kernel/trace/trace_uprobe.c

   1 /*
   2  * uprobes-based tracing events
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  16  *
  17  * Copyright (C) IBM Corporation, 2010-2012
  18  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
  19  */
  20
  21 #include <linux/module.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/uprobes.h>
  24 #include <linux/namei.h>
  25 #include <linux/string.h>
  26
  27 #include "trace_probe.h"
  28
  29 #define UPROBE_EVENT_SYSTEM     "uprobes"
  30
  31 struct uprobe_trace_entry_head {
  32         struct trace_entry      ent;
  33         unsigned long           vaddr[];
  34 };
  35
  36 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  37         (sizeof(struct uprobe_trace_entry_head) +       \
  38          sizeof(unsigned long) * (is_return ? 2 : 1))
  39
  40 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  41         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  42
  43 struct trace_uprobe_filter {
  44         rwlock_t                rwlock;
  45         int                     nr_systemwide;
  46         struct list_head        perf_events;
  47 };
  48
  49 /*
  50  * uprobe event core functions
  51  */
  52 struct trace_uprobe {
  53         struct list_head                list;
  54         struct trace_uprobe_filter      filter;
  55         struct uprobe_consumer          consumer;
  56         struct inode                    *inode;
  57         char                            *filename;
  58         unsigned long                   offset;
  59         unsigned long                   nhit;
  60         struct trace_probe              tp;
  61 };
  62
  63 #define SIZEOF_TRACE_UPROBE(n)                          \
  64         (offsetof(struct trace_uprobe, tp.args) +       \
  65         (sizeof(struct probe_arg) * (n)))
  66
  67 static int register_uprobe_event(struct trace_uprobe *tu);
  68 static int unregister_uprobe_event(struct trace_uprobe *tu);
  69
  70 static DEFINE_MUTEX(uprobe_lock);
  71 static LIST_HEAD(uprobe_list);
  72
  73 struct uprobe_dispatch_data {
  74         struct trace_uprobe     *tu;
  75         unsigned long           bp_addr;
  76 };
  77
  78 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  79 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  80                                 unsigned long func, struct pt_regs *regs);
  81
  82 #ifdef CONFIG_STACK_GROWSUP
  83 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  84 {
  85         return addr - (n * sizeof(long));
  86 }
  87 #else
  88 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  89 {
  90         return addr + (n * sizeof(long));
  91 }
  92 #endif
  93
  94 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  95 {
  96         unsigned long ret;
  97         unsigned long addr = user_stack_pointer(regs);
  98
  99         addr = adjust_stack_addr(addr, n);
 100
 101         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 102                 return 0;
 103
 104         return ret;
 105 }
 106
 107 /*
 108  * Uprobes-specific fetch functions
 109  */
 110 #define DEFINE_FETCH_stack(type)                                        \
 111 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 112                                          void *offset, void *dest)      \
 113 {                                                                       \
 114         *(type *)dest = (type)get_user_stack_nth(regs,                  \
 115                                               ((unsigned long)offset)); \
 116 }
 117 DEFINE_BASIC_FETCH_FUNCS(stack)
 118 /* No string on the stack entry */
 119 #define fetch_stack_string      NULL
 120 #define fetch_stack_string_size NULL
 121
 122 #define DEFINE_FETCH_memory(type)                                       \
 123 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 124                                           void *addr, void *dest)       \
 125 {                                                                       \
 126         type retval;                                                    \
 127         void __user *vaddr = (void __force __user *) addr;              \
 128                                                                         \
 129         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
 130                 *(type *)dest = 0;                                      \
 131         else                                                            \
 132                 *(type *) dest = retval;                                \
 133 }
 134 DEFINE_BASIC_FETCH_FUNCS(memory)
 135 /*
 136  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 137  * length and relative data location.
 138  */
 139 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 140                                             void *addr, void *dest)
 141 {
 142         long ret;
 143         u32 rloc = *(u32 *)dest;
 144         int maxlen  = get_rloc_len(rloc);
 145         u8 *dst = get_rloc_data(dest);
 146         void __user *src = (void __force __user *) addr;
 147
 148         if (!maxlen)
 149                 return;
 150
 151         ret = strncpy_from_user(dst, src, maxlen);
 152
 153         if (ret < 0) {  /* Failed to fetch string */
 154                 ((u8 *)get_rloc_data(dest))[0] = '\0';
 155                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 156         } else {
 157                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 158         }
 159 }
 160
 161 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 162                                                  void *addr, void *dest)
 163 {
 164         int len;
 165         void __user *vaddr = (void __force __user *) addr;
 166
 167         len = strnlen_user(vaddr, MAX_STRING_SIZE);
 168
 169         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 170                 *(u32 *)dest = 0;
 171         else
 172                 *(u32 *)dest = len;
 173 }
 174
 175 static unsigned long translate_user_vaddr(void *file_offset)
 176 {
 177         unsigned long base_addr;
 178         struct uprobe_dispatch_data *udd;
 179
 180         udd = (void *) current->utask->vaddr;
 181
 182         base_addr = udd->bp_addr - udd->tu->offset;
 183         return base_addr + (unsigned long)file_offset;
 184 }
 185
 186 #define DEFINE_FETCH_file_offset(type)                                  \
 187 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,    \
 188                                                void *offset, void *dest)\
 189 {                                                                       \
 190         void *vaddr = (void *)translate_user_vaddr(offset);             \
 191                                                                         \
 192         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
 193 }
 194 DEFINE_BASIC_FETCH_FUNCS(file_offset)
 195 DEFINE_FETCH_file_offset(string)
 196 DEFINE_FETCH_file_offset(string_size)
 197
 198 /* Fetch type information table */
 199 static const struct fetch_type uprobes_fetch_type_table[] = {
 200         /* Special types */
 201         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 202                                         sizeof(u32), 1, "__data_loc char[]"),
 203         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 204                                         string_size, sizeof(u32), 0, "u32"),
 205         /* Basic types */
 206         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 207         ASSIGN_FETCH_TYPE(u16, u16, 0),
 208         ASSIGN_FETCH_TYPE(u32, u32, 0),
 209         ASSIGN_FETCH_TYPE(u64, u64, 0),
 210         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 211         ASSIGN_FETCH_TYPE(s16, u16, 1),
 212         ASSIGN_FETCH_TYPE(s32, u32, 1),
 213         ASSIGN_FETCH_TYPE(s64, u64, 1),
 214
 215         ASSIGN_FETCH_TYPE_END
 216 };
 217
 218 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 219 {
 220         rwlock_init(&filter->rwlock);
 221         filter->nr_systemwide = 0;
 222         INIT_LIST_HEAD(&filter->perf_events);
 223 }
 224
 225 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 226 {
 227         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 228 }
 229
 230 static inline bool is_ret_probe(struct trace_uprobe *tu)
 231 {
 232         return tu->consumer.ret_handler != NULL;
 233 }
 234
 235 /*
 236  * Allocate new trace_uprobe and initialize it (including uprobes).
 237  */
 238 static struct trace_uprobe *
 239 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 240 {
 241         struct trace_uprobe *tu;
 242
 243         if (!event || !is_good_name(event))
 244                 return ERR_PTR(-EINVAL);
 245
 246         if (!group || !is_good_name(group))
 247                 return ERR_PTR(-EINVAL);
 248
 249         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 250         if (!tu)
 251                 return ERR_PTR(-ENOMEM);
 252
 253         tu->tp.call.class = &tu->tp.class;
 254         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 255         if (!tu->tp.call.name)
 256                 goto error;
 257
 258         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 259         if (!tu->tp.class.system)
 260                 goto error;
 261
 262         INIT_LIST_HEAD(&tu->list);
 263         INIT_LIST_HEAD(&tu->tp.files);
 264         tu->consumer.handler = uprobe_dispatcher;
 265         if (is_ret)
 266                 tu->consumer.ret_handler = uretprobe_dispatcher;
 267         init_trace_uprobe_filter(&tu->filter);
 268         return tu;
 269
 270 error:
 271         kfree(tu->tp.call.name);
 272         kfree(tu);
 273
 274         return ERR_PTR(-ENOMEM);
 275 }
 276
 277 static void free_trace_uprobe(struct trace_uprobe *tu)
 278 {
 279         int i;
 280
 281         for (i = 0; i < tu->tp.nr_args; i++)
 282                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 283
 284         iput(tu->inode);
 285         kfree(tu->tp.call.class->system);
 286         kfree(tu->tp.call.name);
 287         kfree(tu->filename);
 288         kfree(tu);
 289 }
 290
 291 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 292 {
 293         struct trace_uprobe *tu;
 294
 295         list_for_each_entry(tu, &uprobe_list, list)
 296                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 297                     strcmp(tu->tp.call.class->system, group) == 0)
 298                         return tu;
 299
 300         return NULL;
 301 }
 302
 303 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 304 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 305 {
 306         int ret;
 307
 308         ret = unregister_uprobe_event(tu);
 309         if (ret)
 310                 return ret;
 311
 312         list_del(&tu->list);
 313         free_trace_uprobe(tu);
 314         return 0;
 315 }
 316
 317 /* Register a trace_uprobe and probe_event */
 318 static int register_trace_uprobe(struct trace_uprobe *tu)
 319 {
 320         struct trace_uprobe *old_tu;
 321         int ret;
 322
 323         mutex_lock(&uprobe_lock);
 324
 325         /* register as an event */
 326         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
 327                         tu->tp.call.class->system);
 328         if (old_tu) {
 329                 /* delete old event */
 330                 ret = unregister_trace_uprobe(old_tu);
 331                 if (ret)
 332                         goto end;
 333         }
 334
 335         ret = register_uprobe_event(tu);
 336         if (ret) {
 337                 pr_warn("Failed to register probe event(%d)\n", ret);
 338                 goto end;
 339         }
 340
 341         list_add_tail(&tu->list, &uprobe_list);
 342
 343 end:
 344         mutex_unlock(&uprobe_lock);
 345
 346         return ret;
 347 }
 348
 349 /*
 350  * Argument syntax:
 351  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 352  *
 353  *  - Remove uprobe: -:[GRP/]EVENT
 354  */
 355 static int create_trace_uprobe(int argc, char **argv)
 356 {
 357         struct trace_uprobe *tu;
 358         struct inode *inode;
 359         char *arg, *event, *group, *filename;
 360         char buf[MAX_EVENT_NAME_LEN];
 361         struct path path;
 362         unsigned long offset;
 363         bool is_delete, is_return;
 364         int i, ret;
 365
 366         inode = NULL;
 367         ret = 0;
 368         is_delete = false;
 369         is_return = false;
 370         event = NULL;
 371         group = NULL;
 372
 373         /* argc must be >= 1 */
 374         if (argv[0][0] == '-')
 375                 is_delete = true;
 376         else if (argv[0][0] == 'r')
 377                 is_return = true;
 378         else if (argv[0][0] != 'p') {
 379                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 380                 return -EINVAL;
 381         }
 382
 383         if (argv[0][1] == ':') {
 384                 event = &argv[0][2];
 385                 arg = strchr(event, '/');
 386
 387                 if (arg) {
 388                         group = event;
 389                         event = arg + 1;
 390                         event[-1] = '\0';
 391
 392                         if (strlen(group) == 0) {
 393                                 pr_info("Group name is not specified\n");
 394                                 return -EINVAL;
 395                         }
 396                 }
 397                 if (strlen(event) == 0) {
 398                         pr_info("Event name is not specified\n");
 399                         return -EINVAL;
 400                 }
 401         }
 402         if (!group)
 403                 group = UPROBE_EVENT_SYSTEM;
 404
 405         if (is_delete) {
 406                 int ret;
 407
 408                 if (!event) {
 409                         pr_info("Delete command needs an event name.\n");
 410                         return -EINVAL;
 411                 }
 412                 mutex_lock(&uprobe_lock);
 413                 tu = find_probe_event(event, group);
 414
 415                 if (!tu) {
 416                         mutex_unlock(&uprobe_lock);
 417                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 418                         return -ENOENT;
 419                 }
 420                 /* delete an event */
 421                 ret = unregister_trace_uprobe(tu);
 422                 mutex_unlock(&uprobe_lock);
 423                 return ret;
 424         }
 425
 426         if (argc < 2) {
 427                 pr_info("Probe point is not specified.\n");
 428                 return -EINVAL;
 429         }
 430         if (isdigit(argv[1][0])) {
 431                 pr_info("probe point must be have a filename.\n");
 432                 return -EINVAL;
 433         }
 434         arg = strchr(argv[1], ':');
 435         if (!arg) {
 436                 ret = -EINVAL;
 437                 goto fail_address_parse;
 438         }
 439
 440         *arg++ = '\0';
 441         filename = argv[1];
 442         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 443         if (ret)
 444                 goto fail_address_parse;
 445
 446         inode = igrab(d_inode(path.dentry));
 447         path_put(&path);
 448
 449         if (!inode || !S_ISREG(inode->i_mode)) {
 450                 ret = -EINVAL;
 451                 goto fail_address_parse;
 452         }
 453
 454         ret = kstrtoul(arg, 0, &offset);
 455         if (ret)
 456                 goto fail_address_parse;
 457
 458         argc -= 2;
 459         argv += 2;
 460
 461         /* setup a probe */
 462         if (!event) {
 463                 char *tail;
 464                 char *ptr;
 465
 466                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 467                 if (!tail) {
 468                         ret = -ENOMEM;
 469                         goto fail_address_parse;
 470                 }
 471
 472                 ptr = strpbrk(tail, ".-_");
 473                 if (ptr)
 474                         *ptr = '\0';
 475
 476                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 477                 event = buf;
 478                 kfree(tail);
 479         }
 480
 481         tu = alloc_trace_uprobe(group, event, argc, is_return);
 482         if (IS_ERR(tu)) {
 483                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 484                 ret = PTR_ERR(tu);
 485                 goto fail_address_parse;
 486         }
 487         tu->offset = offset;
 488         tu->inode = inode;
 489         tu->filename = kstrdup(filename, GFP_KERNEL);
 490
 491         if (!tu->filename) {
 492                 pr_info("Failed to allocate filename.\n");
 493                 ret = -ENOMEM;
 494                 goto error;
 495         }
 496
 497         /* parse arguments */
 498         ret = 0;
 499         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 500                 struct probe_arg *parg = &tu->tp.args[i];
 501
 502                 /* Increment count for freeing args in error case */
 503                 tu->tp.nr_args++;
 504
 505                 /* Parse argument name */
 506                 arg = strchr(argv[i], '=');
 507                 if (arg) {
 508                         *arg++ = '\0';
 509                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 510                 } else {
 511                         arg = argv[i];
 512                         /* If argument name is omitted, set "argN" */
 513                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 514                         parg->name = kstrdup(buf, GFP_KERNEL);
 515                 }
 516
 517                 if (!parg->name) {
 518                         pr_info("Failed to allocate argument[%d] name.\n", i);
 519                         ret = -ENOMEM;
 520                         goto error;
 521                 }
 522
 523                 if (!is_good_name(parg->name)) {
 524                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 525                         ret = -EINVAL;
 526                         goto error;
 527                 }
 528
 529                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 530                         pr_info("Argument[%d] name '%s' conflicts with "
 531                                 "another field.\n", i, argv[i]);
 532                         ret = -EINVAL;
 533                         goto error;
 534                 }
 535
 536                 /* Parse fetch argument */
 537                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 538                                                  is_return, false,
 539                                                  uprobes_fetch_type_table);
 540                 if (ret) {
 541                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 542                         goto error;
 543                 }
 544         }
 545
 546         ret = register_trace_uprobe(tu);
 547         if (ret)
 548                 goto error;
 549         return 0;
 550
 551 error:
 552         free_trace_uprobe(tu);
 553         return ret;
 554
 555 fail_address_parse:
 556         iput(inode);
 557
 558         pr_info("Failed to parse address or file.\n");
 559
 560         return ret;
 561 }
 562
 563 static int cleanup_all_probes(void)
 564 {
 565         struct trace_uprobe *tu;
 566         int ret = 0;
 567
 568         mutex_lock(&uprobe_lock);
 569         while (!list_empty(&uprobe_list)) {
 570                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 571                 ret = unregister_trace_uprobe(tu);
 572                 if (ret)
 573                         break;
 574         }
 575         mutex_unlock(&uprobe_lock);
 576         return ret;
 577 }
 578
 579 /* Probes listing interfaces */
 580 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 581 {
 582         mutex_lock(&uprobe_lock);
 583         return seq_list_start(&uprobe_list, *pos);
 584 }
 585
 586 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 587 {
 588         return seq_list_next(v, &uprobe_list, pos);
 589 }
 590
 591 static void probes_seq_stop(struct seq_file *m, void *v)
 592 {
 593         mutex_unlock(&uprobe_lock);
 594 }
 595
 596 static int probes_seq_show(struct seq_file *m, void *v)
 597 {
 598         struct trace_uprobe *tu = v;
 599         char c = is_ret_probe(tu) ? 'r' : 'p';
 600         int i;
 601
 602         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 603                         trace_event_name(&tu->tp.call));
 604         seq_printf(m, " %s:", tu->filename);
 605
 606         /* Don't print "0x  (null)" when offset is 0 */
 607         if (tu->offset) {
 608                 seq_printf(m, "0x%p", (void *)tu->offset);
 609         } else {
 610                 switch (sizeof(void *)) {
 611                 case 4:
 612                         seq_printf(m, "0x00000000");
 613                         break;
 614                 case 8:
 615                 default:
 616                         seq_printf(m, "0x0000000000000000");
 617                         break;
 618                 }
 619         }
 620
 621         for (i = 0; i < tu->tp.nr_args; i++)
 622                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 623
 624         seq_putc(m, '\n');
 625         return 0;
 626 }
 627
 628 static const struct seq_operations probes_seq_op = {
 629         .start  = probes_seq_start,
 630         .next   = probes_seq_next,
 631         .stop   = probes_seq_stop,
 632         .show   = probes_seq_show
 633 };
 634
 635 static int probes_open(struct inode *inode, struct file *file)
 636 {
 637         int ret;
 638
 639         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 640                 ret = cleanup_all_probes();
 641                 if (ret)
 642                         return ret;
 643         }
 644
 645         return seq_open(file, &probes_seq_op);
 646 }
 647
 648 static ssize_t probes_write(struct file *file, const char __user *buffer,
 649                             size_t count, loff_t *ppos)
 650 {
 651         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 652 }
 653
 654 static const struct file_operations uprobe_events_ops = {
 655         .owner          = THIS_MODULE,
 656         .open           = probes_open,
 657         .read           = seq_read,
 658         .llseek         = seq_lseek,
 659         .release        = seq_release,
 660         .write          = probes_write,
 661 };
 662
 663 /* Probes profiling interfaces */
 664 static int probes_profile_seq_show(struct seq_file *m, void *v)
 665 {
 666         struct trace_uprobe *tu = v;
 667
 668         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 669                         trace_event_name(&tu->tp.call), tu->nhit);
 670         return 0;
 671 }
 672
 673 static const struct seq_operations profile_seq_op = {
 674         .start  = probes_seq_start,
 675         .next   = probes_seq_next,
 676         .stop   = probes_seq_stop,
 677         .show   = probes_profile_seq_show
 678 };
 679
 680 static int profile_open(struct inode *inode, struct file *file)
 681 {
 682         return seq_open(file, &profile_seq_op);
 683 }
 684
 685 static const struct file_operations uprobe_profile_ops = {
 686         .owner          = THIS_MODULE,
 687         .open           = profile_open,
 688         .read           = seq_read,
 689         .llseek         = seq_lseek,
 690         .release        = seq_release,
 691 };
 692
 693 struct uprobe_cpu_buffer {
 694         struct mutex mutex;
 695         void *buf;
 696 };
 697 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 698 static int uprobe_buffer_refcnt;
 699
 700 static int uprobe_buffer_init(void)
 701 {
 702         int cpu, err_cpu;
 703
 704         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 705         if (uprobe_cpu_buffer == NULL)
 706                 return -ENOMEM;
 707
 708         for_each_possible_cpu(cpu) {
 709                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 710                                                   GFP_KERNEL, 0);
 711                 if (p == NULL) {
 712                         err_cpu = cpu;
 713                         goto err;
 714                 }
 715                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 716                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 717         }
 718
 719         return 0;
 720
 721 err:
 722         for_each_possible_cpu(cpu) {
 723                 if (cpu == err_cpu)
 724                         break;
 725                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 726         }
 727
 728         free_percpu(uprobe_cpu_buffer);
 729         return -ENOMEM;
 730 }
 731
 732 static int uprobe_buffer_enable(void)
 733 {
 734         int ret = 0;
 735
 736         BUG_ON(!mutex_is_locked(&event_mutex));
 737
 738         if (uprobe_buffer_refcnt++ == 0) {
 739                 ret = uprobe_buffer_init();
 740                 if (ret < 0)
 741                         uprobe_buffer_refcnt--;
 742         }
 743
 744         return ret;
 745 }
 746
 747 static void uprobe_buffer_disable(void)
 748 {
 749         int cpu;
 750
 751         BUG_ON(!mutex_is_locked(&event_mutex));
 752
 753         if (--uprobe_buffer_refcnt == 0) {
 754                 for_each_possible_cpu(cpu)
 755                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 756                                                              cpu)->buf);
 757
 758                 free_percpu(uprobe_cpu_buffer);
 759                 uprobe_cpu_buffer = NULL;
 760         }
 761 }
 762
 763 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 764 {
 765         struct uprobe_cpu_buffer *ucb;
 766         int cpu;
 767
 768         cpu = raw_smp_processor_id();
 769         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 770
 771         /*
 772          * Use per-cpu buffers for fastest access, but we might migrate
 773          * so the mutex makes sure we have sole access to it.
 774          */
 775         mutex_lock(&ucb->mutex);
 776
 777         return ucb;
 778 }
 779
 780 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 781 {
 782         mutex_unlock(&ucb->mutex);
 783 }
 784
 785 static void __uprobe_trace_func(struct trace_uprobe *tu,
 786                                 unsigned long func, struct pt_regs *regs,
 787                                 struct uprobe_cpu_buffer *ucb, int dsize,
 788                                 struct trace_event_file *trace_file)
 789 {
 790         struct uprobe_trace_entry_head *entry;
 791         struct ring_buffer_event *event;
 792         struct ring_buffer *buffer;
 793         void *data;
 794         int size, esize;
 795         struct trace_event_call *call = &tu->tp.call;
 796
 797         WARN_ON(call != trace_file->event_call);
 798
 799         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 800                 return;
 801
 802         if (trace_trigger_soft_disabled(trace_file))
 803                 return;
 804
 805         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 806         size = esize + tu->tp.size + dsize;
 807         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 808                                                 call->event.type, size, 0, 0);
 809         if (!event)
 810                 return;
 811
 812         entry = ring_buffer_event_data(event);
 813         if (is_ret_probe(tu)) {
 814                 entry->vaddr[0] = func;
 815                 entry->vaddr[1] = instruction_pointer(regs);
 816                 data = DATAOF_TRACE_ENTRY(entry, true);
 817         } else {
 818                 entry->vaddr[0] = instruction_pointer(regs);
 819                 data = DATAOF_TRACE_ENTRY(entry, false);
 820         }
 821
 822         memcpy(data, ucb->buf, tu->tp.size + dsize);
 823
 824         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 825 }
 826
 827 /* uprobe handler */
 828 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 829                              struct uprobe_cpu_buffer *ucb, int dsize)
 830 {
 831         struct event_file_link *link;
 832
 833         if (is_ret_probe(tu))
 834                 return 0;
 835
 836         rcu_read_lock();
 837         list_for_each_entry_rcu(link, &tu->tp.files, list)
 838                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 839         rcu_read_unlock();
 840
 841         return 0;
 842 }
 843
 844 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 845                                  struct pt_regs *regs,
 846                                  struct uprobe_cpu_buffer *ucb, int dsize)
 847 {
 848         struct event_file_link *link;
 849
 850         rcu_read_lock();
 851         list_for_each_entry_rcu(link, &tu->tp.files, list)
 852                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 853         rcu_read_unlock();
 854 }
 855
 856 /* Event entry printers */
 857 static enum print_line_t
 858 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 859 {
 860         struct uprobe_trace_entry_head *entry;
 861         struct trace_seq *s = &iter->seq;
 862         struct trace_uprobe *tu;
 863         u8 *data;
 864         int i;
 865
 866         entry = (struct uprobe_trace_entry_head *)iter->ent;
 867         tu = container_of(event, struct trace_uprobe, tp.call.event);
 868
 869         if (is_ret_probe(tu)) {
 870                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 871                                  trace_event_name(&tu->tp.call),
 872                                  entry->vaddr[1], entry->vaddr[0]);
 873                 data = DATAOF_TRACE_ENTRY(entry, true);
 874         } else {
 875                 trace_seq_printf(s, "%s: (0x%lx)",
 876                                  trace_event_name(&tu->tp.call),
 877                                  entry->vaddr[0]);
 878                 data = DATAOF_TRACE_ENTRY(entry, false);
 879         }
 880
 881         for (i = 0; i < tu->tp.nr_args; i++) {
 882                 struct probe_arg *parg = &tu->tp.args[i];
 883
 884                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 885                         goto out;
 886         }
 887
 888         trace_seq_putc(s, '\n');
 889
 890  out:
 891         return trace_handle_return(s);
 892 }
 893
 894 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 895                                 enum uprobe_filter_ctx ctx,
 896                                 struct mm_struct *mm);
 897
 898 static int
 899 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
 900                    filter_func_t filter)
 901 {
 902         bool enabled = trace_probe_is_enabled(&tu->tp);
 903         struct event_file_link *link = NULL;
 904         int ret;
 905
 906         if (file) {
 907                 if (tu->tp.flags & TP_FLAG_PROFILE)
 908                         return -EINTR;
 909
 910                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 911                 if (!link)
 912                         return -ENOMEM;
 913
 914                 link->file = file;
 915                 list_add_tail_rcu(&link->list, &tu->tp.files);
 916
 917                 tu->tp.flags |= TP_FLAG_TRACE;
 918         } else {
 919                 if (tu->tp.flags & TP_FLAG_TRACE)
 920                         return -EINTR;
 921
 922                 tu->tp.flags |= TP_FLAG_PROFILE;
 923         }
 924
 925         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 926
 927         if (enabled)
 928                 return 0;
 929
 930         ret = uprobe_buffer_enable();
 931         if (ret)
 932                 goto err_flags;
 933
 934         tu->consumer.filter = filter;
 935         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 936         if (ret)
 937                 goto err_buffer;
 938
 939         return 0;
 940
 941  err_buffer:
 942         uprobe_buffer_disable();
 943
 944  err_flags:
 945         if (file) {
 946                 list_del(&link->list);
 947                 kfree(link);
 948                 tu->tp.flags &= ~TP_FLAG_TRACE;
 949         } else {
 950                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 951         }
 952         return ret;
 953 }
 954
 955 static void
 956 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 957 {
 958         if (!trace_probe_is_enabled(&tu->tp))
 959                 return;
 960
 961         if (file) {
 962                 struct event_file_link *link;
 963
 964                 link = find_event_file_link(&tu->tp, file);
 965                 if (!link)
 966                         return;
 967
 968                 list_del_rcu(&link->list);
 969                 /* synchronize with u{,ret}probe_trace_func */
 970                 synchronize_sched();
 971                 kfree(link);
 972
 973                 if (!list_empty(&tu->tp.files))
 974                         return;
 975         }
 976
 977         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 978
 979         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 980         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 981
 982         uprobe_buffer_disable();
 983 }
 984
 985 static int uprobe_event_define_fields(struct trace_event_call *event_call)
 986 {
 987         int ret, i, size;
 988         struct uprobe_trace_entry_head field;
 989         struct trace_uprobe *tu = event_call->data;
 990
 991         if (is_ret_probe(tu)) {
 992                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 993                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 994                 size = SIZEOF_TRACE_ENTRY(true);
 995         } else {
 996                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
 997                 size = SIZEOF_TRACE_ENTRY(false);
 998         }
 999         /* Set argument names as fields */
1000         for (i = 0; i < tu->tp.nr_args; i++) {
1001                 struct probe_arg *parg = &tu->tp.args[i];
1002
1003                 ret = trace_define_field(event_call, parg->type->fmttype,
1004                                          parg->name, size + parg->offset,
1005                                          parg->type->size, parg->type->is_signed,
1006                                          FILTER_OTHER);
1007
1008                 if (ret)
1009                         return ret;
1010         }
1011         return 0;
1012 }
1013
1014 #ifdef CONFIG_PERF_EVENTS
1015 static bool
1016 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1017 {
1018         struct perf_event *event;
1019
1020         if (filter->nr_systemwide)
1021                 return true;
1022
1023         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1024                 if (event->hw.target->mm == mm)
1025                         return true;
1026         }
1027
1028         return false;
1029 }
1030
1031 static inline bool
1032 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1033 {
1034         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1035 }
1036
1037 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1038 {
1039         bool done;
1040
1041         write_lock(&tu->filter.rwlock);
1042         if (event->hw.target) {
1043                 list_del(&event->hw.tp_list);
1044                 done = tu->filter.nr_systemwide ||
1045                         (event->hw.target->flags & PF_EXITING) ||
1046                         uprobe_filter_event(tu, event);
1047         } else {
1048                 tu->filter.nr_systemwide--;
1049                 done = tu->filter.nr_systemwide;
1050         }
1051         write_unlock(&tu->filter.rwlock);
1052
1053         if (!done)
1054                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1055
1056         return 0;
1057 }
1058
1059 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1060 {
1061         bool done;
1062         int err;
1063
1064         write_lock(&tu->filter.rwlock);
1065         if (event->hw.target) {
1066                 /*
1067                  * event->parent != NULL means copy_process(), we can avoid
1068                  * uprobe_apply(). current->mm must be probed and we can rely
1069                  * on dup_mmap() which preserves the already installed bp's.
1070                  *
1071                  * attr.enable_on_exec means that exec/mmap will install the
1072                  * breakpoints we need.
1073                  */
1074                 done = tu->filter.nr_systemwide ||
1075                         event->parent || event->attr.enable_on_exec ||
1076                         uprobe_filter_event(tu, event);
1077                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1078         } else {
1079                 done = tu->filter.nr_systemwide;
1080                 tu->filter.nr_systemwide++;
1081         }
1082         write_unlock(&tu->filter.rwlock);
1083
1084         err = 0;
1085         if (!done) {
1086                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1087                 if (err)
1088                         uprobe_perf_close(tu, event);
1089         }
1090         return err;
1091 }
1092
1093 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1094                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1095 {
1096         struct trace_uprobe *tu;
1097         int ret;
1098
1099         tu = container_of(uc, struct trace_uprobe, consumer);
1100         read_lock(&tu->filter.rwlock);
1101         ret = __uprobe_perf_filter(&tu->filter, mm);
1102         read_unlock(&tu->filter.rwlock);
1103
1104         return ret;
1105 }
1106
1107 static void __uprobe_perf_func(struct trace_uprobe *tu,
1108                                unsigned long func, struct pt_regs *regs,
1109                                struct uprobe_cpu_buffer *ucb, int dsize)
1110 {
1111         struct trace_event_call *call = &tu->tp.call;
1112         struct uprobe_trace_entry_head *entry;
1113         struct bpf_prog *prog = call->prog;
1114         struct hlist_head *head;
1115         void *data;
1116         int size, esize;
1117         int rctx;
1118
1119         if (prog && !trace_call_bpf(prog, regs))
1120                 return;
1121
1122         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1123
1124         size = esize + tu->tp.size + dsize;
1125         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1126         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1127                 return;
1128
1129         preempt_disable();
1130         head = this_cpu_ptr(call->perf_events);
1131         if (hlist_empty(head))
1132                 goto out;
1133
1134         entry = perf_trace_buf_prepare(size, call->event.type, NULL, &rctx);
1135         if (!entry)
1136                 goto out;
1137
1138         if (is_ret_probe(tu)) {
1139                 entry->vaddr[0] = func;
1140                 entry->vaddr[1] = instruction_pointer(regs);
1141                 data = DATAOF_TRACE_ENTRY(entry, true);
1142         } else {
1143                 entry->vaddr[0] = instruction_pointer(regs);
1144                 data = DATAOF_TRACE_ENTRY(entry, false);
1145         }
1146
1147         memcpy(data, ucb->buf, tu->tp.size + dsize);
1148
1149         if (size - esize > tu->tp.size + dsize) {
1150                 int len = tu->tp.size + dsize;
1151
1152                 memset(data + len, 0, size - esize - len);
1153         }
1154
1155         perf_trace_buf_submit(entry, size, rctx, 0, 1, regs, head, NULL);
1156  out:
1157         preempt_enable();
1158 }
1159
1160 /* uprobe profile handler */
1161 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1162                             struct uprobe_cpu_buffer *ucb, int dsize)
1163 {
1164         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1165                 return UPROBE_HANDLER_REMOVE;
1166
1167         if (!is_ret_probe(tu))
1168                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1169         return 0;
1170 }
1171
1172 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1173                                 struct pt_regs *regs,
1174                                 struct uprobe_cpu_buffer *ucb, int dsize)
1175 {
1176         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1177 }
1178 #endif  /* CONFIG_PERF_EVENTS */
1179
1180 static int
1181 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1182                       void *data)
1183 {
1184         struct trace_uprobe *tu = event->data;
1185         struct trace_event_file *file = data;
1186
1187         switch (type) {
1188         case TRACE_REG_REGISTER:
1189                 return probe_event_enable(tu, file, NULL);
1190
1191         case TRACE_REG_UNREGISTER:
1192                 probe_event_disable(tu, file);
1193                 return 0;
1194
1195 #ifdef CONFIG_PERF_EVENTS
1196         case TRACE_REG_PERF_REGISTER:
1197                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1198
1199         case TRACE_REG_PERF_UNREGISTER:
1200                 probe_event_disable(tu, NULL);
1201                 return 0;
1202
1203         case TRACE_REG_PERF_OPEN:
1204                 return uprobe_perf_open(tu, data);
1205
1206         case TRACE_REG_PERF_CLOSE:
1207                 return uprobe_perf_close(tu, data);
1208
1209 #endif
1210         default:
1211                 return 0;
1212         }
1213         return 0;
1214 }
1215
1216 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1217 {
1218         struct trace_uprobe *tu;
1219         struct uprobe_dispatch_data udd;
1220         struct uprobe_cpu_buffer *ucb;
1221         int dsize, esize;
1222         int ret = 0;
1223
1224
1225         tu = container_of(con, struct trace_uprobe, consumer);
1226         tu->nhit++;
1227
1228         udd.tu = tu;
1229         udd.bp_addr = instruction_pointer(regs);
1230
1231         current->utask->vaddr = (unsigned long) &udd;
1232
1233         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1234                 return 0;
1235
1236         dsize = __get_data_size(&tu->tp, regs);
1237         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1238
1239         ucb = uprobe_buffer_get();
1240         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1241
1242         if (tu->tp.flags & TP_FLAG_TRACE)
1243                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1244
1245 #ifdef CONFIG_PERF_EVENTS
1246         if (tu->tp.flags & TP_FLAG_PROFILE)
1247                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1248 #endif
1249         uprobe_buffer_put(ucb);
1250         return ret;
1251 }
1252
1253 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1254                                 unsigned long func, struct pt_regs *regs)
1255 {
1256         struct trace_uprobe *tu;
1257         struct uprobe_dispatch_data udd;
1258         struct uprobe_cpu_buffer *ucb;
1259         int dsize, esize;
1260
1261         tu = container_of(con, struct trace_uprobe, consumer);
1262
1263         udd.tu = tu;
1264         udd.bp_addr = func;
1265
1266         current->utask->vaddr = (unsigned long) &udd;
1267
1268         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1269                 return 0;
1270
1271         dsize = __get_data_size(&tu->tp, regs);
1272         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1273
1274         ucb = uprobe_buffer_get();
1275         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1276
1277         if (tu->tp.flags & TP_FLAG_TRACE)
1278                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1279
1280 #ifdef CONFIG_PERF_EVENTS
1281         if (tu->tp.flags & TP_FLAG_PROFILE)
1282                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1283 #endif
1284         uprobe_buffer_put(ucb);
1285         return 0;
1286 }
1287
1288 static struct trace_event_functions uprobe_funcs = {
1289         .trace          = print_uprobe_event
1290 };
1291
1292 static int register_uprobe_event(struct trace_uprobe *tu)
1293 {
1294         struct trace_event_call *call = &tu->tp.call;
1295         int ret;
1296
1297         /* Initialize trace_event_call */
1298         INIT_LIST_HEAD(&call->class->fields);
1299         call->event.funcs = &uprobe_funcs;
1300         call->class->define_fields = uprobe_event_define_fields;
1301
1302         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1303                 return -ENOMEM;
1304
1305         ret = register_trace_event(&call->event);
1306         if (!ret) {
1307                 kfree(call->print_fmt);
1308                 return -ENODEV;
1309         }
1310
1311         call->flags = TRACE_EVENT_FL_UPROBE;
1312         call->class->reg = trace_uprobe_register;
1313         call->data = tu;
1314         ret = trace_add_event_call(call);
1315
1316         if (ret) {
1317                 pr_info("Failed to register uprobe event: %s\n",
1318                         trace_event_name(call));
1319                 kfree(call->print_fmt);
1320                 unregister_trace_event(&call->event);
1321         }
1322
1323         return ret;
1324 }
1325
1326 static int unregister_uprobe_event(struct trace_uprobe *tu)
1327 {
1328         int ret;
1329
1330         /* tu->event is unregistered in trace_remove_event_call() */
1331         ret = trace_remove_event_call(&tu->tp.call);
1332         if (ret)
1333                 return ret;
1334         kfree(tu->tp.call.print_fmt);
1335         tu->tp.call.print_fmt = NULL;
1336         return 0;
1337 }
1338
1339 /* Make a trace interface for controling probe points */
1340 static __init int init_uprobe_trace(void)
1341 {
1342         struct dentry *d_tracer;
1343
1344         d_tracer = tracing_init_dentry();
1345         if (IS_ERR(d_tracer))
1346                 return 0;
1347
1348         trace_create_file("uprobe_events", 0644, d_tracer,
1349                                     NULL, &uprobe_events_ops);
1350         /* Profile interface */
1351         trace_create_file("uprobe_profile", 0444, d_tracer,
1352                                     NULL, &uprobe_profile_ops);
1353         return 0;
1354 }
1355
1356 fs_initcall(init_uprobe_trace);