kernel/trace/trace_uprobe.c

   1 /*
   2  * uprobes-based tracing events
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  16  *
  17  * Copyright (C) IBM Corporation, 2010-2012
  18  * Author:      Srikar Dronamraju <srikar@linux.vnet.ibm.com>
  19  */
  20 #define pr_fmt(fmt)     "trace_kprobe: " fmt
  21
  22 #include <linux/module.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/uprobes.h>
  25 #include <linux/namei.h>
  26 #include <linux/string.h>
  27 #include <linux/rculist.h>
  28
  29 #include "trace_probe.h"
  30
  31 #define UPROBE_EVENT_SYSTEM     "uprobes"
  32
  33 struct uprobe_trace_entry_head {
  34         struct trace_entry      ent;
  35         unsigned long           vaddr[];
  36 };
  37
  38 #define SIZEOF_TRACE_ENTRY(is_return)                   \
  39         (sizeof(struct uprobe_trace_entry_head) +       \
  40          sizeof(unsigned long) * (is_return ? 2 : 1))
  41
  42 #define DATAOF_TRACE_ENTRY(entry, is_return)            \
  43         ((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
  44
  45 struct trace_uprobe_filter {
  46         rwlock_t                rwlock;
  47         int                     nr_systemwide;
  48         struct list_head        perf_events;
  49 };
  50
  51 /*
  52  * uprobe event core functions
  53  */
  54 struct trace_uprobe {
  55         struct list_head                list;
  56         struct trace_uprobe_filter      filter;
  57         struct uprobe_consumer          consumer;
  58         struct inode                    *inode;
  59         char                            *filename;
  60         unsigned long                   offset;
  61         unsigned long                   nhit;
  62         struct trace_probe              tp;
  63 };
  64
  65 #define SIZEOF_TRACE_UPROBE(n)                          \
  66         (offsetof(struct trace_uprobe, tp.args) +       \
  67         (sizeof(struct probe_arg) * (n)))
  68
  69 static int register_uprobe_event(struct trace_uprobe *tu);
  70 static int unregister_uprobe_event(struct trace_uprobe *tu);
  71
  72 static DEFINE_MUTEX(uprobe_lock);
  73 static LIST_HEAD(uprobe_list);
  74
  75 struct uprobe_dispatch_data {
  76         struct trace_uprobe     *tu;
  77         unsigned long           bp_addr;
  78 };
  79
  80 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs);
  81 static int uretprobe_dispatcher(struct uprobe_consumer *con,
  82                                 unsigned long func, struct pt_regs *regs);
  83
  84 #ifdef CONFIG_STACK_GROWSUP
  85 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  86 {
  87         return addr - (n * sizeof(long));
  88 }
  89 #else
  90 static unsigned long adjust_stack_addr(unsigned long addr, unsigned int n)
  91 {
  92         return addr + (n * sizeof(long));
  93 }
  94 #endif
  95
  96 static unsigned long get_user_stack_nth(struct pt_regs *regs, unsigned int n)
  97 {
  98         unsigned long ret;
  99         unsigned long addr = user_stack_pointer(regs);
 100
 101         addr = adjust_stack_addr(addr, n);
 102
 103         if (copy_from_user(&ret, (void __force __user *) addr, sizeof(ret)))
 104                 return 0;
 105
 106         return ret;
 107 }
 108
 109 /*
 110  * Uprobes-specific fetch functions
 111  */
 112 #define DEFINE_FETCH_stack(type)                                        \
 113 static void FETCH_FUNC_NAME(stack, type)(struct pt_regs *regs,          \
 114                                          void *offset, void *dest)      \
 115 {                                                                       \
 116         *(type *)dest = (type)get_user_stack_nth(regs,                  \
 117                                               ((unsigned long)offset)); \
 118 }
 119 DEFINE_BASIC_FETCH_FUNCS(stack)
 120 /* No string on the stack entry */
 121 #define fetch_stack_string      NULL
 122 #define fetch_stack_string_size NULL
 123
 124 #define DEFINE_FETCH_memory(type)                                       \
 125 static void FETCH_FUNC_NAME(memory, type)(struct pt_regs *regs,         \
 126                                           void *addr, void *dest)       \
 127 {                                                                       \
 128         type retval;                                                    \
 129         void __user *vaddr = (void __force __user *) addr;              \
 130                                                                         \
 131         if (copy_from_user(&retval, vaddr, sizeof(type)))               \
 132                 *(type *)dest = 0;                                      \
 133         else                                                            \
 134                 *(type *) dest = retval;                                \
 135 }
 136 DEFINE_BASIC_FETCH_FUNCS(memory)
 137 /*
 138  * Fetch a null-terminated string. Caller MUST set *(u32 *)dest with max
 139  * length and relative data location.
 140  */
 141 static void FETCH_FUNC_NAME(memory, string)(struct pt_regs *regs,
 142                                             void *addr, void *dest)
 143 {
 144         long ret;
 145         u32 rloc = *(u32 *)dest;
 146         int maxlen  = get_rloc_len(rloc);
 147         u8 *dst = get_rloc_data(dest);
 148         void __user *src = (void __force __user *) addr;
 149
 150         if (!maxlen)
 151                 return;
 152
 153         ret = strncpy_from_user(dst, src, maxlen);
 154
 155         if (ret < 0) {  /* Failed to fetch string */
 156                 ((u8 *)get_rloc_data(dest))[0] = '\0';
 157                 *(u32 *)dest = make_data_rloc(0, get_rloc_offs(rloc));
 158         } else {
 159                 *(u32 *)dest = make_data_rloc(ret, get_rloc_offs(rloc));
 160         }
 161 }
 162
 163 static void FETCH_FUNC_NAME(memory, string_size)(struct pt_regs *regs,
 164                                                  void *addr, void *dest)
 165 {
 166         int len;
 167         void __user *vaddr = (void __force __user *) addr;
 168
 169         len = strnlen_user(vaddr, MAX_STRING_SIZE);
 170
 171         if (len == 0 || len > MAX_STRING_SIZE)  /* Failed to check length */
 172                 *(u32 *)dest = 0;
 173         else
 174                 *(u32 *)dest = len;
 175 }
 176
 177 static unsigned long translate_user_vaddr(void *file_offset)
 178 {
 179         unsigned long base_addr;
 180         struct uprobe_dispatch_data *udd;
 181
 182         udd = (void *) current->utask->vaddr;
 183
 184         base_addr = udd->bp_addr - udd->tu->offset;
 185         return base_addr + (unsigned long)file_offset;
 186 }
 187
 188 #define DEFINE_FETCH_file_offset(type)                                  \
 189 static void FETCH_FUNC_NAME(file_offset, type)(struct pt_regs *regs,    \
 190                                                void *offset, void *dest)\
 191 {                                                                       \
 192         void *vaddr = (void *)translate_user_vaddr(offset);             \
 193                                                                         \
 194         FETCH_FUNC_NAME(memory, type)(regs, vaddr, dest);               \
 195 }
 196 DEFINE_BASIC_FETCH_FUNCS(file_offset)
 197 DEFINE_FETCH_file_offset(string)
 198 DEFINE_FETCH_file_offset(string_size)
 199
 200 /* Fetch type information table */
 201 static const struct fetch_type uprobes_fetch_type_table[] = {
 202         /* Special types */
 203         [FETCH_TYPE_STRING] = __ASSIGN_FETCH_TYPE("string", string, string,
 204                                         sizeof(u32), 1, "__data_loc char[]"),
 205         [FETCH_TYPE_STRSIZE] = __ASSIGN_FETCH_TYPE("string_size", u32,
 206                                         string_size, sizeof(u32), 0, "u32"),
 207         /* Basic types */
 208         ASSIGN_FETCH_TYPE(u8,  u8,  0),
 209         ASSIGN_FETCH_TYPE(u16, u16, 0),
 210         ASSIGN_FETCH_TYPE(u32, u32, 0),
 211         ASSIGN_FETCH_TYPE(u64, u64, 0),
 212         ASSIGN_FETCH_TYPE(s8,  u8,  1),
 213         ASSIGN_FETCH_TYPE(s16, u16, 1),
 214         ASSIGN_FETCH_TYPE(s32, u32, 1),
 215         ASSIGN_FETCH_TYPE(s64, u64, 1),
 216         ASSIGN_FETCH_TYPE_ALIAS(x8,  u8,  u8,  0),
 217         ASSIGN_FETCH_TYPE_ALIAS(x16, u16, u16, 0),
 218         ASSIGN_FETCH_TYPE_ALIAS(x32, u32, u32, 0),
 219         ASSIGN_FETCH_TYPE_ALIAS(x64, u64, u64, 0),
 220
 221         ASSIGN_FETCH_TYPE_END
 222 };
 223
 224 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 225 {
 226         rwlock_init(&filter->rwlock);
 227         filter->nr_systemwide = 0;
 228         INIT_LIST_HEAD(&filter->perf_events);
 229 }
 230
 231 static inline bool uprobe_filter_is_empty(struct trace_uprobe_filter *filter)
 232 {
 233         return !filter->nr_systemwide && list_empty(&filter->perf_events);
 234 }
 235
 236 static inline bool is_ret_probe(struct trace_uprobe *tu)
 237 {
 238         return tu->consumer.ret_handler != NULL;
 239 }
 240
 241 /*
 242  * Allocate new trace_uprobe and initialize it (including uprobes).
 243  */
 244 static struct trace_uprobe *
 245 alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 246 {
 247         struct trace_uprobe *tu;
 248
 249         if (!event || !is_good_name(event))
 250                 return ERR_PTR(-EINVAL);
 251
 252         if (!group || !is_good_name(group))
 253                 return ERR_PTR(-EINVAL);
 254
 255         tu = kzalloc(SIZEOF_TRACE_UPROBE(nargs), GFP_KERNEL);
 256         if (!tu)
 257                 return ERR_PTR(-ENOMEM);
 258
 259         tu->tp.call.class = &tu->tp.class;
 260         tu->tp.call.name = kstrdup(event, GFP_KERNEL);
 261         if (!tu->tp.call.name)
 262                 goto error;
 263
 264         tu->tp.class.system = kstrdup(group, GFP_KERNEL);
 265         if (!tu->tp.class.system)
 266                 goto error;
 267
 268         INIT_LIST_HEAD(&tu->list);
 269         INIT_LIST_HEAD(&tu->tp.files);
 270         tu->consumer.handler = uprobe_dispatcher;
 271         if (is_ret)
 272                 tu->consumer.ret_handler = uretprobe_dispatcher;
 273         init_trace_uprobe_filter(&tu->filter);
 274         return tu;
 275
 276 error:
 277         kfree(tu->tp.call.name);
 278         kfree(tu);
 279
 280         return ERR_PTR(-ENOMEM);
 281 }
 282
 283 static void free_trace_uprobe(struct trace_uprobe *tu)
 284 {
 285         int i;
 286
 287         for (i = 0; i < tu->tp.nr_args; i++)
 288                 traceprobe_free_probe_arg(&tu->tp.args[i]);
 289
 290         iput(tu->inode);
 291         kfree(tu->tp.call.class->system);
 292         kfree(tu->tp.call.name);
 293         kfree(tu->filename);
 294         kfree(tu);
 295 }
 296
 297 static struct trace_uprobe *find_probe_event(const char *event, const char *group)
 298 {
 299         struct trace_uprobe *tu;
 300
 301         list_for_each_entry(tu, &uprobe_list, list)
 302                 if (strcmp(trace_event_name(&tu->tp.call), event) == 0 &&
 303                     strcmp(tu->tp.call.class->system, group) == 0)
 304                         return tu;
 305
 306         return NULL;
 307 }
 308
 309 /* Unregister a trace_uprobe and probe_event: call with locking uprobe_lock */
 310 static int unregister_trace_uprobe(struct trace_uprobe *tu)
 311 {
 312         int ret;
 313
 314         ret = unregister_uprobe_event(tu);
 315         if (ret)
 316                 return ret;
 317
 318         list_del(&tu->list);
 319         free_trace_uprobe(tu);
 320         return 0;
 321 }
 322
 323 /* Register a trace_uprobe and probe_event */
 324 static int register_trace_uprobe(struct trace_uprobe *tu)
 325 {
 326         struct trace_uprobe *old_tu;
 327         int ret;
 328
 329         mutex_lock(&uprobe_lock);
 330
 331         /* register as an event */
 332         old_tu = find_probe_event(trace_event_name(&tu->tp.call),
 333                         tu->tp.call.class->system);
 334         if (old_tu) {
 335                 /* delete old event */
 336                 ret = unregister_trace_uprobe(old_tu);
 337                 if (ret)
 338                         goto end;
 339         }
 340
 341         ret = register_uprobe_event(tu);
 342         if (ret) {
 343                 pr_warn("Failed to register probe event(%d)\n", ret);
 344                 goto end;
 345         }
 346
 347         list_add_tail(&tu->list, &uprobe_list);
 348
 349 end:
 350         mutex_unlock(&uprobe_lock);
 351
 352         return ret;
 353 }
 354
 355 /*
 356  * Argument syntax:
 357  *  - Add uprobe: p|r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS]
 358  *
 359  *  - Remove uprobe: -:[GRP/]EVENT
 360  */
 361 static int create_trace_uprobe(int argc, char **argv)
 362 {
 363         struct trace_uprobe *tu;
 364         struct inode *inode;
 365         char *arg, *event, *group, *filename;
 366         char buf[MAX_EVENT_NAME_LEN];
 367         struct path path;
 368         unsigned long offset;
 369         bool is_delete, is_return;
 370         int i, ret;
 371
 372         inode = NULL;
 373         ret = 0;
 374         is_delete = false;
 375         is_return = false;
 376         event = NULL;
 377         group = NULL;
 378
 379         /* argc must be >= 1 */
 380         if (argv[0][0] == '-')
 381                 is_delete = true;
 382         else if (argv[0][0] == 'r')
 383                 is_return = true;
 384         else if (argv[0][0] != 'p') {
 385                 pr_info("Probe definition must be started with 'p', 'r' or '-'.\n");
 386                 return -EINVAL;
 387         }
 388
 389         if (argv[0][1] == ':') {
 390                 event = &argv[0][2];
 391                 arg = strchr(event, '/');
 392
 393                 if (arg) {
 394                         group = event;
 395                         event = arg + 1;
 396                         event[-1] = '\0';
 397
 398                         if (strlen(group) == 0) {
 399                                 pr_info("Group name is not specified\n");
 400                                 return -EINVAL;
 401                         }
 402                 }
 403                 if (strlen(event) == 0) {
 404                         pr_info("Event name is not specified\n");
 405                         return -EINVAL;
 406                 }
 407         }
 408         if (!group)
 409                 group = UPROBE_EVENT_SYSTEM;
 410
 411         if (is_delete) {
 412                 int ret;
 413
 414                 if (!event) {
 415                         pr_info("Delete command needs an event name.\n");
 416                         return -EINVAL;
 417                 }
 418                 mutex_lock(&uprobe_lock);
 419                 tu = find_probe_event(event, group);
 420
 421                 if (!tu) {
 422                         mutex_unlock(&uprobe_lock);
 423                         pr_info("Event %s/%s doesn't exist.\n", group, event);
 424                         return -ENOENT;
 425                 }
 426                 /* delete an event */
 427                 ret = unregister_trace_uprobe(tu);
 428                 mutex_unlock(&uprobe_lock);
 429                 return ret;
 430         }
 431
 432         if (argc < 2) {
 433                 pr_info("Probe point is not specified.\n");
 434                 return -EINVAL;
 435         }
 436         /* Find the last occurrence, in case the path contains ':' too. */
 437         arg = strrchr(argv[1], ':');
 438         if (!arg) {
 439                 ret = -EINVAL;
 440                 goto fail_address_parse;
 441         }
 442
 443         *arg++ = '\0';
 444         filename = argv[1];
 445         ret = kern_path(filename, LOOKUP_FOLLOW, &path);
 446         if (ret)
 447                 goto fail_address_parse;
 448
 449         inode = igrab(d_inode(path.dentry));
 450         path_put(&path);
 451
 452         if (!inode || !S_ISREG(inode->i_mode)) {
 453                 ret = -EINVAL;
 454                 goto fail_address_parse;
 455         }
 456
 457         ret = kstrtoul(arg, 0, &offset);
 458         if (ret)
 459                 goto fail_address_parse;
 460
 461         argc -= 2;
 462         argv += 2;
 463
 464         /* setup a probe */
 465         if (!event) {
 466                 char *tail;
 467                 char *ptr;
 468
 469                 tail = kstrdup(kbasename(filename), GFP_KERNEL);
 470                 if (!tail) {
 471                         ret = -ENOMEM;
 472                         goto fail_address_parse;
 473                 }
 474
 475                 ptr = strpbrk(tail, ".-_");
 476                 if (ptr)
 477                         *ptr = '\0';
 478
 479                 snprintf(buf, MAX_EVENT_NAME_LEN, "%c_%s_0x%lx", 'p', tail, offset);
 480                 event = buf;
 481                 kfree(tail);
 482         }
 483
 484         tu = alloc_trace_uprobe(group, event, argc, is_return);
 485         if (IS_ERR(tu)) {
 486                 pr_info("Failed to allocate trace_uprobe.(%d)\n", (int)PTR_ERR(tu));
 487                 ret = PTR_ERR(tu);
 488                 goto fail_address_parse;
 489         }
 490         tu->offset = offset;
 491         tu->inode = inode;
 492         tu->filename = kstrdup(filename, GFP_KERNEL);
 493
 494         if (!tu->filename) {
 495                 pr_info("Failed to allocate filename.\n");
 496                 ret = -ENOMEM;
 497                 goto error;
 498         }
 499
 500         /* parse arguments */
 501         ret = 0;
 502         for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
 503                 struct probe_arg *parg = &tu->tp.args[i];
 504
 505                 /* Increment count for freeing args in error case */
 506                 tu->tp.nr_args++;
 507
 508                 /* Parse argument name */
 509                 arg = strchr(argv[i], '=');
 510                 if (arg) {
 511                         *arg++ = '\0';
 512                         parg->name = kstrdup(argv[i], GFP_KERNEL);
 513                 } else {
 514                         arg = argv[i];
 515                         /* If argument name is omitted, set "argN" */
 516                         snprintf(buf, MAX_EVENT_NAME_LEN, "arg%d", i + 1);
 517                         parg->name = kstrdup(buf, GFP_KERNEL);
 518                 }
 519
 520                 if (!parg->name) {
 521                         pr_info("Failed to allocate argument[%d] name.\n", i);
 522                         ret = -ENOMEM;
 523                         goto error;
 524                 }
 525
 526                 if (!is_good_name(parg->name)) {
 527                         pr_info("Invalid argument[%d] name: %s\n", i, parg->name);
 528                         ret = -EINVAL;
 529                         goto error;
 530                 }
 531
 532                 if (traceprobe_conflict_field_name(parg->name, tu->tp.args, i)) {
 533                         pr_info("Argument[%d] name '%s' conflicts with "
 534                                 "another field.\n", i, argv[i]);
 535                         ret = -EINVAL;
 536                         goto error;
 537                 }
 538
 539                 /* Parse fetch argument */
 540                 ret = traceprobe_parse_probe_arg(arg, &tu->tp.size, parg,
 541                                                  is_return, false,
 542                                                  uprobes_fetch_type_table);
 543                 if (ret) {
 544                         pr_info("Parse error at argument[%d]. (%d)\n", i, ret);
 545                         goto error;
 546                 }
 547         }
 548
 549         ret = register_trace_uprobe(tu);
 550         if (ret)
 551                 goto error;
 552         return 0;
 553
 554 error:
 555         free_trace_uprobe(tu);
 556         return ret;
 557
 558 fail_address_parse:
 559         iput(inode);
 560
 561         pr_info("Failed to parse address or file.\n");
 562
 563         return ret;
 564 }
 565
 566 static int cleanup_all_probes(void)
 567 {
 568         struct trace_uprobe *tu;
 569         int ret = 0;
 570
 571         mutex_lock(&uprobe_lock);
 572         while (!list_empty(&uprobe_list)) {
 573                 tu = list_entry(uprobe_list.next, struct trace_uprobe, list);
 574                 ret = unregister_trace_uprobe(tu);
 575                 if (ret)
 576                         break;
 577         }
 578         mutex_unlock(&uprobe_lock);
 579         return ret;
 580 }
 581
 582 /* Probes listing interfaces */
 583 static void *probes_seq_start(struct seq_file *m, loff_t *pos)
 584 {
 585         mutex_lock(&uprobe_lock);
 586         return seq_list_start(&uprobe_list, *pos);
 587 }
 588
 589 static void *probes_seq_next(struct seq_file *m, void *v, loff_t *pos)
 590 {
 591         return seq_list_next(v, &uprobe_list, pos);
 592 }
 593
 594 static void probes_seq_stop(struct seq_file *m, void *v)
 595 {
 596         mutex_unlock(&uprobe_lock);
 597 }
 598
 599 static int probes_seq_show(struct seq_file *m, void *v)
 600 {
 601         struct trace_uprobe *tu = v;
 602         char c = is_ret_probe(tu) ? 'r' : 'p';
 603         int i;
 604
 605         seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
 606                         trace_event_name(&tu->tp.call));
 607         seq_printf(m, " %s:", tu->filename);
 608
 609         /* Don't print "0x  (null)" when offset is 0 */
 610         if (tu->offset) {
 611                 seq_printf(m, "0x%p", (void *)tu->offset);
 612         } else {
 613                 switch (sizeof(void *)) {
 614                 case 4:
 615                         seq_printf(m, "0x00000000");
 616                         break;
 617                 case 8:
 618                 default:
 619                         seq_printf(m, "0x0000000000000000");
 620                         break;
 621                 }
 622         }
 623
 624         for (i = 0; i < tu->tp.nr_args; i++)
 625                 seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
 626
 627         seq_putc(m, '\n');
 628         return 0;
 629 }
 630
 631 static const struct seq_operations probes_seq_op = {
 632         .start  = probes_seq_start,
 633         .next   = probes_seq_next,
 634         .stop   = probes_seq_stop,
 635         .show   = probes_seq_show
 636 };
 637
 638 static int probes_open(struct inode *inode, struct file *file)
 639 {
 640         int ret;
 641
 642         if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
 643                 ret = cleanup_all_probes();
 644                 if (ret)
 645                         return ret;
 646         }
 647
 648         return seq_open(file, &probes_seq_op);
 649 }
 650
 651 static ssize_t probes_write(struct file *file, const char __user *buffer,
 652                             size_t count, loff_t *ppos)
 653 {
 654         return traceprobe_probes_write(file, buffer, count, ppos, create_trace_uprobe);
 655 }
 656
 657 static const struct file_operations uprobe_events_ops = {
 658         .owner          = THIS_MODULE,
 659         .open           = probes_open,
 660         .read           = seq_read,
 661         .llseek         = seq_lseek,
 662         .release        = seq_release,
 663         .write          = probes_write,
 664 };
 665
 666 /* Probes profiling interfaces */
 667 static int probes_profile_seq_show(struct seq_file *m, void *v)
 668 {
 669         struct trace_uprobe *tu = v;
 670
 671         seq_printf(m, "  %s %-44s %15lu\n", tu->filename,
 672                         trace_event_name(&tu->tp.call), tu->nhit);
 673         return 0;
 674 }
 675
 676 static const struct seq_operations profile_seq_op = {
 677         .start  = probes_seq_start,
 678         .next   = probes_seq_next,
 679         .stop   = probes_seq_stop,
 680         .show   = probes_profile_seq_show
 681 };
 682
 683 static int profile_open(struct inode *inode, struct file *file)
 684 {
 685         return seq_open(file, &profile_seq_op);
 686 }
 687
 688 static const struct file_operations uprobe_profile_ops = {
 689         .owner          = THIS_MODULE,
 690         .open           = profile_open,
 691         .read           = seq_read,
 692         .llseek         = seq_lseek,
 693         .release        = seq_release,
 694 };
 695
 696 struct uprobe_cpu_buffer {
 697         struct mutex mutex;
 698         void *buf;
 699 };
 700 static struct uprobe_cpu_buffer __percpu *uprobe_cpu_buffer;
 701 static int uprobe_buffer_refcnt;
 702
 703 static int uprobe_buffer_init(void)
 704 {
 705         int cpu, err_cpu;
 706
 707         uprobe_cpu_buffer = alloc_percpu(struct uprobe_cpu_buffer);
 708         if (uprobe_cpu_buffer == NULL)
 709                 return -ENOMEM;
 710
 711         for_each_possible_cpu(cpu) {
 712                 struct page *p = alloc_pages_node(cpu_to_node(cpu),
 713                                                   GFP_KERNEL, 0);
 714                 if (p == NULL) {
 715                         err_cpu = cpu;
 716                         goto err;
 717                 }
 718                 per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf = page_address(p);
 719                 mutex_init(&per_cpu_ptr(uprobe_cpu_buffer, cpu)->mutex);
 720         }
 721
 722         return 0;
 723
 724 err:
 725         for_each_possible_cpu(cpu) {
 726                 if (cpu == err_cpu)
 727                         break;
 728                 free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer, cpu)->buf);
 729         }
 730
 731         free_percpu(uprobe_cpu_buffer);
 732         return -ENOMEM;
 733 }
 734
 735 static int uprobe_buffer_enable(void)
 736 {
 737         int ret = 0;
 738
 739         BUG_ON(!mutex_is_locked(&event_mutex));
 740
 741         if (uprobe_buffer_refcnt++ == 0) {
 742                 ret = uprobe_buffer_init();
 743                 if (ret < 0)
 744                         uprobe_buffer_refcnt--;
 745         }
 746
 747         return ret;
 748 }
 749
 750 static void uprobe_buffer_disable(void)
 751 {
 752         int cpu;
 753
 754         BUG_ON(!mutex_is_locked(&event_mutex));
 755
 756         if (--uprobe_buffer_refcnt == 0) {
 757                 for_each_possible_cpu(cpu)
 758                         free_page((unsigned long)per_cpu_ptr(uprobe_cpu_buffer,
 759                                                              cpu)->buf);
 760
 761                 free_percpu(uprobe_cpu_buffer);
 762                 uprobe_cpu_buffer = NULL;
 763         }
 764 }
 765
 766 static struct uprobe_cpu_buffer *uprobe_buffer_get(void)
 767 {
 768         struct uprobe_cpu_buffer *ucb;
 769         int cpu;
 770
 771         cpu = raw_smp_processor_id();
 772         ucb = per_cpu_ptr(uprobe_cpu_buffer, cpu);
 773
 774         /*
 775          * Use per-cpu buffers for fastest access, but we might migrate
 776          * so the mutex makes sure we have sole access to it.
 777          */
 778         mutex_lock(&ucb->mutex);
 779
 780         return ucb;
 781 }
 782
 783 static void uprobe_buffer_put(struct uprobe_cpu_buffer *ucb)
 784 {
 785         mutex_unlock(&ucb->mutex);
 786 }
 787
 788 static void __uprobe_trace_func(struct trace_uprobe *tu,
 789                                 unsigned long func, struct pt_regs *regs,
 790                                 struct uprobe_cpu_buffer *ucb, int dsize,
 791                                 struct trace_event_file *trace_file)
 792 {
 793         struct uprobe_trace_entry_head *entry;
 794         struct ring_buffer_event *event;
 795         struct ring_buffer *buffer;
 796         void *data;
 797         int size, esize;
 798         struct trace_event_call *call = &tu->tp.call;
 799
 800         WARN_ON(call != trace_file->event_call);
 801
 802         if (WARN_ON_ONCE(tu->tp.size + dsize > PAGE_SIZE))
 803                 return;
 804
 805         if (trace_trigger_soft_disabled(trace_file))
 806                 return;
 807
 808         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
 809         size = esize + tu->tp.size + dsize;
 810         event = trace_event_buffer_lock_reserve(&buffer, trace_file,
 811                                                 call->event.type, size, 0, 0);
 812         if (!event)
 813                 return;
 814
 815         entry = ring_buffer_event_data(event);
 816         if (is_ret_probe(tu)) {
 817                 entry->vaddr[0] = func;
 818                 entry->vaddr[1] = instruction_pointer(regs);
 819                 data = DATAOF_TRACE_ENTRY(entry, true);
 820         } else {
 821                 entry->vaddr[0] = instruction_pointer(regs);
 822                 data = DATAOF_TRACE_ENTRY(entry, false);
 823         }
 824
 825         memcpy(data, ucb->buf, tu->tp.size + dsize);
 826
 827         event_trigger_unlock_commit(trace_file, buffer, event, entry, 0, 0);
 828 }
 829
 830 /* uprobe handler */
 831 static int uprobe_trace_func(struct trace_uprobe *tu, struct pt_regs *regs,
 832                              struct uprobe_cpu_buffer *ucb, int dsize)
 833 {
 834         struct event_file_link *link;
 835
 836         if (is_ret_probe(tu))
 837                 return 0;
 838
 839         rcu_read_lock();
 840         list_for_each_entry_rcu(link, &tu->tp.files, list)
 841                 __uprobe_trace_func(tu, 0, regs, ucb, dsize, link->file);
 842         rcu_read_unlock();
 843
 844         return 0;
 845 }
 846
 847 static void uretprobe_trace_func(struct trace_uprobe *tu, unsigned long func,
 848                                  struct pt_regs *regs,
 849                                  struct uprobe_cpu_buffer *ucb, int dsize)
 850 {
 851         struct event_file_link *link;
 852
 853         rcu_read_lock();
 854         list_for_each_entry_rcu(link, &tu->tp.files, list)
 855                 __uprobe_trace_func(tu, func, regs, ucb, dsize, link->file);
 856         rcu_read_unlock();
 857 }
 858
 859 /* Event entry printers */
 860 static enum print_line_t
 861 print_uprobe_event(struct trace_iterator *iter, int flags, struct trace_event *event)
 862 {
 863         struct uprobe_trace_entry_head *entry;
 864         struct trace_seq *s = &iter->seq;
 865         struct trace_uprobe *tu;
 866         u8 *data;
 867         int i;
 868
 869         entry = (struct uprobe_trace_entry_head *)iter->ent;
 870         tu = container_of(event, struct trace_uprobe, tp.call.event);
 871
 872         if (is_ret_probe(tu)) {
 873                 trace_seq_printf(s, "%s: (0x%lx <- 0x%lx)",
 874                                  trace_event_name(&tu->tp.call),
 875                                  entry->vaddr[1], entry->vaddr[0]);
 876                 data = DATAOF_TRACE_ENTRY(entry, true);
 877         } else {
 878                 trace_seq_printf(s, "%s: (0x%lx)",
 879                                  trace_event_name(&tu->tp.call),
 880                                  entry->vaddr[0]);
 881                 data = DATAOF_TRACE_ENTRY(entry, false);
 882         }
 883
 884         for (i = 0; i < tu->tp.nr_args; i++) {
 885                 struct probe_arg *parg = &tu->tp.args[i];
 886
 887                 if (!parg->type->print(s, parg->name, data + parg->offset, entry))
 888                         goto out;
 889         }
 890
 891         trace_seq_putc(s, '\n');
 892
 893  out:
 894         return trace_handle_return(s);
 895 }
 896
 897 typedef bool (*filter_func_t)(struct uprobe_consumer *self,
 898                                 enum uprobe_filter_ctx ctx,
 899                                 struct mm_struct *mm);
 900
 901 static int
 902 probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
 903                    filter_func_t filter)
 904 {
 905         bool enabled = trace_probe_is_enabled(&tu->tp);
 906         struct event_file_link *link = NULL;
 907         int ret;
 908
 909         if (file) {
 910                 if (tu->tp.flags & TP_FLAG_PROFILE)
 911                         return -EINTR;
 912
 913                 link = kmalloc(sizeof(*link), GFP_KERNEL);
 914                 if (!link)
 915                         return -ENOMEM;
 916
 917                 link->file = file;
 918                 list_add_tail_rcu(&link->list, &tu->tp.files);
 919
 920                 tu->tp.flags |= TP_FLAG_TRACE;
 921         } else {
 922                 if (tu->tp.flags & TP_FLAG_TRACE)
 923                         return -EINTR;
 924
 925                 tu->tp.flags |= TP_FLAG_PROFILE;
 926         }
 927
 928         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 929
 930         if (enabled)
 931                 return 0;
 932
 933         ret = uprobe_buffer_enable();
 934         if (ret)
 935                 goto err_flags;
 936
 937         tu->consumer.filter = filter;
 938         ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
 939         if (ret)
 940                 goto err_buffer;
 941
 942         return 0;
 943
 944  err_buffer:
 945         uprobe_buffer_disable();
 946
 947  err_flags:
 948         if (file) {
 949                 list_del(&link->list);
 950                 kfree(link);
 951                 tu->tp.flags &= ~TP_FLAG_TRACE;
 952         } else {
 953                 tu->tp.flags &= ~TP_FLAG_PROFILE;
 954         }
 955         return ret;
 956 }
 957
 958 static void
 959 probe_event_disable(struct trace_uprobe *tu, struct trace_event_file *file)
 960 {
 961         if (!trace_probe_is_enabled(&tu->tp))
 962                 return;
 963
 964         if (file) {
 965                 struct event_file_link *link;
 966
 967                 link = find_event_file_link(&tu->tp, file);
 968                 if (!link)
 969                         return;
 970
 971                 list_del_rcu(&link->list);
 972                 /* synchronize with u{,ret}probe_trace_func */
 973                 synchronize_sched();
 974                 kfree(link);
 975
 976                 if (!list_empty(&tu->tp.files))
 977                         return;
 978         }
 979
 980         WARN_ON(!uprobe_filter_is_empty(&tu->filter));
 981
 982         uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 983         tu->tp.flags &= file ? ~TP_FLAG_TRACE : ~TP_FLAG_PROFILE;
 984
 985         uprobe_buffer_disable();
 986 }
 987
 988 static int uprobe_event_define_fields(struct trace_event_call *event_call)
 989 {
 990         int ret, i, size;
 991         struct uprobe_trace_entry_head field;
 992         struct trace_uprobe *tu = event_call->data;
 993
 994         if (is_ret_probe(tu)) {
 995                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_FUNC, 0);
 996                 DEFINE_FIELD(unsigned long, vaddr[1], FIELD_STRING_RETIP, 0);
 997                 size = SIZEOF_TRACE_ENTRY(true);
 998         } else {
 999                 DEFINE_FIELD(unsigned long, vaddr[0], FIELD_STRING_IP, 0);
1000                 size = SIZEOF_TRACE_ENTRY(false);
1001         }
1002         /* Set argument names as fields */
1003         for (i = 0; i < tu->tp.nr_args; i++) {
1004                 struct probe_arg *parg = &tu->tp.args[i];
1005
1006                 ret = trace_define_field(event_call, parg->type->fmttype,
1007                                          parg->name, size + parg->offset,
1008                                          parg->type->size, parg->type->is_signed,
1009                                          FILTER_OTHER);
1010
1011                 if (ret)
1012                         return ret;
1013         }
1014         return 0;
1015 }
1016
1017 #ifdef CONFIG_PERF_EVENTS
1018 static bool
1019 __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
1020 {
1021         struct perf_event *event;
1022
1023         if (filter->nr_systemwide)
1024                 return true;
1025
1026         list_for_each_entry(event, &filter->perf_events, hw.tp_list) {
1027                 if (event->hw.target->mm == mm)
1028                         return true;
1029         }
1030
1031         return false;
1032 }
1033
1034 static inline bool
1035 uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
1036 {
1037         return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
1038 }
1039
1040 static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
1041 {
1042         bool done;
1043
1044         write_lock(&tu->filter.rwlock);
1045         if (event->hw.target) {
1046                 list_del(&event->hw.tp_list);
1047                 done = tu->filter.nr_systemwide ||
1048                         (event->hw.target->flags & PF_EXITING) ||
1049                         uprobe_filter_event(tu, event);
1050         } else {
1051                 tu->filter.nr_systemwide--;
1052                 done = tu->filter.nr_systemwide;
1053         }
1054         write_unlock(&tu->filter.rwlock);
1055
1056         if (!done)
1057                 return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
1058
1059         return 0;
1060 }
1061
1062 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
1063 {
1064         bool done;
1065         int err;
1066
1067         write_lock(&tu->filter.rwlock);
1068         if (event->hw.target) {
1069                 /*
1070                  * event->parent != NULL means copy_process(), we can avoid
1071                  * uprobe_apply(). current->mm must be probed and we can rely
1072                  * on dup_mmap() which preserves the already installed bp's.
1073                  *
1074                  * attr.enable_on_exec means that exec/mmap will install the
1075                  * breakpoints we need.
1076                  */
1077                 done = tu->filter.nr_systemwide ||
1078                         event->parent || event->attr.enable_on_exec ||
1079                         uprobe_filter_event(tu, event);
1080                 list_add(&event->hw.tp_list, &tu->filter.perf_events);
1081         } else {
1082                 done = tu->filter.nr_systemwide;
1083                 tu->filter.nr_systemwide++;
1084         }
1085         write_unlock(&tu->filter.rwlock);
1086
1087         err = 0;
1088         if (!done) {
1089                 err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
1090                 if (err)
1091                         uprobe_perf_close(tu, event);
1092         }
1093         return err;
1094 }
1095
1096 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
1097                                 enum uprobe_filter_ctx ctx, struct mm_struct *mm)
1098 {
1099         struct trace_uprobe *tu;
1100         int ret;
1101
1102         tu = container_of(uc, struct trace_uprobe, consumer);
1103         read_lock(&tu->filter.rwlock);
1104         ret = __uprobe_perf_filter(&tu->filter, mm);
1105         read_unlock(&tu->filter.rwlock);
1106
1107         return ret;
1108 }
1109
1110 static void __uprobe_perf_func(struct trace_uprobe *tu,
1111                                unsigned long func, struct pt_regs *regs,
1112                                struct uprobe_cpu_buffer *ucb, int dsize)
1113 {
1114         struct trace_event_call *call = &tu->tp.call;
1115         struct uprobe_trace_entry_head *entry;
1116         struct bpf_prog *prog = call->prog;
1117         struct hlist_head *head;
1118         void *data;
1119         int size, esize;
1120         int rctx;
1121
1122         if (prog && !trace_call_bpf(prog, regs))
1123                 return;
1124
1125         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1126
1127         size = esize + tu->tp.size + dsize;
1128         size = ALIGN(size + sizeof(u32), sizeof(u64)) - sizeof(u32);
1129         if (WARN_ONCE(size > PERF_MAX_TRACE_SIZE, "profile buffer not large enough"))
1130                 return;
1131
1132         preempt_disable();
1133         head = this_cpu_ptr(call->perf_events);
1134         if (hlist_empty(head))
1135                 goto out;
1136
1137         entry = perf_trace_buf_alloc(size, NULL, &rctx);
1138         if (!entry)
1139                 goto out;
1140
1141         if (is_ret_probe(tu)) {
1142                 entry->vaddr[0] = func;
1143                 entry->vaddr[1] = instruction_pointer(regs);
1144                 data = DATAOF_TRACE_ENTRY(entry, true);
1145         } else {
1146                 entry->vaddr[0] = instruction_pointer(regs);
1147                 data = DATAOF_TRACE_ENTRY(entry, false);
1148         }
1149
1150         memcpy(data, ucb->buf, tu->tp.size + dsize);
1151
1152         if (size - esize > tu->tp.size + dsize) {
1153                 int len = tu->tp.size + dsize;
1154
1155                 memset(data + len, 0, size - esize - len);
1156         }
1157
1158         perf_trace_buf_submit(entry, size, rctx, call->event.type, 1, regs,
1159                               head, NULL);
1160  out:
1161         preempt_enable();
1162 }
1163
1164 /* uprobe profile handler */
1165 static int uprobe_perf_func(struct trace_uprobe *tu, struct pt_regs *regs,
1166                             struct uprobe_cpu_buffer *ucb, int dsize)
1167 {
1168         if (!uprobe_perf_filter(&tu->consumer, 0, current->mm))
1169                 return UPROBE_HANDLER_REMOVE;
1170
1171         if (!is_ret_probe(tu))
1172                 __uprobe_perf_func(tu, 0, regs, ucb, dsize);
1173         return 0;
1174 }
1175
1176 static void uretprobe_perf_func(struct trace_uprobe *tu, unsigned long func,
1177                                 struct pt_regs *regs,
1178                                 struct uprobe_cpu_buffer *ucb, int dsize)
1179 {
1180         __uprobe_perf_func(tu, func, regs, ucb, dsize);
1181 }
1182 #endif  /* CONFIG_PERF_EVENTS */
1183
1184 static int
1185 trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
1186                       void *data)
1187 {
1188         struct trace_uprobe *tu = event->data;
1189         struct trace_event_file *file = data;
1190
1191         switch (type) {
1192         case TRACE_REG_REGISTER:
1193                 return probe_event_enable(tu, file, NULL);
1194
1195         case TRACE_REG_UNREGISTER:
1196                 probe_event_disable(tu, file);
1197                 return 0;
1198
1199 #ifdef CONFIG_PERF_EVENTS
1200         case TRACE_REG_PERF_REGISTER:
1201                 return probe_event_enable(tu, NULL, uprobe_perf_filter);
1202
1203         case TRACE_REG_PERF_UNREGISTER:
1204                 probe_event_disable(tu, NULL);
1205                 return 0;
1206
1207         case TRACE_REG_PERF_OPEN:
1208                 return uprobe_perf_open(tu, data);
1209
1210         case TRACE_REG_PERF_CLOSE:
1211                 return uprobe_perf_close(tu, data);
1212
1213 #endif
1214         default:
1215                 return 0;
1216         }
1217         return 0;
1218 }
1219
1220 static int uprobe_dispatcher(struct uprobe_consumer *con, struct pt_regs *regs)
1221 {
1222         struct trace_uprobe *tu;
1223         struct uprobe_dispatch_data udd;
1224         struct uprobe_cpu_buffer *ucb;
1225         int dsize, esize;
1226         int ret = 0;
1227
1228
1229         tu = container_of(con, struct trace_uprobe, consumer);
1230         tu->nhit++;
1231
1232         udd.tu = tu;
1233         udd.bp_addr = instruction_pointer(regs);
1234
1235         current->utask->vaddr = (unsigned long) &udd;
1236
1237         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1238                 return 0;
1239
1240         dsize = __get_data_size(&tu->tp, regs);
1241         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1242
1243         ucb = uprobe_buffer_get();
1244         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1245
1246         if (tu->tp.flags & TP_FLAG_TRACE)
1247                 ret |= uprobe_trace_func(tu, regs, ucb, dsize);
1248
1249 #ifdef CONFIG_PERF_EVENTS
1250         if (tu->tp.flags & TP_FLAG_PROFILE)
1251                 ret |= uprobe_perf_func(tu, regs, ucb, dsize);
1252 #endif
1253         uprobe_buffer_put(ucb);
1254         return ret;
1255 }
1256
1257 static int uretprobe_dispatcher(struct uprobe_consumer *con,
1258                                 unsigned long func, struct pt_regs *regs)
1259 {
1260         struct trace_uprobe *tu;
1261         struct uprobe_dispatch_data udd;
1262         struct uprobe_cpu_buffer *ucb;
1263         int dsize, esize;
1264
1265         tu = container_of(con, struct trace_uprobe, consumer);
1266
1267         udd.tu = tu;
1268         udd.bp_addr = func;
1269
1270         current->utask->vaddr = (unsigned long) &udd;
1271
1272         if (WARN_ON_ONCE(!uprobe_cpu_buffer))
1273                 return 0;
1274
1275         dsize = __get_data_size(&tu->tp, regs);
1276         esize = SIZEOF_TRACE_ENTRY(is_ret_probe(tu));
1277
1278         ucb = uprobe_buffer_get();
1279         store_trace_args(esize, &tu->tp, regs, ucb->buf, dsize);
1280
1281         if (tu->tp.flags & TP_FLAG_TRACE)
1282                 uretprobe_trace_func(tu, func, regs, ucb, dsize);
1283
1284 #ifdef CONFIG_PERF_EVENTS
1285         if (tu->tp.flags & TP_FLAG_PROFILE)
1286                 uretprobe_perf_func(tu, func, regs, ucb, dsize);
1287 #endif
1288         uprobe_buffer_put(ucb);
1289         return 0;
1290 }
1291
1292 static struct trace_event_functions uprobe_funcs = {
1293         .trace          = print_uprobe_event
1294 };
1295
1296 static int register_uprobe_event(struct trace_uprobe *tu)
1297 {
1298         struct trace_event_call *call = &tu->tp.call;
1299         int ret;
1300
1301         /* Initialize trace_event_call */
1302         INIT_LIST_HEAD(&call->class->fields);
1303         call->event.funcs = &uprobe_funcs;
1304         call->class->define_fields = uprobe_event_define_fields;
1305
1306         if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
1307                 return -ENOMEM;
1308
1309         ret = register_trace_event(&call->event);
1310         if (!ret) {
1311                 kfree(call->print_fmt);
1312                 return -ENODEV;
1313         }
1314
1315         call->flags = TRACE_EVENT_FL_UPROBE;
1316         call->class->reg = trace_uprobe_register;
1317         call->data = tu;
1318         ret = trace_add_event_call(call);
1319
1320         if (ret) {
1321                 pr_info("Failed to register uprobe event: %s\n",
1322                         trace_event_name(call));
1323                 kfree(call->print_fmt);
1324                 unregister_trace_event(&call->event);
1325         }
1326
1327         return ret;
1328 }
1329
1330 static int unregister_uprobe_event(struct trace_uprobe *tu)
1331 {
1332         int ret;
1333
1334         /* tu->event is unregistered in trace_remove_event_call() */
1335         ret = trace_remove_event_call(&tu->tp.call);
1336         if (ret)
1337                 return ret;
1338         kfree(tu->tp.call.print_fmt);
1339         tu->tp.call.print_fmt = NULL;
1340         return 0;
1341 }
1342
1343 /* Make a trace interface for controling probe points */
1344 static __init int init_uprobe_trace(void)
1345 {
1346         struct dentry *d_tracer;
1347
1348         d_tracer = tracing_init_dentry();
1349         if (IS_ERR(d_tracer))
1350                 return 0;
1351
1352         trace_create_file("uprobe_events", 0644, d_tracer,
1353                                     NULL, &uprobe_events_ops);
1354         /* Profile interface */
1355         trace_create_file("uprobe_profile", 0444, d_tracer,
1356                                     NULL, &uprobe_profile_ops);
1357         return 0;
1358 }
1359
1360 fs_initcall(init_uprobe_trace);