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         if (ret == maxlen)
 153                 dst[ret - 1] = '\0';
 154         else if (ret >= 0)
 155                 /*
 156                  * Include the terminating null byte. In this case it
 157                  * was copied by strncpy_from_user but not accounted
 158                  * for in ret.
 159                  */
 160                 ret++;
 161
 162         if (ret < 0) {  /* Failed to fetch string */
 163                 ((u8 *)get_rloc_data(dest))[0] = '\0';
 164                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 165         } else {
 166                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 167         }
 168 }
 169
 170 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 171                                                  void *addr, void *dest)
 172 {
 173         int len;
 174         void __user *vaddr = (void __force __user *) addr;
 175
 176         len = strnlen_user(vaddr, MAX_STRING_SIZE);
 177
 178         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 179                 *(u32 *)dest = 0;
 180         else
 181                 *(u32 *)dest = len;
 182 }
 183
 184 static unsigned long translate_user_vaddr(void *file_offset)
 185 {
 186         unsigned long base_addr;
 187         struct uprobe_dispatch_data *udd;
 188
 189         udd = (void *) current->utask->vaddr;
 190
 191         base_addr = udd->bp_addr - udd->tu->offset;
 192         return base_addr + (unsigned long)file_offset;
 193 }
 194
 195 #define DEFINE_FETCH_file_offset(type)                                  \
 196 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,    \
 197                                                void *offset, void *dest)\
 198 {                                                                       \
 199         void *vaddr = (void *)translate_user_vaddr(offset);             \
 200                                                                         \
 201         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
 202 }
 203 DEFINE_BASIC_FETCH_FUNCS(file_offset)
 204 DEFINE_FETCH_file_offset(string)
 205 DEFINE_FETCH_file_offset(string_size)
 206
 207 /* Fetch type information table */
 208 static const struct fetch_type uprobes_fetch_type_table[] = {
 209         /* Special types */
 210         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 211                                         sizeof(u32), 1, "__data_loc char[]"),
 212         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 213                                         string_size, sizeof(u32), 0, "u32"),
 214         /* Basic types */
 215         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 216         ASSIGN_FETCH_TYPE(u16, u16, 0),
 217         ASSIGN_FETCH_TYPE(u32, u32, 0),
 218         ASSIGN_FETCH_TYPE(u64, u64, 0),
 219         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 220         ASSIGN_FETCH_TYPE(s16, u16, 1),
 221         ASSIGN_FETCH_TYPE(s32, u32, 1),
 222         ASSIGN_FETCH_TYPE(s64, u64, 1),
 223
 224         ASSIGN_FETCH_TYPE_END
 225 };
 226
 227 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 228 {
 229         rwlock_init(&filter->rwlock);
 230         filter->nr_systemwide = 0;
 231         INIT_LIST_HEAD(&filter->perf_events);
 232 }
 233
 234 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 235 {
 236         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 237 }
 238
 239 static inline bool is_ret_probe(struct trace_uprobe *tu)
 240 {
 241         return tu->consumer.ret_handler != NULL;
 242 }
 243
 244 /*
 245  * Allocate new trace_uprobe and initialize it (including uprobes).
 246  */
 247 static struct trace_uprobe *
 248 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 249 {
 250         struct trace_uprobe *tu;
 251
 252         if (!event || !is_good_name(event))
 253                 return ERR_PTR(-EINVAL);
 254
 255         if (!group || !is_good_name(group))
 256                 return ERR_PTR(-EINVAL);
 257
 258         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 259         if (!tu)
 260                 return ERR_PTR(-ENOMEM);
 261
 262         tu->tp.call.class = &tu->tp.class;
 263         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 264         if (!tu->tp.call.name)
 265                 goto error;
 266
 267         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 268         if (!tu->tp.class.system)
 269                 goto error;
 270
 271         INIT_LIST_HEAD(&tu->list);
 272         INIT_LIST_HEAD(&tu->tp.files);
 273         tu->consumer.handler = uprobe_dispatcher;
 274         if (is_ret)
 275                 tu->consumer.ret_handler = uretprobe_dispatcher;
 276         init_trace_uprobe_filter(&tu->filter);
 277         return tu;
 278
 279 error:
 280         kfree(tu->tp.call.name);
 281         kfree(tu);
 282
 283         return ERR_PTR(-ENOMEM);
 284 }
 285
 286 static void free_trace_uprobe(struct trace_uprobe *tu)
 287 {
 288         int i;
 289
 290         for (i = 0; i < tu->tp.nr_args; i++)
 291                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 292
 293         iput(tu->inode);
 294         kfree(tu->tp.call.class->system);
 295         kfree(tu->tp.call.name);
 296         kfree(tu->filename);
 297         kfree(tu);
 298 }
 299
 300 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 301 {
 302         struct trace_uprobe *tu;
 303
 304         list_for_each_entry(tu, &uprobe_list, list)
 305                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 306                     strcmp(tu->tp.call.class->system, group) == 0)
 307                         return tu;
 308
 309         return NULL;
 310 }
 311
 312 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 313 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 314 {
 315         int ret;
 316
 317         ret = unregister_uprobe_event(tu);
 318         if (ret)
 319                 return ret;
 320
 321         list_del(&tu->list);
 322         free_trace_uprobe(tu);
 323         return 0;
 324 }
 325
 326 /* Register a trace_uprobe and probe_event */
 327 static int register_trace_uprobe(struct trace_uprobe *tu)
 328 {
 329         struct trace_uprobe *old_tu;
 330         int ret;
 331
 332         mutex_lock(&uprobe_lock);
 333
 334         /* register as an event */
 335         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
 336                         tu->tp.call.class->system);
 337         if (old_tu) {
 338                 /* delete old event */
 339                 ret = unregister_trace_uprobe(old_tu);
 340                 if (ret)
 341                         goto end;
 342         }
 343
 344         ret = register_uprobe_event(tu);
 345         if (ret) {
 346                 pr_warning("Failed to register probe event(%d)\n", ret);
 347                 goto end;
 348         }
 349
 350         list_add_tail(&tu->list, &uprobe_list);
 351
 352 end:
 353         mutex_unlock(&uprobe_lock);
 354
 355         return ret;
 356 }
 357
 358 /*
 359  * Argument syntax:
 360  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 361  *
 362  *  - Remove uprobe: -:[GRP/]EVENT
 363  */
 364 static int create_trace_uprobe(int argc, char **argv)
 365 {
 366         struct trace_uprobe *tu;
 367         struct inode *inode;
 368         char *arg, *event, *group, *filename;
 369         char buf[MAX_EVENT_NAME_LEN];
 370         struct path path;
 371         unsigned long offset;
 372         bool is_delete, is_return;
 373         int i, ret;
 374
 375         inode = NULL;
 376         ret = 0;
 377         is_delete = false;
 378         is_return = false;
 379         event = NULL;
 380         group = NULL;
 381
 382         /* argc must be >= 1 */
 383         if (argv[0][0] == '-')
 384                 is_delete = true;
 385         else if (argv[0][0] == 'r')
 386                 is_return = true;
 387         else if (argv[0][0] != 'p') {
 388                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 389                 return -EINVAL;
 390         }
 391
 392         if (argv[0][1] == ':') {
 393                 event = &argv[0][2];
 394                 arg = strchr(event, '/');
 395
 396                 if (arg) {
 397                         group = event;
 398                         event = arg + 1;
 399                         event[-1] = '\0';
 400
 401                         if (strlen(group) == 0) {
 402                                 pr_info("Group name is not specified\n");
 403                                 return -EINVAL;
 404                         }
 405                 }
 406                 if (strlen(event) == 0) {
 407                         pr_info("Event name is not specified\n");
 408                         return -EINVAL;
 409                 }
 410         }
 411         if (!group)
 412                 group = UPROBE_EVENT_SYSTEM;
 413
 414         if (is_delete) {
 415                 int ret;
 416
 417                 if (!event) {
 418                         pr_info("Delete command needs an event name.\n");
 419                         return -EINVAL;
 420                 }
 421                 mutex_lock(&uprobe_lock);
 422                 tu = find_probe_event(event, group);
 423
 424                 if (!tu) {
 425                         mutex_unlock(&uprobe_lock);
 426                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 427                         return -ENOENT;
 428                 }
 429                 /* delete an event */
 430                 ret = unregister_trace_uprobe(tu);
 431                 mutex_unlock(&uprobe_lock);
 432                 return ret;
 433         }
 434
 435         if (argc < 2) {
 436                 pr_info("Probe point is not specified.\n");
 437                 return -EINVAL;
 438         }
 439         if (isdigit(argv[1][0])) {
 440                 pr_info("probe point must be have a filename.\n");
 441                 return -EINVAL;
 442         }
 443         arg = strchr(argv[1], ':');
 444         if (!arg) {
 445                 ret = -EINVAL;
 446                 goto fail_address_parse;
 447         }
 448
 449         *arg++ = '\0';
 450         filename = argv[1];
 451         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 452         if (ret)
 453                 goto fail_address_parse;
 454
 455         inode = igrab(d_inode(path.dentry));
 456         path_put(&path);
 457
 458         if (!inode || !S_ISREG(inode->i_mode)) {
 459                 ret = -EINVAL;
 460                 goto fail_address_parse;
 461         }
 462
 463         ret = kstrtoul(arg, 0, &offset);
 464         if (ret)
 465                 goto fail_address_parse;
 466
 467         argc -= 2;
 468         argv += 2;
 469
 470         /* setup a probe */
 471         if (!event) {
 472                 char *tail;
 473                 char *ptr;
 474
 475                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 476                 if (!tail) {
 477                         ret = -ENOMEM;
 478                         goto fail_address_parse;
 479                 }
 480
 481                 ptr = strpbrk(tail, ".-_");
 482                 if (ptr)
 483                         *ptr = '\0';
 484
 485                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 486                 event = buf;
 487                 kfree(tail);
 488         }
 489
 490         tu = alloc_trace_uprobe(group, event, argc, is_return);
 491         if (IS_ERR(tu)) {
 492                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 493                 ret = PTR_ERR(tu);
 494                 goto fail_address_parse;
 495         }
 496         tu->offset = offset;
 497         tu->inode = inode;
 498         tu->filename = kstrdup(filename, GFP_KERNEL);
 499
 500         if (!tu->filename) {
 501                 pr_info("Failed to allocate filename.\n");
 502                 ret = -ENOMEM;
 503                 goto error;
 504         }
 505
 506         /* parse arguments */
 507         ret = 0;
 508         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 509                 struct probe_arg *parg = &tu->tp.args[i];
 510
 511                 /* Increment count for freeing args in error case */
 512                 tu->tp.nr_args++;
 513
 514                 /* Parse argument name */
 515                 arg = strchr(argv[i], '=');
 516                 if (arg) {
 517                         *arg++ = '\0';
 518                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 519                 } else {
 520                         arg = argv[i];
 521                         /* If argument name is omitted, set "argN" */
 522                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 523                         parg->name = kstrdup(buf, GFP_KERNEL);
 524                 }
 525
 526                 if (!parg->name) {
 527                         pr_info("Failed to allocate argument[%d] name.\n", i);
 528                         ret = -ENOMEM;
 529                         goto error;
 530                 }
 531
 532                 if (!is_good_name(parg->name)) {
 533                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 534                         ret = -EINVAL;
 535                         goto error;
 536                 }
 537
 538                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 539                         pr_info("Argument[%d] name '%s' conflicts with "
 540                                 "another field.\n", i, argv[i]);
 541                         ret = -EINVAL;
 542                         goto error;
 543                 }
 544
 545                 /* Parse fetch argument */
 546                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 547                                                  is_return, false,
 548                                                  uprobes_fetch_type_table);
 549                 if (ret) {
 550                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 551                         goto error;
 552                 }
 553         }
 554
 555         ret = register_trace_uprobe(tu);
 556         if (ret)
 557                 goto error;
 558         return 0;
 559
 560 error:
 561         free_trace_uprobe(tu);
 562         return ret;
 563
 564 fail_address_parse:
 565         iput(inode);
 566
 567         pr_info("Failed to parse address or file.\n");
 568
 569         return ret;
 570 }
 571
 572 static int cleanup_all_probes(void)
 573 {
 574         struct trace_uprobe *tu;
 575         int ret = 0;
 576
 577         mutex_lock(&uprobe_lock);
 578         while (!list_empty(&uprobe_list)) {
 579                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 580                 ret = unregister_trace_uprobe(tu);
 581                 if (ret)
 582                         break;
 583         }
 584         mutex_unlock(&uprobe_lock);
 585         return ret;
 586 }
 587
 588 /* Probes listing interfaces */
 589 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 590 {
 591         mutex_lock(&uprobe_lock);
 592         return seq_list_start(&uprobe_list, *pos);
 593 }
 594
 595 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 596 {
 597         return seq_list_next(v, &uprobe_list, pos);
 598 }
 599
 600 static void probes_seq_stop(struct seq_file *m, void *v)
 601 {
 602         mutex_unlock(&uprobe_lock);
 603 }
 604
 605 static int probes_seq_show(struct seq_file *m, void *v)
 606 {
 607         struct trace_uprobe *tu = v;
 608         char c = is_ret_probe(tu) ? 'r' : 'p';
 609         int i;
 610
 611         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 612                         trace_event_name(&tu->tp.call));
 613         seq_printf(m, " %s:", tu->filename);
 614
 615         /* Don't print "0x  (null)" when offset is 0 */
 616         if (tu->offset) {
 617                 seq_printf(m, "0x%p", (void *)tu->offset);
 618         } else {
 619                 switch (sizeof(void *)) {
 620                 case 4:
 621                         seq_printf(m, "0x00000000");
 622                         break;
 623                 case 8:
 624                 default:
 625                         seq_printf(m, "0x0000000000000000");
 626                         break;
 627                 }
 628         }
 629
 630         for (i = 0; i < tu->tp.nr_args; i++)
 631                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 632
 633         seq_putc(m, '\n');
 634         return 0;
 635 }
 636
 637 static const struct seq_operations probes_seq_op = {
 638         .start  = probes_seq_start,
 639         .next   = probes_seq_next,
 640         .stop   = probes_seq_stop,
 641         .show   = probes_seq_show
 642 };
 643
 644 static int probes_open(struct inode *inode, struct file *file)
 645 {
 646         int ret;
 647
 648         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 649                 ret = cleanup_all_probes();
 650                 if (ret)
 651                         return ret;
 652         }
 653
 654         return seq_open(file, &probes_seq_op);
 655 }
 656
 657 static ssize_t probes_write(struct file *file, const char __user *buffer,
 658                             size_t count, loff_t *ppos)
 659 {
 660         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 661 }
 662
 663 static const struct file_operations uprobe_events_ops = {
 664         .owner          = THIS_MODULE,
 665         .open           = probes_open,
 666         .read           = seq_read,
 667         .llseek         = seq_lseek,
 668         .release        = seq_release,
 669         .write          = probes_write,
 670 };
 671
 672 /* Probes profiling interfaces */
 673 static int probes_profile_seq_show(struct seq_file *m, void *v)
 674 {
 675         struct trace_uprobe *tu = v;
 676
 677         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 678                         trace_event_name(&tu->tp.call), tu->nhit);
 679         return 0;
 680 }
 681
 682 static const struct seq_operations profile_seq_op = {
 683         .start  = probes_seq_start,
 684         .next   = probes_seq_next,
 685         .stop   = probes_seq_stop,
 686         .show   = probes_profile_seq_show
 687 };
 688
 689 static int profile_open(struct inode *inode, struct file *file)
 690 {
 691         return seq_open(file, &profile_seq_op);
 692 }
 693
 694 static const struct file_operations uprobe_profile_ops = {
 695         .owner          = THIS_MODULE,
 696         .open           = profile_open,
 697         .read           = seq_read,
 698         .llseek         = seq_lseek,
 699         .release        = seq_release,
 700 };
 701
 702 struct uprobe_cpu_buffer {
 703         struct mutex mutex;
 704         void *buf;
 705 };
 706 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 707 static int uprobe_buffer_refcnt;
 708
 709 static int uprobe_buffer_init(void)
 710 {
 711         int cpu, err_cpu;
 712
 713         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 714         if (uprobe_cpu_buffer == NULL)
 715                 return -ENOMEM;
 716
 717         for_each_possible_cpu(cpu) {
 718                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 719                                                   GFP_KERNEL, 0);
 720                 if (p == NULL) {
 721                         err_cpu = cpu;
 722                         goto err;
 723                 }
 724                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 725                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 726         }
 727
 728         return 0;
 729
 730 err:
 731         for_each_possible_cpu(cpu) {
 732                 if (cpu == err_cpu)
 733                         break;
 734                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 735         }
 736
 737         free_percpu(uprobe_cpu_buffer);
 738         return -ENOMEM;
 739 }
 740
 741 static int uprobe_buffer_enable(void)
 742 {
 743         int ret = 0;
 744
 745         BUG_ON(!mutex_is_locked(&event_mutex));
 746
 747         if (uprobe_buffer_refcnt++ == 0) {
 748                 ret = uprobe_buffer_init();
 749                 if (ret < 0)
 750                         uprobe_buffer_refcnt--;
 751         }
 752
 753         return ret;
 754 }
 755
 756 static void uprobe_buffer_disable(void)
 757 {
 758         int cpu;
 759
 760         BUG_ON(!mutex_is_locked(&event_mutex));
 761
 762         if (--uprobe_buffer_refcnt == 0) {
 763                 for_each_possible_cpu(cpu)
 764                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 765                                                              cpu)->buf);
 766
 767                 free_percpu(uprobe_cpu_buffer);
 768                 uprobe_cpu_buffer = NULL;
 769         }
 770 }
 771
 772 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 773 {
 774         struct uprobe_cpu_buffer *ucb;
 775         int cpu;
 776
 777         cpu = raw_smp_processor_id();
 778         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 779
 780         /*
 781          * Use per-cpu buffers for fastest access, but we might migrate
 782          * so the mutex makes sure we have sole access to it.
 783          */
 784         mutex_lock(&ucb->mutex);
 785
 786         return ucb;
 787 }
 788
 789 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 790 {
 791         mutex_unlock(&ucb->mutex);
 792 }
 793
 794 static void __uprobe_trace_func(struct trace_uprobe *tu,
 795                                 unsigned long func, struct pt_regs *regs,
 796                                 struct uprobe_cpu_buffer *ucb, int dsize,
 797                                 struct trace_event_file *trace_file)
 798 {
 799         struct uprobe_trace_entry_head *entry;
 800         struct ring_buffer_event *event;
 801         struct ring_buffer *buffer;
 802         void *data;
 803         int size, esize;
 804         struct trace_event_call *call = &tu->tp.call;
 805
 806         WARN_ON(call != trace_file->event_call);
 807
 808         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 809                 return;
 810
 811         if (trace_trigger_soft_disabled(trace_file))
 812                 return;
 813
 814         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 815         size = esize + tu->tp.size + dsize;
 816         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 817                                                 call->event.type, size, 0, 0);
 818         if (!event)
 819                 return;
 820
 821         entry = ring_buffer_event_data(event);
 822         if (is_ret_probe(tu)) {
 823                 entry->vaddr[0] = func;
 824                 entry->vaddr[1] = instruction_pointer(regs);
 825                 data = DATAOF_TRACE_ENTRY(entry, true);
 826         } else {
 827                 entry->vaddr[0] = instruction_pointer(regs);
 828                 data = DATAOF_TRACE_ENTRY(entry, false);
 829         }
 830
 831         memcpy(data, ucb->buf, tu->tp.size + dsize);
 832
 833         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 834 }
 835
 836 /* uprobe handler */
 837 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 838                              struct uprobe_cpu_buffer *ucb, int dsize)
 839 {
 840         struct event_file_link *link;
 841
 842         if (is_ret_probe(tu))
 843                 return 0;
 844
 845         rcu_read_lock();
 846         list_for_each_entry_rcu(link, &tu->tp.files, list)
 847                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 848         rcu_read_unlock();
 849
 850         return 0;
 851 }
 852
 853 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 854                                  struct pt_regs *regs,
 855                                  struct uprobe_cpu_buffer *ucb, int dsize)
 856 {
 857         struct event_file_link *link;
 858
 859         rcu_read_lock();
 860         list_for_each_entry_rcu(link, &tu->tp.files, list)
 861                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 862         rcu_read_unlock();
 863 }
 864
 865 /* Event entry printers */
 866 static enum print_line_t
 867 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 868 {
 869         struct uprobe_trace_entry_head *entry;
 870         struct trace_seq *s = &iter->seq;
 871         struct trace_uprobe *tu;
 872         u8 *data;
 873         int i;
 874
 875         entry = (struct uprobe_trace_entry_head *)iter->ent;
 876         tu = container_of(event, struct trace_uprobe, tp.call.event);
 877
 878         if (is_ret_probe(tu)) {
 879                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 880                                  trace_event_name(&tu->tp.call),
 881                                  entry->vaddr[1], entry->vaddr[0]);
 882                 data = DATAOF_TRACE_ENTRY(entry, true);
 883         } else {
 884                 trace_seq_printf(s, "%s: (0x%lx)",
 885                                  trace_event_name(&tu->tp.call),
 886                                  entry->vaddr[0]);
 887                 data = DATAOF_TRACE_ENTRY(entry, false);
 888         }
 889
 890         for (i = 0; i < tu->tp.nr_args; i++) {
 891                 struct probe_arg *parg = &tu->tp.args[i];
 892
 893                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 894                         goto out;
 895         }
 896
 897         trace_seq_putc(s, '\n');
 898
 899  out:
 900         return trace_handle_return(s);
 901 }
 902
 903 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 904                                 enum uprobe_filter_ctx ctx,
 905                                 struct mm_struct *mm);
 906
 907 static int
 908 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
 909                    filter_func_t filter)
 910 {
 911         bool enabled = trace_probe_is_enabled(&tu->tp);
 912         struct event_file_link *link = NULL;
 913         int ret;
 914
 915         if (file) {
 916                 if (tu->tp.flags & TP_FLAG_PROFILE)
 917                         return -EINTR;
 918
 919                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 920                 if (!link)
 921                         return -ENOMEM;
 922
 923                 link->file = file;
 924                 list_add_tail_rcu(&link->list, &tu->tp.files);
 925
 926                 tu->tp.flags |= TP_FLAG_TRACE;
 927         } else {
 928                 if (tu->tp.flags & TP_FLAG_TRACE)
 929                         return -EINTR;
 930
 931                 tu->tp.flags |= TP_FLAG_PROFILE;
 932         }
 933
 934         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 935
 936         if (enabled)
 937                 return 0;
 938
 939         ret = uprobe_buffer_enable();
 940         if (ret)
 941                 goto err_flags;
 942
 943         tu->consumer.filter = filter;
 944         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 945         if (ret)
 946                 goto err_buffer;
 947
 948         return 0;
 949
 950  err_buffer:
 951         uprobe_buffer_disable();
 952
 953  err_flags:
 954         if (file) {
 955                 list_del(&link->list);
 956                 kfree(link);
 957                 tu->tp.flags &= ~TP_FLAG_TRACE;
 958         } else {
 959                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 960         }
 961         return ret;
 962 }
 963
 964 static void
 965 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 966 {
 967         if (!trace_probe_is_enabled(&tu->tp))
 968                 return;
 969
 970         if (file) {
 971                 struct event_file_link *link;
 972
 973                 link = find_event_file_link(&tu->tp, file);
 974                 if (!link)
 975                         return;
 976
 977                 list_del_rcu(&link->list);
 978                 /* synchronize with u{,ret}probe_trace_func */
 979                 synchronize_rcu();
 980                 kfree(link);
 981
 982                 if (!list_empty(&tu->tp.files))
 983                         return;
 984         }
 985
 986         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 987
 988         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 989         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 990
 991         uprobe_buffer_disable();
 992 }
 993
 994 static int uprobe_event_define_fields(struct trace_event_call *event_call)
 995 {
 996         int ret, i, size;
 997         struct uprobe_trace_entry_head field;
 998         struct trace_uprobe *tu = event_call->data;
 999
1000         if (is_ret_probe(tu)) {
1001                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
1002                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
1003                 size = SIZEOF_TRACE_ENTRY(true);
1004         } else {
1005                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1006                 size = SIZEOF_TRACE_ENTRY(false);
1007         }
1008         /* Set argument names as fields */
1009         for (i = 0; i < tu->tp.nr_args; i++) {
1010                 struct probe_arg *parg = &tu->tp.args[i];
1011
1012                 ret = trace_define_field(event_call, parg->type->fmttype,
1013                                          parg->name, size + parg->offset,
1014                                          parg->type->size, parg->type->is_signed,
1015                                          FILTER_OTHER);
1016
1017                 if (ret)
1018                         return ret;
1019         }
1020         return 0;
1021 }
1022
1023 #ifdef CONFIG_PERF_EVENTS
1024 static bool
1025 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1026 {
1027         struct perf_event *event;
1028
1029         if (filter->nr_systemwide)
1030                 return true;
1031
1032         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1033                 if (event->hw.target->mm == mm)
1034                         return true;
1035         }
1036
1037         return false;
1038 }
1039
1040 static inline bool
1041 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1042 {
1043         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1044 }
1045
1046 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1047 {
1048         bool done;
1049
1050         write_lock(&tu->filter.rwlock);
1051         if (event->hw.target) {
1052                 list_del(&event->hw.tp_list);
1053                 done = tu->filter.nr_systemwide ||
1054                         (event->hw.target->flags & PF_EXITING) ||
1055                         uprobe_filter_event(tu, event);
1056         } else {
1057                 tu->filter.nr_systemwide--;
1058                 done = tu->filter.nr_systemwide;
1059         }
1060         write_unlock(&tu->filter.rwlock);
1061
1062         if (!done)
1063                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1064
1065         return 0;
1066 }
1067
1068 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1069 {
1070         bool done;
1071         int err;
1072
1073         write_lock(&tu->filter.rwlock);
1074         if (event->hw.target) {
1075                 /*
1076                  * event->parent != NULL means copy_process(), we can avoid
1077                  * uprobe_apply(). current->mm must be probed and we can rely
1078                  * on dup_mmap() which preserves the already installed bp's.
1079                  *
1080                  * attr.enable_on_exec means that exec/mmap will install the
1081                  * breakpoints we need.
1082                  */
1083                 done = tu->filter.nr_systemwide ||
1084                         event->parent || event->attr.enable_on_exec ||
1085                         uprobe_filter_event(tu, event);
1086                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1087         } else {
1088                 done = tu->filter.nr_systemwide;
1089                 tu->filter.nr_systemwide++;
1090         }
1091         write_unlock(&tu->filter.rwlock);
1092
1093         err = 0;
1094         if (!done) {
1095                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1096                 if (err)
1097                         uprobe_perf_close(tu, event);
1098         }
1099         return err;
1100 }
1101
1102 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1103                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1104 {
1105         struct trace_uprobe *tu;
1106         int ret;
1107
1108         tu = container_of(uc, struct trace_uprobe, consumer);
1109         read_lock(&tu->filter.rwlock);
1110         ret = __uprobe_perf_filter(&tu->filter, mm);
1111         read_unlock(&tu->filter.rwlock);
1112
1113         return ret;
1114 }
1115
1116 static void __uprobe_perf_func(struct trace_uprobe *tu,
1117                                unsigned long func, struct pt_regs *regs,
1118                                struct uprobe_cpu_buffer *ucb, int dsize)
1119 {
1120         struct trace_event_call *call = &tu->tp.call;
1121         struct uprobe_trace_entry_head *entry;
1122         struct bpf_prog *prog = call->prog;
1123         struct hlist_head *head;
1124         void *data;
1125         int size, esize;
1126         int rctx;
1127
1128         if (prog && !trace_call_bpf(prog, regs))
1129                 return;
1130
1131         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1132
1133         size = esize + tu->tp.size + dsize;
1134         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1135         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1136                 return;
1137
1138         preempt_disable();
1139         head = this_cpu_ptr(call->perf_events);
1140         if (hlist_empty(head))
1141                 goto out;
1142
1143         entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1144         if (!entry)
1145                 goto out;
1146
1147         if (is_ret_probe(tu)) {
1148                 entry->vaddr[0] = func;
1149                 entry->vaddr[1] = instruction_pointer(regs);
1150                 data = DATAOF_TRACE_ENTRY(entry, true);
1151         } else {
1152                 entry->vaddr[0] = instruction_pointer(regs);
1153                 data = DATAOF_TRACE_ENTRY(entry, false);
1154         }
1155
1156         memcpy(data, ucb->buf, tu->tp.size + dsize);
1157
1158         if (size - esize > tu->tp.size + dsize) {
1159                 int len = tu->tp.size + dsize;
1160
1161                 memset(data + len, 0, size - esize - len);
1162         }
1163
1164         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1165  out:
1166         preempt_enable();
1167 }
1168
1169 /* uprobe profile handler */
1170 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1171                             struct uprobe_cpu_buffer *ucb, int dsize)
1172 {
1173         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1174                 return UPROBE_HANDLER_REMOVE;
1175
1176         if (!is_ret_probe(tu))
1177                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1178         return 0;
1179 }
1180
1181 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1182                                 struct pt_regs *regs,
1183                                 struct uprobe_cpu_buffer *ucb, int dsize)
1184 {
1185         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1186 }
1187 #endif  /* CONFIG_PERF_EVENTS */
1188
1189 static int
1190 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1191                       void *data)
1192 {
1193         struct trace_uprobe *tu = event->data;
1194         struct trace_event_file *file = data;
1195
1196         switch (type) {
1197         case TRACE_REG_REGISTER:
1198                 return probe_event_enable(tu, file, NULL);
1199
1200         case TRACE_REG_UNREGISTER:
1201                 probe_event_disable(tu, file);
1202                 return 0;
1203
1204 #ifdef CONFIG_PERF_EVENTS
1205         case TRACE_REG_PERF_REGISTER:
1206                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1207
1208         case TRACE_REG_PERF_UNREGISTER:
1209                 probe_event_disable(tu, NULL);
1210                 return 0;
1211
1212         case TRACE_REG_PERF_OPEN:
1213                 return uprobe_perf_open(tu, data);
1214
1215         case TRACE_REG_PERF_CLOSE:
1216                 return uprobe_perf_close(tu, data);
1217
1218 #endif
1219         default:
1220                 return 0;
1221         }
1222         return 0;
1223 }
1224
1225 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1226 {
1227         struct trace_uprobe *tu;
1228         struct uprobe_dispatch_data udd;
1229         struct uprobe_cpu_buffer *ucb;
1230         int dsize, esize;
1231         int ret = 0;
1232
1233
1234         tu = container_of(con, struct trace_uprobe, consumer);
1235         tu->nhit++;
1236
1237         udd.tu = tu;
1238         udd.bp_addr = instruction_pointer(regs);
1239
1240         current->utask->vaddr = (unsigned long) &udd;
1241
1242         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1243                 return 0;
1244
1245         dsize = __get_data_size(&tu->tp, regs);
1246         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1247
1248         ucb = uprobe_buffer_get();
1249         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1250
1251         if (tu->tp.flags & TP_FLAG_TRACE)
1252                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1253
1254 #ifdef CONFIG_PERF_EVENTS
1255         if (tu->tp.flags & TP_FLAG_PROFILE)
1256                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1257 #endif
1258         uprobe_buffer_put(ucb);
1259         return ret;
1260 }
1261
1262 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1263                                 unsigned long func, struct pt_regs *regs)
1264 {
1265         struct trace_uprobe *tu;
1266         struct uprobe_dispatch_data udd;
1267         struct uprobe_cpu_buffer *ucb;
1268         int dsize, esize;
1269
1270         tu = container_of(con, struct trace_uprobe, consumer);
1271
1272         udd.tu = tu;
1273         udd.bp_addr = func;
1274
1275         current->utask->vaddr = (unsigned long) &udd;
1276
1277         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1278                 return 0;
1279
1280         dsize = __get_data_size(&tu->tp, regs);
1281         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1282
1283         ucb = uprobe_buffer_get();
1284         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1285
1286         if (tu->tp.flags & TP_FLAG_TRACE)
1287                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1288
1289 #ifdef CONFIG_PERF_EVENTS
1290         if (tu->tp.flags & TP_FLAG_PROFILE)
1291                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1292 #endif
1293         uprobe_buffer_put(ucb);
1294         return 0;
1295 }
1296
1297 static struct trace_event_functions uprobe_funcs = {
1298         .trace          = print_uprobe_event
1299 };
1300
1301 static int register_uprobe_event(struct trace_uprobe *tu)
1302 {
1303         struct trace_event_call *call = &tu->tp.call;
1304         int ret;
1305
1306         /* Initialize trace_event_call */
1307         INIT_LIST_HEAD(&call->class->fields);
1308         call->event.funcs = &uprobe_funcs;
1309         call->class->define_fields = uprobe_event_define_fields;
1310
1311         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1312                 return -ENOMEM;
1313
1314         ret = register_trace_event(&call->event);
1315         if (!ret) {
1316                 kfree(call->print_fmt);
1317                 return -ENODEV;
1318         }
1319
1320         call->flags = TRACE_EVENT_FL_UPROBE;
1321         call->class->reg = trace_uprobe_register;
1322         call->data = tu;
1323         ret = trace_add_event_call(call);
1324
1325         if (ret) {
1326                 pr_info("Failed to register uprobe event: %s\n",
1327                         trace_event_name(call));
1328                 kfree(call->print_fmt);
1329                 unregister_trace_event(&call->event);
1330         }
1331
1332         return ret;
1333 }
1334
1335 static int unregister_uprobe_event(struct trace_uprobe *tu)
1336 {
1337         int ret;
1338
1339         /* tu->event is unregistered in trace_remove_event_call() */
1340         ret = trace_remove_event_call(&tu->tp.call);
1341         if (ret)
1342                 return ret;
1343         kfree(tu->tp.call.print_fmt);
1344         tu->tp.call.print_fmt = NULL;
1345         return 0;
1346 }
1347
1348 /* Make a trace interface for controling probe points */
1349 static __init int init_uprobe_trace(void)
1350 {
1351         struct dentry *d_tracer;
1352
1353         d_tracer = tracing_init_dentry();
1354         if (IS_ERR(d_tracer))
1355                 return 0;
1356
1357         trace_create_file("uprobe_events", 0644, d_tracer,
1358                                     NULL, &uprobe_events_ops);
1359         /* Profile interface */
1360         trace_create_file("uprobe_profile", 0444, d_tracer,
1361                                     NULL, &uprobe_profile_ops);
1362         return 0;
1363 }
1364
1365 fs_initcall(init_uprobe_trace);