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         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 215         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 216         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 217         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 218
 219         ASSIGN_FETCH_TYPE_END
 220 };
 221
 222 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 223 {
 224         rwlock_init(&filter->rwlock);
 225         filter->nr_systemwide = 0;
 226         INIT_LIST_HEAD(&filter->perf_events);
 227 }
 228
 229 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 230 {
 231         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 232 }
 233
 234 static inline bool is_ret_probe(struct trace_uprobe *tu)
 235 {
 236         return tu->consumer.ret_handler != NULL;
 237 }
 238
 239 /*
 240  * Allocate new trace_uprobe and initialize it (including uprobes).
 241  */
 242 static struct trace_uprobe *
 243 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 244 {
 245         struct trace_uprobe *tu;
 246
 247         if (!event || !is_good_name(event))
 248                 return ERR_PTR(-EINVAL);
 249
 250         if (!group || !is_good_name(group))
 251                 return ERR_PTR(-EINVAL);
 252
 253         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 254         if (!tu)
 255                 return ERR_PTR(-ENOMEM);
 256
 257         tu->tp.call.class = &tu->tp.class;
 258         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 259         if (!tu->tp.call.name)
 260                 goto error;
 261
 262         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 263         if (!tu->tp.class.system)
 264                 goto error;
 265
 266         INIT_LIST_HEAD(&tu->list);
 267         INIT_LIST_HEAD(&tu->tp.files);
 268         tu->consumer.handler = uprobe_dispatcher;
 269         if (is_ret)
 270                 tu->consumer.ret_handler = uretprobe_dispatcher;
 271         init_trace_uprobe_filter(&tu->filter);
 272         return tu;
 273
 274 error:
 275         kfree(tu->tp.call.name);
 276         kfree(tu);
 277
 278         return ERR_PTR(-ENOMEM);
 279 }
 280
 281 static void free_trace_uprobe(struct trace_uprobe *tu)
 282 {
 283         int i;
 284
 285         for (i = 0; i < tu->tp.nr_args; i++)
 286                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 287
 288         iput(tu->inode);
 289         kfree(tu->tp.call.class->system);
 290         kfree(tu->tp.call.name);
 291         kfree(tu->filename);
 292         kfree(tu);
 293 }
 294
 295 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 296 {
 297         struct trace_uprobe *tu;
 298
 299         list_for_each_entry(tu, &uprobe_list, list)
 300                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 301                     strcmp(tu->tp.call.class->system, group) == 0)
 302                         return tu;
 303
 304         return NULL;
 305 }
 306
 307 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 308 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 309 {
 310         int ret;
 311
 312         ret = unregister_uprobe_event(tu);
 313         if (ret)
 314                 return ret;
 315
 316         list_del(&tu->list);
 317         free_trace_uprobe(tu);
 318         return 0;
 319 }
 320
 321 /* Register a trace_uprobe and probe_event */
 322 static int register_trace_uprobe(struct trace_uprobe *tu)
 323 {
 324         struct trace_uprobe *old_tu;
 325         int ret;
 326
 327         mutex_lock(&uprobe_lock);
 328
 329         /* register as an event */
 330         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
 331                         tu->tp.call.class->system);
 332         if (old_tu) {
 333                 /* delete old event */
 334                 ret = unregister_trace_uprobe(old_tu);
 335                 if (ret)
 336                         goto end;
 337         }
 338
 339         ret = register_uprobe_event(tu);
 340         if (ret) {
 341                 pr_warn("Failed to register probe event(%d)\n", ret);
 342                 goto end;
 343         }
 344
 345         list_add_tail(&tu->list, &uprobe_list);
 346
 347 end:
 348         mutex_unlock(&uprobe_lock);
 349
 350         return ret;
 351 }
 352
 353 /*
 354  * Argument syntax:
 355  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 356  *
 357  *  - Remove uprobe: -:[GRP/]EVENT
 358  */
 359 static int create_trace_uprobe(int argc, char **argv)
 360 {
 361         struct trace_uprobe *tu;
 362         struct inode *inode;
 363         char *arg, *event, *group, *filename;
 364         char buf[MAX_EVENT_NAME_LEN];
 365         struct path path;
 366         unsigned long offset;
 367         bool is_delete, is_return;
 368         int i, ret;
 369
 370         inode = NULL;
 371         ret = 0;
 372         is_delete = false;
 373         is_return = false;
 374         event = NULL;
 375         group = NULL;
 376
 377         /* argc must be >= 1 */
 378         if (argv[0][0] == '-')
 379                 is_delete = true;
 380         else if (argv[0][0] == 'r')
 381                 is_return = true;
 382         else if (argv[0][0] != 'p') {
 383                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 384                 return -EINVAL;
 385         }
 386
 387         if (argv[0][1] == ':') {
 388                 event = &argv[0][2];
 389                 arg = strchr(event, '/');
 390
 391                 if (arg) {
 392                         group = event;
 393                         event = arg + 1;
 394                         event[-1] = '\0';
 395
 396                         if (strlen(group) == 0) {
 397                                 pr_info("Group name is not specified\n");
 398                                 return -EINVAL;
 399                         }
 400                 }
 401                 if (strlen(event) == 0) {
 402                         pr_info("Event name is not specified\n");
 403                         return -EINVAL;
 404                 }
 405         }
 406         if (!group)
 407                 group = UPROBE_EVENT_SYSTEM;
 408
 409         if (is_delete) {
 410                 int ret;
 411
 412                 if (!event) {
 413                         pr_info("Delete command needs an event name.\n");
 414                         return -EINVAL;
 415                 }
 416                 mutex_lock(&uprobe_lock);
 417                 tu = find_probe_event(event, group);
 418
 419                 if (!tu) {
 420                         mutex_unlock(&uprobe_lock);
 421                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 422                         return -ENOENT;
 423                 }
 424                 /* delete an event */
 425                 ret = unregister_trace_uprobe(tu);
 426                 mutex_unlock(&uprobe_lock);
 427                 return ret;
 428         }
 429
 430         if (argc < 2) {
 431                 pr_info("Probe point is not specified.\n");
 432                 return -EINVAL;
 433         }
 434         if (isdigit(argv[1][0])) {
 435                 pr_info("probe point must be have a filename.\n");
 436                 return -EINVAL;
 437         }
 438         arg = strchr(argv[1], ':');
 439         if (!arg) {
 440                 ret = -EINVAL;
 441                 goto fail_address_parse;
 442         }
 443
 444         *arg++ = '\0';
 445         filename = argv[1];
 446         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 447         if (ret)
 448                 goto fail_address_parse;
 449
 450         inode = igrab(d_inode(path.dentry));
 451         path_put(&path);
 452
 453         if (!inode || !S_ISREG(inode->i_mode)) {
 454                 ret = -EINVAL;
 455                 goto fail_address_parse;
 456         }
 457
 458         ret = kstrtoul(arg, 0, &offset);
 459         if (ret)
 460                 goto fail_address_parse;
 461
 462         argc -= 2;
 463         argv += 2;
 464
 465         /* setup a probe */
 466         if (!event) {
 467                 char *tail;
 468                 char *ptr;
 469
 470                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 471                 if (!tail) {
 472                         ret = -ENOMEM;
 473                         goto fail_address_parse;
 474                 }
 475
 476                 ptr = strpbrk(tail, ".-_");
 477                 if (ptr)
 478                         *ptr = '\0';
 479
 480                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 481                 event = buf;
 482                 kfree(tail);
 483         }
 484
 485         tu = alloc_trace_uprobe(group, event, argc, is_return);
 486         if (IS_ERR(tu)) {
 487                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 488                 ret = PTR_ERR(tu);
 489                 goto fail_address_parse;
 490         }
 491         tu->offset = offset;
 492         tu->inode = inode;
 493         tu->filename = kstrdup(filename, GFP_KERNEL);
 494
 495         if (!tu->filename) {
 496                 pr_info("Failed to allocate filename.\n");
 497                 ret = -ENOMEM;
 498                 goto error;
 499         }
 500
 501         /* parse arguments */
 502         ret = 0;
 503         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 504                 struct probe_arg *parg = &tu->tp.args[i];
 505
 506                 /* Increment count for freeing args in error case */
 507                 tu->tp.nr_args++;
 508
 509                 /* Parse argument name */
 510                 arg = strchr(argv[i], '=');
 511                 if (arg) {
 512                         *arg++ = '\0';
 513                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 514                 } else {
 515                         arg = argv[i];
 516                         /* If argument name is omitted, set "argN" */
 517                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 518                         parg->name = kstrdup(buf, GFP_KERNEL);
 519                 }
 520
 521                 if (!parg->name) {
 522                         pr_info("Failed to allocate argument[%d] name.\n", i);
 523                         ret = -ENOMEM;
 524                         goto error;
 525                 }
 526
 527                 if (!is_good_name(parg->name)) {
 528                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 529                         ret = -EINVAL;
 530                         goto error;
 531                 }
 532
 533                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 534                         pr_info("Argument[%d] name '%s' conflicts with "
 535                                 "another field.\n", i, argv[i]);
 536                         ret = -EINVAL;
 537                         goto error;
 538                 }
 539
 540                 /* Parse fetch argument */
 541                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 542                                                  is_return, false,
 543                                                  uprobes_fetch_type_table);
 544                 if (ret) {
 545                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 546                         goto error;
 547                 }
 548         }
 549
 550         ret = register_trace_uprobe(tu);
 551         if (ret)
 552                 goto error;
 553         return 0;
 554
 555 error:
 556         free_trace_uprobe(tu);
 557         return ret;
 558
 559 fail_address_parse:
 560         iput(inode);
 561
 562         pr_info("Failed to parse address or file.\n");
 563
 564         return ret;
 565 }
 566
 567 static int cleanup_all_probes(void)
 568 {
 569         struct trace_uprobe *tu;
 570         int ret = 0;
 571
 572         mutex_lock(&uprobe_lock);
 573         while (!list_empty(&uprobe_list)) {
 574                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 575                 ret = unregister_trace_uprobe(tu);
 576                 if (ret)
 577                         break;
 578         }
 579         mutex_unlock(&uprobe_lock);
 580         return ret;
 581 }
 582
 583 /* Probes listing interfaces */
 584 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 585 {
 586         mutex_lock(&uprobe_lock);
 587         return seq_list_start(&uprobe_list, *pos);
 588 }
 589
 590 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 591 {
 592         return seq_list_next(v, &uprobe_list, pos);
 593 }
 594
 595 static void probes_seq_stop(struct seq_file *m, void *v)
 596 {
 597         mutex_unlock(&uprobe_lock);
 598 }
 599
 600 static int probes_seq_show(struct seq_file *m, void *v)
 601 {
 602         struct trace_uprobe *tu = v;
 603         char c = is_ret_probe(tu) ? 'r' : 'p';
 604         int i;
 605
 606         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 607                         trace_event_name(&tu->tp.call));
 608         seq_printf(m, " %s:", tu->filename);
 609
 610         /* Don't print "0x  (null)" when offset is 0 */
 611         if (tu->offset) {
 612                 seq_printf(m, "0x%p", (void *)tu->offset);
 613         } else {
 614                 switch (sizeof(void *)) {
 615                 case 4:
 616                         seq_printf(m, "0x00000000");
 617                         break;
 618                 case 8:
 619                 default:
 620                         seq_printf(m, "0x0000000000000000");
 621                         break;
 622                 }
 623         }
 624
 625         for (i = 0; i < tu->tp.nr_args; i++)
 626                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 627
 628         seq_putc(m, '\n');
 629         return 0;
 630 }
 631
 632 static const struct seq_operations probes_seq_op = {
 633         .start  = probes_seq_start,
 634         .next   = probes_seq_next,
 635         .stop   = probes_seq_stop,
 636         .show   = probes_seq_show
 637 };
 638
 639 static int probes_open(struct inode *inode, struct file *file)
 640 {
 641         int ret;
 642
 643         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 644                 ret = cleanup_all_probes();
 645                 if (ret)
 646                         return ret;
 647         }
 648
 649         return seq_open(file, &probes_seq_op);
 650 }
 651
 652 static ssize_t probes_write(struct file *file, const char __user *buffer,
 653                             size_t count, loff_t *ppos)
 654 {
 655         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 656 }
 657
 658 static const struct file_operations uprobe_events_ops = {
 659         .owner          = THIS_MODULE,
 660         .open           = probes_open,
 661         .read           = seq_read,
 662         .llseek         = seq_lseek,
 663         .release        = seq_release,
 664         .write          = probes_write,
 665 };
 666
 667 /* Probes profiling interfaces */
 668 static int probes_profile_seq_show(struct seq_file *m, void *v)
 669 {
 670         struct trace_uprobe *tu = v;
 671
 672         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 673                         trace_event_name(&tu->tp.call), tu->nhit);
 674         return 0;
 675 }
 676
 677 static const struct seq_operations profile_seq_op = {
 678         .start  = probes_seq_start,
 679         .next   = probes_seq_next,
 680         .stop   = probes_seq_stop,
 681         .show   = probes_profile_seq_show
 682 };
 683
 684 static int profile_open(struct inode *inode, struct file *file)
 685 {
 686         return seq_open(file, &profile_seq_op);
 687 }
 688
 689 static const struct file_operations uprobe_profile_ops = {
 690         .owner          = THIS_MODULE,
 691         .open           = profile_open,
 692         .read           = seq_read,
 693         .llseek         = seq_lseek,
 694         .release        = seq_release,
 695 };
 696
 697 struct uprobe_cpu_buffer {
 698         struct mutex mutex;
 699         void *buf;
 700 };
 701 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 702 static int uprobe_buffer_refcnt;
 703
 704 static int uprobe_buffer_init(void)
 705 {
 706         int cpu, err_cpu;
 707
 708         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 709         if (uprobe_cpu_buffer == NULL)
 710                 return -ENOMEM;
 711
 712         for_each_possible_cpu(cpu) {
 713                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 714                                                   GFP_KERNEL, 0);
 715                 if (p == NULL) {
 716                         err_cpu = cpu;
 717                         goto err;
 718                 }
 719                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 720                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 721         }
 722
 723         return 0;
 724
 725 err:
 726         for_each_possible_cpu(cpu) {
 727                 if (cpu == err_cpu)
 728                         break;
 729                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 730         }
 731
 732         free_percpu(uprobe_cpu_buffer);
 733         return -ENOMEM;
 734 }
 735
 736 static int uprobe_buffer_enable(void)
 737 {
 738         int ret = 0;
 739
 740         BUG_ON(!mutex_is_locked(&event_mutex));
 741
 742         if (uprobe_buffer_refcnt++ == 0) {
 743                 ret = uprobe_buffer_init();
 744                 if (ret < 0)
 745                         uprobe_buffer_refcnt--;
 746         }
 747
 748         return ret;
 749 }
 750
 751 static void uprobe_buffer_disable(void)
 752 {
 753         int cpu;
 754
 755         BUG_ON(!mutex_is_locked(&event_mutex));
 756
 757         if (--uprobe_buffer_refcnt == 0) {
 758                 for_each_possible_cpu(cpu)
 759                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 760                                                              cpu)->buf);
 761
 762                 free_percpu(uprobe_cpu_buffer);
 763                 uprobe_cpu_buffer = NULL;
 764         }
 765 }
 766
 767 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 768 {
 769         struct uprobe_cpu_buffer *ucb;
 770         int cpu;
 771
 772         cpu = raw_smp_processor_id();
 773         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 774
 775         /*
 776          * Use per-cpu buffers for fastest access, but we might migrate
 777          * so the mutex makes sure we have sole access to it.
 778          */
 779         mutex_lock(&ucb->mutex);
 780
 781         return ucb;
 782 }
 783
 784 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 785 {
 786         mutex_unlock(&ucb->mutex);
 787 }
 788
 789 static void __uprobe_trace_func(struct trace_uprobe *tu,
 790                                 unsigned long func, struct pt_regs *regs,
 791                                 struct uprobe_cpu_buffer *ucb, int dsize,
 792                                 struct trace_event_file *trace_file)
 793 {
 794         struct uprobe_trace_entry_head *entry;
 795         struct ring_buffer_event *event;
 796         struct ring_buffer *buffer;
 797         void *data;
 798         int size, esize;
 799         struct trace_event_call *call = &tu->tp.call;
 800
 801         WARN_ON(call != trace_file->event_call);
 802
 803         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 804                 return;
 805
 806         if (trace_trigger_soft_disabled(trace_file))
 807                 return;
 808
 809         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 810         size = esize + tu->tp.size + dsize;
 811         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 812                                                 call->event.type, size, 0, 0);
 813         if (!event)
 814                 return;
 815
 816         entry = ring_buffer_event_data(event);
 817         if (is_ret_probe(tu)) {
 818                 entry->vaddr[0] = func;
 819                 entry->vaddr[1] = instruction_pointer(regs);
 820                 data = DATAOF_TRACE_ENTRY(entry, true);
 821         } else {
 822                 entry->vaddr[0] = instruction_pointer(regs);
 823                 data = DATAOF_TRACE_ENTRY(entry, false);
 824         }
 825
 826         memcpy(data, ucb->buf, tu->tp.size + dsize);
 827
 828         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 829 }
 830
 831 /* uprobe handler */
 832 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 833                              struct uprobe_cpu_buffer *ucb, int dsize)
 834 {
 835         struct event_file_link *link;
 836
 837         if (is_ret_probe(tu))
 838                 return 0;
 839
 840         rcu_read_lock();
 841         list_for_each_entry_rcu(link, &tu->tp.files, list)
 842                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 843         rcu_read_unlock();
 844
 845         return 0;
 846 }
 847
 848 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 849                                  struct pt_regs *regs,
 850                                  struct uprobe_cpu_buffer *ucb, int dsize)
 851 {
 852         struct event_file_link *link;
 853
 854         rcu_read_lock();
 855         list_for_each_entry_rcu(link, &tu->tp.files, list)
 856                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 857         rcu_read_unlock();
 858 }
 859
 860 /* Event entry printers */
 861 static enum print_line_t
 862 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 863 {
 864         struct uprobe_trace_entry_head *entry;
 865         struct trace_seq *s = &iter->seq;
 866         struct trace_uprobe *tu;
 867         u8 *data;
 868         int i;
 869
 870         entry = (struct uprobe_trace_entry_head *)iter->ent;
 871         tu = container_of(event, struct trace_uprobe, tp.call.event);
 872
 873         if (is_ret_probe(tu)) {
 874                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 875                                  trace_event_name(&tu->tp.call),
 876                                  entry->vaddr[1], entry->vaddr[0]);
 877                 data = DATAOF_TRACE_ENTRY(entry, true);
 878         } else {
 879                 trace_seq_printf(s, "%s: (0x%lx)",
 880                                  trace_event_name(&tu->tp.call),
 881                                  entry->vaddr[0]);
 882                 data = DATAOF_TRACE_ENTRY(entry, false);
 883         }
 884
 885         for (i = 0; i < tu->tp.nr_args; i++) {
 886                 struct probe_arg *parg = &tu->tp.args[i];
 887
 888                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 889                         goto out;
 890         }
 891
 892         trace_seq_putc(s, '\n');
 893
 894  out:
 895         return trace_handle_return(s);
 896 }
 897
 898 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 899                                 enum uprobe_filter_ctx ctx,
 900                                 struct mm_struct *mm);
 901
 902 static int
 903 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
 904                    filter_func_t filter)
 905 {
 906         bool enabled = trace_probe_is_enabled(&tu->tp);
 907         struct event_file_link *link = NULL;
 908         int ret;
 909
 910         if (file) {
 911                 if (tu->tp.flags & TP_FLAG_PROFILE)
 912                         return -EINTR;
 913
 914                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 915                 if (!link)
 916                         return -ENOMEM;
 917
 918                 link->file = file;
 919                 list_add_tail_rcu(&link->list, &tu->tp.files);
 920
 921                 tu->tp.flags |= TP_FLAG_TRACE;
 922         } else {
 923                 if (tu->tp.flags & TP_FLAG_TRACE)
 924                         return -EINTR;
 925
 926                 tu->tp.flags |= TP_FLAG_PROFILE;
 927         }
 928
 929         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 930
 931         if (enabled)
 932                 return 0;
 933
 934         ret = uprobe_buffer_enable();
 935         if (ret)
 936                 goto err_flags;
 937
 938         tu->consumer.filter = filter;
 939         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 940         if (ret)
 941                 goto err_buffer;
 942
 943         return 0;
 944
 945  err_buffer:
 946         uprobe_buffer_disable();
 947
 948  err_flags:
 949         if (file) {
 950                 list_del(&link->list);
 951                 kfree(link);
 952                 tu->tp.flags &= ~TP_FLAG_TRACE;
 953         } else {
 954                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 955         }
 956         return ret;
 957 }
 958
 959 static void
 960 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 961 {
 962         if (!trace_probe_is_enabled(&tu->tp))
 963                 return;
 964
 965         if (file) {
 966                 struct event_file_link *link;
 967
 968                 link = find_event_file_link(&tu->tp, file);
 969                 if (!link)
 970                         return;
 971
 972                 list_del_rcu(&link->list);
 973                 /* synchronize with u{,ret}probe_trace_func */
 974                 synchronize_sched();
 975                 kfree(link);
 976
 977                 if (!list_empty(&tu->tp.files))
 978                         return;
 979         }
 980
 981         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 982
 983         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 984         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 985
 986         uprobe_buffer_disable();
 987 }
 988
 989 static int uprobe_event_define_fields(struct trace_event_call *event_call)
 990 {
 991         int ret, i, size;
 992         struct uprobe_trace_entry_head field;
 993         struct trace_uprobe *tu = event_call->data;
 994
 995         if (is_ret_probe(tu)) {
 996                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 997                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 998                 size = SIZEOF_TRACE_ENTRY(true);
 999         } else {
1000                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1001                 size = SIZEOF_TRACE_ENTRY(false);
1002         }
1003         /* Set argument names as fields */
1004         for (i = 0; i < tu->tp.nr_args; i++) {
1005                 struct probe_arg *parg = &tu->tp.args[i];
1006
1007                 ret = trace_define_field(event_call, parg->type->fmttype,
1008                                          parg->name, size + parg->offset,
1009                                          parg->type->size, parg->type->is_signed,
1010                                          FILTER_OTHER);
1011
1012                 if (ret)
1013                         return ret;
1014         }
1015         return 0;
1016 }
1017
1018 #ifdef CONFIG_PERF_EVENTS
1019 static bool
1020 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1021 {
1022         struct perf_event *event;
1023
1024         if (filter->nr_systemwide)
1025                 return true;
1026
1027         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1028                 if (event->hw.target->mm == mm)
1029                         return true;
1030         }
1031
1032         return false;
1033 }
1034
1035 static inline bool
1036 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1037 {
1038         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1039 }
1040
1041 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1042 {
1043         bool done;
1044
1045         write_lock(&tu->filter.rwlock);
1046         if (event->hw.target) {
1047                 list_del(&event->hw.tp_list);
1048                 done = tu->filter.nr_systemwide ||
1049                         (event->hw.target->flags & PF_EXITING) ||
1050                         uprobe_filter_event(tu, event);
1051         } else {
1052                 tu->filter.nr_systemwide--;
1053                 done = tu->filter.nr_systemwide;
1054         }
1055         write_unlock(&tu->filter.rwlock);
1056
1057         if (!done)
1058                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1059
1060         return 0;
1061 }
1062
1063 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1064 {
1065         bool done;
1066         int err;
1067
1068         write_lock(&tu->filter.rwlock);
1069         if (event->hw.target) {
1070                 /*
1071                  * event->parent != NULL means copy_process(), we can avoid
1072                  * uprobe_apply(). current->mm must be probed and we can rely
1073                  * on dup_mmap() which preserves the already installed bp's.
1074                  *
1075                  * attr.enable_on_exec means that exec/mmap will install the
1076                  * breakpoints we need.
1077                  */
1078                 done = tu->filter.nr_systemwide ||
1079                         event->parent || event->attr.enable_on_exec ||
1080                         uprobe_filter_event(tu, event);
1081                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1082         } else {
1083                 done = tu->filter.nr_systemwide;
1084                 tu->filter.nr_systemwide++;
1085         }
1086         write_unlock(&tu->filter.rwlock);
1087
1088         err = 0;
1089         if (!done) {
1090                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1091                 if (err)
1092                         uprobe_perf_close(tu, event);
1093         }
1094         return err;
1095 }
1096
1097 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1098                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1099 {
1100         struct trace_uprobe *tu;
1101         int ret;
1102
1103         tu = container_of(uc, struct trace_uprobe, consumer);
1104         read_lock(&tu->filter.rwlock);
1105         ret = __uprobe_perf_filter(&tu->filter, mm);
1106         read_unlock(&tu->filter.rwlock);
1107
1108         return ret;
1109 }
1110
1111 static void __uprobe_perf_func(struct trace_uprobe *tu,
1112                                unsigned long func, struct pt_regs *regs,
1113                                struct uprobe_cpu_buffer *ucb, int dsize)
1114 {
1115         struct trace_event_call *call = &tu->tp.call;
1116         struct uprobe_trace_entry_head *entry;
1117         struct bpf_prog *prog = call->prog;
1118         struct hlist_head *head;
1119         void *data;
1120         int size, esize;
1121         int rctx;
1122
1123         if (prog && !trace_call_bpf(prog, regs))
1124                 return;
1125
1126         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1127
1128         size = esize + tu->tp.size + dsize;
1129         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1130         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1131                 return;
1132
1133         preempt_disable();
1134         head = this_cpu_ptr(call->perf_events);
1135         if (hlist_empty(head))
1136                 goto out;
1137
1138         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1139         if (!entry)
1140                 goto out;
1141
1142         if (is_ret_probe(tu)) {
1143                 entry->vaddr[0] = func;
1144                 entry->vaddr[1] = instruction_pointer(regs);
1145                 data = DATAOF_TRACE_ENTRY(entry, true);
1146         } else {
1147                 entry->vaddr[0] = instruction_pointer(regs);
1148                 data = DATAOF_TRACE_ENTRY(entry, false);
1149         }
1150
1151         memcpy(data, ucb->buf, tu->tp.size + dsize);
1152
1153         if (size - esize > tu->tp.size + dsize) {
1154                 int len = tu->tp.size + dsize;
1155
1156                 memset(data + len, 0, size - esize - len);
1157         }
1158
1159         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1160                               head, NULL);
1161  out:
1162         preempt_enable();
1163 }
1164
1165 /* uprobe profile handler */
1166 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1167                             struct uprobe_cpu_buffer *ucb, int dsize)
1168 {
1169         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1170                 return UPROBE_HANDLER_REMOVE;
1171
1172         if (!is_ret_probe(tu))
1173                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1174         return 0;
1175 }
1176
1177 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1178                                 struct pt_regs *regs,
1179                                 struct uprobe_cpu_buffer *ucb, int dsize)
1180 {
1181         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1182 }
1183 #endif  /* CONFIG_PERF_EVENTS */
1184
1185 static int
1186 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1187                       void *data)
1188 {
1189         struct trace_uprobe *tu = event->data;
1190         struct trace_event_file *file = data;
1191
1192         switch (type) {
1193         case TRACE_REG_REGISTER:
1194                 return probe_event_enable(tu, file, NULL);
1195
1196         case TRACE_REG_UNREGISTER:
1197                 probe_event_disable(tu, file);
1198                 return 0;
1199
1200 #ifdef CONFIG_PERF_EVENTS
1201         case TRACE_REG_PERF_REGISTER:
1202                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1203
1204         case TRACE_REG_PERF_UNREGISTER:
1205                 probe_event_disable(tu, NULL);
1206                 return 0;
1207
1208         case TRACE_REG_PERF_OPEN:
1209                 return uprobe_perf_open(tu, data);
1210
1211         case TRACE_REG_PERF_CLOSE:
1212                 return uprobe_perf_close(tu, data);
1213
1214 #endif
1215         default:
1216                 return 0;
1217         }
1218         return 0;
1219 }
1220
1221 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1222 {
1223         struct trace_uprobe *tu;
1224         struct uprobe_dispatch_data udd;
1225         struct uprobe_cpu_buffer *ucb;
1226         int dsize, esize;
1227         int ret = 0;
1228
1229
1230         tu = container_of(con, struct trace_uprobe, consumer);
1231         tu->nhit++;
1232
1233         udd.tu = tu;
1234         udd.bp_addr = instruction_pointer(regs);
1235
1236         current->utask->vaddr = (unsigned long) &udd;
1237
1238         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1239                 return 0;
1240
1241         dsize = __get_data_size(&tu->tp, regs);
1242         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1243
1244         ucb = uprobe_buffer_get();
1245         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1246
1247         if (tu->tp.flags & TP_FLAG_TRACE)
1248                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1249
1250 #ifdef CONFIG_PERF_EVENTS
1251         if (tu->tp.flags & TP_FLAG_PROFILE)
1252                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1253 #endif
1254         uprobe_buffer_put(ucb);
1255         return ret;
1256 }
1257
1258 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1259                                 unsigned long func, struct pt_regs *regs)
1260 {
1261         struct trace_uprobe *tu;
1262         struct uprobe_dispatch_data udd;
1263         struct uprobe_cpu_buffer *ucb;
1264         int dsize, esize;
1265
1266         tu = container_of(con, struct trace_uprobe, consumer);
1267
1268         udd.tu = tu;
1269         udd.bp_addr = func;
1270
1271         current->utask->vaddr = (unsigned long) &udd;
1272
1273         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1274                 return 0;
1275
1276         dsize = __get_data_size(&tu->tp, regs);
1277         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1278
1279         ucb = uprobe_buffer_get();
1280         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1281
1282         if (tu->tp.flags & TP_FLAG_TRACE)
1283                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1284
1285 #ifdef CONFIG_PERF_EVENTS
1286         if (tu->tp.flags & TP_FLAG_PROFILE)
1287                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1288 #endif
1289         uprobe_buffer_put(ucb);
1290         return 0;
1291 }
1292
1293 static struct trace_event_functions uprobe_funcs = {
1294         .trace          = print_uprobe_event
1295 };
1296
1297 static int register_uprobe_event(struct trace_uprobe *tu)
1298 {
1299         struct trace_event_call *call = &tu->tp.call;
1300         int ret;
1301
1302         /* Initialize trace_event_call */
1303         INIT_LIST_HEAD(&call->class->fields);
1304         call->event.funcs = &uprobe_funcs;
1305         call->class->define_fields = uprobe_event_define_fields;
1306
1307         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1308                 return -ENOMEM;
1309
1310         ret = register_trace_event(&call->event);
1311         if (!ret) {
1312                 kfree(call->print_fmt);
1313                 return -ENODEV;
1314         }
1315
1316         call->flags = TRACE_EVENT_FL_UPROBE;
1317         call->class->reg = trace_uprobe_register;
1318         call->data = tu;
1319         ret = trace_add_event_call(call);
1320
1321         if (ret) {
1322                 pr_info("Failed to register uprobe event: %s\n",
1323                         trace_event_name(call));
1324                 kfree(call->print_fmt);
1325                 unregister_trace_event(&call->event);
1326         }
1327
1328         return ret;
1329 }
1330
1331 static int unregister_uprobe_event(struct trace_uprobe *tu)
1332 {
1333         int ret;
1334
1335         /* tu->event is unregistered in trace_remove_event_call() */
1336         ret = trace_remove_event_call(&tu->tp.call);
1337         if (ret)
1338                 return ret;
1339         kfree(tu->tp.call.print_fmt);
1340         tu->tp.call.print_fmt = NULL;
1341         return 0;
1342 }
1343
1344 /* Make a trace interface for controling probe points */
1345 static __init int init_uprobe_trace(void)
1346 {
1347         struct dentry *d_tracer;
1348
1349         d_tracer = tracing_init_dentry();
1350         if (IS_ERR(d_tracer))
1351                 return 0;
1352
1353         trace_create_file("uprobe_events", 0644, d_tracer,
1354                                     NULL, &uprobe_events_ops);
1355         /* Profile interface */
1356         trace_create_file("uprobe_profile", 0444, d_tracer,
1357                                     NULL, &uprobe_profile_ops);
1358         return 0;
1359 }
1360
1361 fs_initcall(init_uprobe_trace);