           = probes_open,
 659         .read           = seq_read,
 660         .llseek         = seq_lseek,
 661         .release        = seq_release,
 662         .write          = probes_write,
 663 };
 664
 665 /* Probes profiling interfaces */
 666 static int probes_profile_seq_show(struct seq_file *m, void *v)
 667 {
 668         struct trace_uprobe *tu = v;
 669
 670         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 671                         trace_event_name(&tu->tp.call), tu->nhit);
 672         return 0;
 673 }
 674
 675 static const struct seq_operations profile_seq_op = {
 676         .start  = probes_seq_start,
 677         .next   = probes_seq_next,
 678         .stop   = probes_seq_stop,
 679         .show   = probes_profile_seq_show
 680 };
 681
 682 static int profile_open(struct inode *inode, struct file *file)
 683 {
 684         return seq_open(file, &profile_seq_op);
 685 }
 686
 687 static const struct file_operations uprobe_profile_ops = {
 688         .owner          = THIS_MODULE,
 689         .open           = profile_open,
 690         .read           = seq_read,
 691         .llseek         = seq_lseek,
 692         .release        = seq_release,
 693 };
 694
 695 struct uprobe_cpu_buffer {
 696         struct mutex mutex;
 697         void *buf;
 698 };
 699 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 700 static int uprobe_buffer_refcnt;
 701
 702 static int uprobe_buffer_init(void)
 703 {
 704         int cpu, err_cpu;
 705
 706         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 707         if (uprobe_cpu_buffer == NULL)
 708                 return -ENOMEM;
 709
 710         for_each_possible_cpu(cpu) {
 711                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 712                                                   GFP_KERNEL, 0);
 713                 if (p == NULL) {
 714                         err_cpu = cpu;
 715                         goto err;
 716                 }
 717                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 718                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 719         }
 720
 721         return 0;
 722
 723 err:
 724         for_each_possible_cpu(cpu) {
 725                 if (cpu == err_cpu)
 726                         break;
 727                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 728         }
 729
 730         free_percpu(uprobe_cpu_buffer);
 731         return -ENOMEM;
 732 }
 733
 734 static int uprobe_buffer_enable(void)
 735 {
 736         int ret = 0;
 737
 738         BUG_ON(!mutex_is_locked(&event_mutex));
 739
 740         if (uprobe_buffer_refcnt++ == 0) {
 741                 ret = uprobe_buffer_init();
 742                 if (ret < 0)
 743                         uprobe_buffer_refcnt--;
 744         }
 745
 746         return ret;
 747 }
 748
 749 static void uprobe_buffer_disable(void)
 750 {
 751         int cpu;
 752
 753         BUG_ON(!mutex_is_locked(&event_mutex));
 754
 755         if (--uprobe_buffer_refcnt == 0) {
 756                 for_each_possible_cpu(cpu)
 757                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 758                                                              cpu)->buf);
 759
 760                 free_percpu(uprobe_cpu_buffer);
 761                 uprobe_cpu_buffer = NULL;
 762         }
 763 }
 764
 765 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 766 {
 767         struct uprobe_cpu_buffer *ucb;
 768         int cpu;
 769
 770         cpu = raw_smp_processor_id();
 771         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 772
 773         /*
 774          * Use per-cpu buffers for fastest access, but we might migrate
 775          * so the mutex makes sure we have sole access to it.
 776          */
 777         mutex_lock(&ucb->mutex);
 778
 779         return ucb;
 780 }
 781
 782 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 783 {
 784         mutex_unlock(&ucb->mutex);
 785 }
 786
 787 static void __uprobe_trace_func(struct trace_uprobe *tu,
 788                                 unsigned long func, struct pt_regs *regs,
 789                                 struct uprobe_cpu_buffer *ucb, int dsize,
 790                                 struct trace_event_file *trace_file)
 791 {
 792         struct uprobe_trace_entry_head *entry;
 793         struct ring_buffer_event *event;
 794         struct ring_buffer *buffer;
 795         void *data;
 796         int size, esize;
 797         struct trace_event_call *call = &tu->tp.call;
 798
 799         WARN_ON(call != trace_file->event_call);
 800
 801         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 802                 return;
 803
 804         if (trace_trigger_soft_disabled(trace_file))
 805                 return;
 806
 807         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 808         size = esize + tu->tp.size + dsize;
 809         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 810                                                 call->event.type, size, 0, 0);
 811         if (!event)
 812                 return;
 813
 814         entry = ring_buffer_event_data(event);
 815         if (is_ret_probe(tu)) {
 816                 entry->vaddr[0] = func;
 817                 entry->vaddr[1] = instruction_pointer(regs);
 818                 data = DATAOF_TRACE_ENTRY(entry, true);
 819         } else {
 820                 entry->vaddr[0] = instruction_pointer(regs);
 821                 data = DATAOF_TRACE_ENTRY(entry, false);
 822         }
 823
 824         memcpy(data, ucb->buf, tu->tp.size + dsize);
 825
 826         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 827 }
 828
 829 /* uprobe handler */
 830 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 831                              struct uprobe_cpu_buffer *ucb, int dsize)
 832 {
 833         struct event_file_link *link;
 834
 835         if (is_ret_probe(tu))
 836                 return 0;
 837
 838         rcu_read_lock();
 839         list_for_each_entry_rcu(link, &tu->tp.files, list)
 840                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 841         rcu_read_unlock();
 842
 843         return 0;
 844 }
 845
 846 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 847                                  struct pt_regs *regs,
 848                                  struct uprobe_cpu_buffer *ucb, int dsize)
 849 {
 850         struct event_file_link *link;
 851
 852         rcu_read_lock();
 853         list_for_each_entry_rcu(link, &tu->tp.files, list)
 854                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 855         rcu_read_unlock();
 856 }
 857
 858 /* Event entry printers */
 859 static enum print_line_t
 860 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 861 {
 862         struct uprobe_trace_entry_head *entry;
 863         struct trace_seq *s = &iter->seq;
 864         struct trace_uprobe *tu;
 865         u8 *data;
 866         int i;
 867
 868         entry = (struct uprobe_trace_entry_head *)iter->ent;
 869         tu = container_of(event, struct trace_uprobe, tp.call.event);
 870
 871         if (is_ret_probe(tu)) {
 872                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 873                                  trace_event_name(&tu->tp.call),
 874                                  entry->vaddr[1], entry->vaddr[0]);
 875                 data = DATAOF_TRACE_ENTRY(entry, true);
 876         } else {
 877                 trace_seq_printf(s, "%s: (0x%lx)",
 878                                  trace_event_name(&tu->tp.call),
 879                                  entry->vaddr[0]);
 880                 data = DATAOF_TRACE_ENTRY(entry, false);
 881         }
 882
 883         for (i = 0; i < tu->tp.nr_args; i++) {
 884                 struct probe_arg *parg = &tu->tp.args[i];
 885
 886                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 887                         goto out;
 888         }
 889
 890         trace_seq_putc(s, '\n');
 891
 892  out:
 893         return trace_handle_return(s);
 894 }
 895
 896 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 897                                 enum uprobe_filter_ctx ctx,
 898                                 struct mm_struct *mm);
 899
 900 static int
 901 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
 902                    filter_func_t filter)
 903 {
 904         bool enabled = trace_probe_is_enabled(&tu->tp);
 905         struct event_file_link *link = NULL;
 906         int ret;
 907
 908         if (file) {
 909                 if (tu->tp.flags & TP_FLAG_PROFILE)
 910                         return -EINTR;
 911
 912                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 913                 if (!link)
 914                         return -ENOMEM;
 915
 916                 link->file = file;
 917                 list_add_tail_rcu(&link->list, &tu->tp.files);
 918
 919                 tu->tp.flags |= TP_FLAG_TRACE;
 920         } else {
 921                 if (tu->tp.flags & TP_FLAG_TRACE)
 922                         return -EINTR;
 923
 924                 tu->tp.flags |= TP_FLAG_PROFILE;
 925         }
 926
 927         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 928
 929         if (enabled)
 930                 return 0;
 931
 932         ret = uprobe_buffer_enable();
 933         if (ret)
 934                 goto err_flags;
 935
 936         tu->consumer.filter = filter;
 937         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 938         if (ret)
 939                 goto err_buffer;
 940
 941         return 0;
 942
 943  err_buffer:
 944         uprobe_buffer_disable();
 945
 946  err_flags:
 947         if (file) {
 948                 list_del(&link->list);
 949                 kfree(link);
 950                 tu->tp.flags &= ~TP_FLAG_TRACE;
 951         } else {
 952                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 953         }
 954         return ret;
 955 }
 956
 957 static void
 958 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 959 {
 960         if (!trace_probe_is_enabled(&tu->tp))
 961                 return;
 962
 963         if (file) {
 964                 struct event_file_link *link;
 965
 966                 link = find_event_file_link(&tu->tp, file);
 967                 if (!link)
 968                         return;
 969
 970                 list_del_rcu(&link->list);
 971                 /* synchronize with u{,ret}probe_trace_func */
 972                 synchronize_sched();
 973                 kfree(link);
 974
 975                 if (!list_empty(&tu->tp.files))
 976                         return;
 977         }
 978
 979         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 980
 981         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 982         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 983
 984         uprobe_buffer_disable();
 985 }
 986
 987 static int uprobe_event_define_fields(struct trace_event_call *event_call)
 988 {
 989         int ret, i, size;
 990         struct uprobe_trace_entry_head field;
 991         struct trace_uprobe *tu = event_call->data;
 992
 993         if (is_ret_probe(tu)) {
 994                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 995                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 996                 size = SIZEOF_TRACE_ENTRY(true);
 997         } else {
 998                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 999                 size = SIZEOF_TRACE_ENTRY(false);
1000         }
1001         /* Set argument names as fields */
1002         for (i = 0; i < tu->tp.nr_args; i++) {
1003                 struct probe_arg *parg = &tu->tp.args[i];
1004
1005                 ret = trace_define_field(event_call, parg->type->fmttype,
1006                                          parg->name, size + parg->offset,
1007                                          parg->type->size, parg->type->is_signed,
1008                                          FILTER_OTHER);
1009
1010                 if (ret)
1011                         return ret;
1012         }
1013         return 0;
1014 }
1015
1016 #ifdef CONFIG_PERF_EVENTS
1017 static bool
1018 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1019 {
1020         struct perf_event *event;
1021
1022         if (filter->nr_systemwide)
1023                 return true;
1024
1025         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1026                 if (event->hw.target->mm == mm)
1027                         return true;
1028         }
1029
1030         return false;
1031 }
1032
1033 static inline bool
1034 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1035 {
1036         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1037 }
1038
1039 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1040 {
1041         bool done;
1042
1043         write_lock(&tu->filter.rwlock);
1044         if (event->hw.target) {
1045                 list_del(&event->hw.tp_list);
1046                 done = tu->filter.nr_systemwide ||
1047                         (event->hw.target->flags & PF_EXITING) ||
1048                         uprobe_filter_event(tu, event);
1049         } else {
1050                 tu->filter.nr_systemwide--;
1051                 done = tu->filter.nr_systemwide;
1052         }
1053         write_unlock(&tu->filter.rwlock);
1054
1055         if (!done)
1056                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1057
1058         return 0;
1059 }
1060
1061 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1062 {
1063         bool done;
1064         int err;
1065
1066         write_lock(&tu->filter.rwlock);
1067         if (event->hw.target) {
1068                 /*
1069                  * event->parent != NULL means copy_process(), we can avoid
1070                  * uprobe_apply(). current->mm must be probed and we can rely
1071                  * on dup_mmap() which preserves the already installed bp's.
1072                  *
1073                  * attr.enable_on_exec means that exec/mmap will install the
1074                  * breakpoints we need.
1075                  */
1076                 done = tu->filter.nr_systemwide ||
1077                         event->parent || event->attr.enable_on_exec ||
1078                         uprobe_filter_event(tu, event);
1079                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1080         } else {
1081                 done = tu->filter.nr_systemwide;
1082                 tu->filter.nr_systemwide++;
1083         }
1084         write_unlock(&tu->filter.rwlock);
1085
1086         err = 0;
1087         if (!done) {
1088                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1089                 if (err)
1090                         uprobe_perf_close(tu, event);
1091         }
1092         return err;
1093 }
1094
1095 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1096                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1097 {
1098         struct trace_uprobe *tu;
1099         int ret;
1100
1101         tu = container_of(uc, struct trace_uprobe, consumer);
1102         read_lock(&tu->filter.rwlock);
1103         ret = __uprobe_perf_filter(&tu->filter, mm);
1104         read_unlock(&tu->filter.rwlock);
1105
1106         return ret;
1107 }
1108
1109 static void __uprobe_perf_func(struct trace_uprobe *tu,
1110                                unsigned long func, struct pt_regs *regs,
1111                                struct uprobe_cpu_buffer *ucb, int dsize)
1112 {
1113         struct trace_event_call *call = &tu->tp.call;
1114         struct uprobe_trace_entry_head *entry;
1115         struct bpf_prog *prog = call->prog;
1116         struct hlist_head *head;
1117         void *data;
1118         int size, esize;
1119         int rctx;
1120
1121         if (prog && !trace_call_bpf(prog, regs))
1122                 return;
1123
1124         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1125
1126         size = esize + tu->tp.size + dsize;
1127         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1128         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1129                 return;
1130
1131         preempt_disable();
1132         head = this_cpu_ptr(call->perf_events);
1133         if (hlist_empty(head))
1134                 goto out;
1135
1136         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1137         if (!entry)
1138                 goto out;
1139
1140         if (is_ret_probe(tu)) {
1141                 entry->vaddr[0] = func;
1142                 entry->vaddr[1] = instruction_pointer(regs);
1143                 data = DATAOF_TRACE_ENTRY(entry, true);
1144         } else {
1145                 entry->vaddr[0] = instruction_pointer(regs);
1146                 data = DATAOF_TRACE_ENTRY(entry, false);
1147         }
1148
1149         memcpy(data, ucb->buf, tu->tp.size + dsize);
1150
1151         if (size - esize > tu->tp.size + dsize) {
1152                 int len = tu->tp.size + dsize;
1153
1154                 memset(data + len, 0, size - esize - len);
1155         }
1156
1157         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1158                               head, NULL);
1159  out:
1160         preempt_enable();
1161 }
1162
1163 /* uprobe profile handler */
1164 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1165                             struct uprobe_cpu_buffer *ucb, int dsize)
1166 {
1167         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1168                 return UPROBE_HANDLER_REMOVE;
1169
1170         if (!is_ret_probe(tu))
1171                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1172         return 0;
1173 }
1174
1175 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1176                                 struct pt_regs *regs,
1177                                 struct uprobe_cpu_buffer *ucb, int dsize)
1178 {
1179         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1180 }
1181 #endif  /* CONFIG_PERF_EVENTS */
1182
1183 static int
1184 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1185                       void *data)
1186 {
1187         struct trace_uprobe *tu = event->data;
1188         struct trace_event_file *file = data;
1189
1190         switch (type) {
1191         case TRACE_REG_REGISTER:
1192                 return probe_event_enable(tu, file, NULL);
1193
1194         case TRACE_REG_UNREGISTER:
1195                 probe_event_disable(tu, file);
1196                 return 0;
1197
1198 #ifdef CONFIG_PERF_EVENTS
1199         case TRACE_REG_PERF_REGISTER:
1200                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1201
1202         case TRACE_REG_PERF_UNREGISTER:
1203                 probe_event_disable(tu, NULL);
1204                 return 0;
1205
1206         case TRACE_REG_PERF_OPEN:
1207                 return uprobe_perf_open(tu, data);
1208
1209         case TRACE_REG_PERF_CLOSE:
1210                 return uprobe_perf_close(tu, data);
1211
1212 #endif
1213         default:
1214                 return 0;
1215         }
1216         return 0;
1217 }
1218
1219 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1220 {
1221         struct trace_uprobe *tu;
1222         struct uprobe_dispatch_data udd;
1223         struct uprobe_cpu_buffer *ucb;
1224         int dsize, esize;
1225         int ret = 0;
1226
1227
1228         tu = container_of(con, struct trace_uprobe, consumer);
1229         tu->nhit++;
1230
1231         udd.tu = tu;
1232         udd.bp_addr = instruction_pointer(regs);
1233
1234         current->utask->vaddr = (unsigned long) &udd;
1235
1236         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1237                 return 0;
1238
1239         dsize = __get_data_size(&tu->tp, regs);
1240         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1241
1242         ucb = uprobe_buffer_get();
1243         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1244
1245         if (tu->tp.flags & TP_FLAG_TRACE)
1246                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1247
1248 #ifdef CONFIG_PERF_EVENTS
1249         if (tu->tp.flags & TP_FLAG_PROFILE)
1250                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1251 #endif
1252         uprobe_buffer_put(ucb);
1253         return ret;
1254 }
1255
1256 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1257                                 unsigned long func, struct pt_regs *regs)
1258 {
1259         struct trace_uprobe *tu;
1260         struct uprobe_dispatch_data udd;
1261         struct uprobe_cpu_buffer *ucb;
1262         int dsize, esize;
1263
1264         tu = container_of(con, struct trace_uprobe, consumer);
1265
1266         udd.tu = tu;
1267         udd.bp_addr = func;
1268
1269         current->utask->vaddr = (unsigned long) &udd;
1270
1271         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1272                 return 0;
1273
1274         dsize = __get_data_size(&tu->tp, regs);
1275         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1276
1277         ucb = uprobe_buffer_get();
1278         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1279
1280         if (tu->tp.flags & TP_FLAG_TRACE)
1281                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1282
1283 #ifdef CONFIG_PERF_EVENTS
1284         if (tu->tp.flags & TP_FLAG_PROFILE)
1285                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1286 #endif
1287         uprobe_buffer_put(ucb);
1288         return 0;
1289 }
1290
1291 static struct trace_event_functions uprobe_funcs = {
1292         .trace          = print_uprobe_event
1293 };
1294
1295 static int register_uprobe_event(struct trace_uprobe *tu)
1296 {
1297         struct trace_event_call *call = &tu->tp.call;
1298         int ret;
1299
1300         /* Initialize trace_event_call */
1301         INIT_LIST_HEAD(&call->class->fields);
1302         call->event.funcs = &uprobe_funcs;
1303         call->class->define_fields = uprobe_event_define_fields;
1304
1305         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1306                 return -ENOMEM;
1307
1308         ret = register_trace_event(&call->event);
1309         if (!ret) {
1310                 kfree(call->print_fmt);
1311                 return -ENODEV;
1312         }
1313
1314         call->flags = TRACE_EVENT_FL_UPROBE;
1315         call->class->reg = trace_uprobe_register;
1316         call->data = tu;
1317         ret = trace_add_event_call(call);
1318
1319         if (ret) {
1320                 pr_info("Failed to register uprobe event: %s\n",
1321                         trace_event_name(call));
1322                 kfree(call->print_fmt);
1323                 unregister_trace_event(&call->event);
1324         }
1325
1326         return ret;
1327 }
1328
1329 static int unregister_uprobe_event(struct trace_uprobe *tu)
1330 {
1331         int ret;
1332
1333         /* tu->event is unregistered in trace_remove_event_call() */
1334         ret = trace_remove_event_call(&tu->tp.call);
1335         if (ret)
1336                 return ret;
1337         kfree(tu->tp.call.print_fmt);
1338         tu->tp.call.print_fmt = NULL;
1339         return 0;
1340 }
1341
1342 /* Make a trace interface for controling probe points */
1343 static __init int init_uprobe_trace(void)
1344 {
1345         struct dentry *d_tracer;
1346
1347         d_tracer = tracing_init_dentry();
1348         if (IS_ERR(d_tracer))
1349                 return 0;
1350
1351         trace_create_file("uprobe_events", 0644, d_tracer,
1352                                     NULL, &uprobe_events_ops);
1353         /* Profile interface */
1354         trace_create_file("uprobe_profile", 0444, d_tracer,
1355                                     NULL, &uprobe_profile_ops);
1356         return 0;
1357 }
1358
1359 fs_initcall(init_uprobe_trace);