kernel/uprobes.c

   1 /*
   2  * Userspace Probes (UProbes)
   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 as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright (C) IBM Corporation, 2008-2011
  19  * Authors:
  20  *      Srikar Dronamraju
  21  *      Jim Keniston
  22  */
  23
  24 #include <linux/kernel.h>
  25 #include <linux/highmem.h>
  26 #include <linux/pagemap.h>      /* read_mapping_page */
  27 #include <linux/slab.h>
  28 #include <linux/sched.h>
  29 #include <linux/rmap.h>         /* anon_vma_prepare */
  30 #include <linux/mmu_notifier.h> /* set_pte_at_notify */
  31 #include <linux/swap.h>         /* try_to_free_swap */
  32 #include <linux/ptrace.h>       /* user_enable_single_step */
  33 #include <linux/kdebug.h>       /* notifier mechanism */
  34 #include <linux/uprobes.h>
  35
  36 #define UINSNS_PER_PAGE (PAGE_SIZE/UPROBES_XOL_SLOT_BYTES)
  37 #define MAX_UPROBES_XOL_SLOTS UINSNS_PER_PAGE
  38
  39 static bulkref_t uprobes_srcu;
  40 static struct rb_root uprobes_tree = RB_ROOT;
  41 static DEFINE_SPINLOCK(uprobes_treelock);       /* serialize rbtree access */
  42
  43 #define UPROBES_HASH_SZ 13
  44 /* serialize (un)register */
  45 static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
  46 #define uprobes_hash(v) (&uprobes_mutex[((unsigned long)(v)) %\
  47                                                 UPROBES_HASH_SZ])
  48
  49 /* serialize uprobe->pending_list */
  50 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
  51 #define uprobes_mmap_hash(v)    (&uprobes_mmap_mutex[((unsigned long)(v)) %\
  52                                                 UPROBES_HASH_SZ])
  53
  54 /*
  55  * uprobe_events allows us to skip the mmap_uprobe if there are no uprobe
  56  * events active at this time.  Probably a fine grained per inode count is
  57  * better?
  58  */
  59 static atomic_t uprobe_events = ATOMIC_INIT(0);
  60
  61 /*
  62  * Maintain a temporary per vma info that can be used to search if a vma
  63  * has already been handled. This structure is introduced since extending
  64  * vm_area_struct wasnt recommended.
  65  */
  66 struct vma_info {
  67         struct list_head probe_list;
  68         struct mm_struct *mm;
  69         loff_t vaddr;
  70 };
  71
  72 /*
  73  * valid_vma: Verify if the specified vma is an executable vma
  74  * Relax restrictions while unregistering: vm_flags might have
  75  * changed after breakpoint was inserted.
  76  *      - is_reg: indicates if we are in register context.
  77  *      - Return 1 if the specified virtual address is in an
  78  *        executable vma.
  79  */
  80 static bool valid_vma(struct vm_area_struct *vma, bool is_reg)
  81 {
  82         if (!vma->vm_file)
  83                 return false;
  84
  85         if (!is_reg)
  86                 return true;
  87
  88         if ((vma->vm_flags & (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)) ==
  89                                                 (VM_READ|VM_EXEC))
  90                 return true;
  91
  92         return false;
  93 }
  94
  95 /**
  96  * __replace_page - replace page in vma by new page.
  97  * based on replace_page in mm/ksm.c
  98  *
  99  * @vma:      vma that holds the pte pointing to page
 100  * @page:     the cowed page we are replacing by kpage
 101  * @kpage:    the modified page we replace page by
 102  *
 103  * Returns 0 on success, -EFAULT on failure.
 104  */
 105 static int __replace_page(struct vm_area_struct *vma, struct page *page,
 106                                         struct page *kpage)
 107 {
 108         struct mm_struct *mm = vma->vm_mm;
 109         pgd_t *pgd;
 110         pud_t *pud;
 111         pmd_t *pmd;
 112         pte_t *ptep;
 113         spinlock_t *ptl;
 114         unsigned long addr;
 115         int err = -EFAULT;
 116
 117         addr = page_address_in_vma(page, vma);
 118         if (addr == -EFAULT)
 119                 goto out;
 120
 121         pgd = pgd_offset(mm, addr);
 122         if (!pgd_present(*pgd))
 123                 goto out;
 124
 125         pud = pud_offset(pgd, addr);
 126         if (!pud_present(*pud))
 127                 goto out;
 128
 129         pmd = pmd_offset(pud, addr);
 130         if (!pmd_present(*pmd))
 131                 goto out;
 132
 133         ptep = pte_offset_map_lock(mm, pmd, addr, &ptl);
 134         if (!ptep)
 135                 goto out;
 136
 137         get_page(kpage);
 138         page_add_new_anon_rmap(kpage, vma, addr);
 139
 140         flush_cache_page(vma, addr, pte_pfn(*ptep));
 141         ptep_clear_flush(vma, addr, ptep);
 142         set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
 143
 144         page_remove_rmap(page);
 145         if (!page_mapped(page))
 146                 try_to_free_swap(page);
 147         put_page(page);
 148         pte_unmap_unlock(ptep, ptl);
 149         err = 0;
 150
 151 out:
 152         return err;
 153 }
 154
 155 /*
 156  * NOTE:
 157  * Expect the breakpoint instruction to be the smallest size instruction for
 158  * the architecture. If an arch has variable length instruction and the
 159  * breakpoint instruction is not of the smallest length instruction
 160  * supported by that architecture then we need to modify read_opcode /
 161  * write_opcode accordingly. This would never be a problem for archs that
 162  * have fixed length instructions.
 163  */
 164
 165 /*
 166  * write_opcode - write the opcode at a given virtual address.
 167  * @mm: the probed process address space.
 168  * @uprobe: the breakpointing information.
 169  * @vaddr: the virtual address to store the opcode.
 170  * @opcode: opcode to be written at @vaddr.
 171  *
 172  * Called with mm->mmap_sem held (for read and with a reference to
 173  * mm).
 174  *
 175  * For mm @mm, write the opcode at @vaddr.
 176  * Return 0 (success) or a negative errno.
 177  */
 178 static int write_opcode(struct mm_struct *mm, struct uprobe *uprobe,
 179                         unsigned long vaddr, uprobe_opcode_t opcode)
 180 {
 181         struct page *old_page, *new_page;
 182         struct address_space *mapping;
 183         void *vaddr_old, *vaddr_new;
 184         struct vm_area_struct *vma;
 185         unsigned long addr;
 186         int ret;
 187
 188         /* Read the page with vaddr into memory */
 189         ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
 190         if (ret <= 0)
 191                 return ret;
 192         ret = -EINVAL;
 193
 194         /*
 195          * We are interested in text pages only. Our pages of interest
 196          * should be mapped for read and execute only. We desist from
 197          * adding probes in write mapped pages since the breakpoints
 198          * might end up in the file copy.
 199          */
 200         if (!valid_vma(vma, opcode == UPROBES_BKPT_INSN))
 201                 goto put_out;
 202
 203         mapping = uprobe->inode->i_mapping;
 204         if (mapping != vma->vm_file->f_mapping)
 205                 goto put_out;
 206
 207         addr = vma->vm_start + uprobe->offset;
 208         addr -= vma->vm_pgoff << PAGE_SHIFT;
 209         if (vaddr != (unsigned long)addr)
 210                 goto put_out;
 211
 212         ret = -ENOMEM;
 213         new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
 214         if (!new_page)
 215                 goto put_out;
 216
 217         __SetPageUptodate(new_page);
 218
 219         /*
 220          * lock page will serialize against do_wp_page()'s
 221          * PageAnon() handling
 222          */
 223         lock_page(old_page);
 224         /* copy the page now that we've got it stable */
 225         vaddr_old = kmap_atomic(old_page);
 226         vaddr_new = kmap_atomic(new_page);
 227
 228         memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
 229         /* poke the new insn in, ASSUMES we don't cross page boundary */
 230         vaddr &= ~PAGE_MASK;
 231         memcpy(vaddr_new + vaddr, &opcode, uprobe_opcode_sz);
 232
 233         kunmap_atomic(vaddr_new);
 234         kunmap_atomic(vaddr_old);
 235
 236         ret = anon_vma_prepare(vma);
 237         if (ret)
 238                 goto unlock_out;
 239
 240         lock_page(new_page);
 241         ret = __replace_page(vma, old_page, new_page);
 242         unlock_page(new_page);
 243
 244 unlock_out:
 245         unlock_page(old_page);
 246         page_cache_release(new_page);
 247
 248 put_out:
 249         put_page(old_page);     /* we did a get_page in the beginning */
 250         return ret;
 251 }
 252
 253 /**
 254  * read_opcode - read the opcode at a given virtual address.
 255  * @mm: the probed process address space.
 256  * @vaddr: the virtual address to read the opcode.
 257  * @opcode: location to store the read opcode.
 258  *
 259  * Called with mm->mmap_sem held (for read and with a reference to
 260  * mm.
 261  *
 262  * For mm @mm, read the opcode at @vaddr and store it in @opcode.
 263  * Return 0 (success) or a negative errno.
 264  */
 265 static int read_opcode(struct mm_struct *mm, unsigned long vaddr,
 266                                                 uprobe_opcode_t *opcode)
 267 {
 268         struct page *page;
 269         void *vaddr_new;
 270         int ret;
 271
 272         ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &page, NULL);
 273         if (ret <= 0)
 274                 return ret;
 275
 276         lock_page(page);
 277         vaddr_new = kmap_atomic(page);
 278         vaddr &= ~PAGE_MASK;
 279         memcpy(opcode, vaddr_new + vaddr, uprobe_opcode_sz);
 280         kunmap_atomic(vaddr_new);
 281         unlock_page(page);
 282         put_page(page);         /* we did a get_user_pages in the beginning */
 283         return 0;
 284 }
 285
 286 /**
 287  * set_bkpt - store breakpoint at a given address.
 288  * @mm: the probed process address space.
 289  * @uprobe: the probepoint information.
 290  * @vaddr: the virtual address to insert the opcode.
 291  *
 292  * For mm @mm, store the breakpoint instruction at @vaddr.
 293  * Return 0 (success) or a negative errno.
 294  */
 295 int __weak set_bkpt(struct mm_struct *mm, struct uprobe *uprobe,
 296                                                 unsigned long vaddr)
 297 {
 298         return write_opcode(mm, uprobe, vaddr, UPROBES_BKPT_INSN);
 299 }
 300
 301 /**
 302  * set_orig_insn - Restore the original instruction.
 303  * @mm: the probed process address space.
 304  * @uprobe: the probepoint information.
 305  * @vaddr: the virtual address to insert the opcode.
 306  * @verify: if true, verify existance of breakpoint instruction.
 307  *
 308  * For mm @mm, restore the original opcode (opcode) at @vaddr.
 309  * Return 0 (success) or a negative errno.
 310  */
 311 int __weak set_orig_insn(struct mm_struct *mm, struct uprobe *uprobe,
 312                                         unsigned long vaddr, bool verify)
 313 {
 314         if (verify) {
 315                 uprobe_opcode_t opcode;
 316                 int result = read_opcode(mm, vaddr, &opcode);
 317
 318                 if (result)
 319                         return result;
 320
 321                 if (opcode != UPROBES_BKPT_INSN)
 322                         return -EINVAL;
 323         }
 324         return write_opcode(mm, uprobe, vaddr,
 325                                 *(uprobe_opcode_t *)uprobe->insn);
 326 }
 327
 328 /**
 329  * is_bkpt_insn - check if instruction is breakpoint instruction.
 330  * @insn: instruction to be checked.
 331  * Default implementation of is_bkpt_insn
 332  * Returns true if @insn is a breakpoint instruction.
 333  */
 334 bool __weak is_bkpt_insn(u8 *insn)
 335 {
 336         return (insn[0] == UPROBES_BKPT_INSN);
 337 }
 338
 339 static int match_uprobe(struct uprobe *l, struct uprobe *r)
 340 {
 341         if (l->inode < r->inode)
 342                 return -1;
 343         if (l->inode > r->inode)
 344                 return 1;
 345         else {
 346                 if (l->offset < r->offset)
 347                         return -1;
 348
 349                 if (l->offset > r->offset)
 350                         return 1;
 351         }
 352
 353         return 0;
 354 }
 355
 356 static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
 357 {
 358         struct uprobe u = { .inode = inode, .offset = offset };
 359         struct rb_node *n = uprobes_tree.rb_node;
 360         struct uprobe *uprobe;
 361         int match;
 362
 363         while (n) {
 364                 uprobe = rb_entry(n, struct uprobe, rb_node);
 365                 match = match_uprobe(&u, uprobe);
 366                 if (!match) {
 367                         atomic_inc(&uprobe->ref);
 368                         return uprobe;
 369                 }
 370                 if (match < 0)
 371                         n = n->rb_left;
 372                 else
 373                         n = n->rb_right;
 374         }
 375         return NULL;
 376 }
 377
 378 /*
 379  * Find a uprobe corresponding to a given inode:offset
 380  * Acquires uprobes_treelock
 381  */
 382 static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
 383 {
 384         struct uprobe *uprobe;
 385         unsigned long flags;
 386
 387         spin_lock_irqsave(&uprobes_treelock, flags);
 388         uprobe = __find_uprobe(inode, offset);
 389         spin_unlock_irqrestore(&uprobes_treelock, flags);
 390         return uprobe;
 391 }
 392
 393 static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
 394 {
 395         struct rb_node **p = &uprobes_tree.rb_node;
 396         struct rb_node *parent = NULL;
 397         struct uprobe *u;
 398         int match;
 399
 400         while (*p) {
 401                 parent = *p;
 402                 u = rb_entry(parent, struct uprobe, rb_node);
 403                 match = match_uprobe(uprobe, u);
 404                 if (!match) {
 405                         atomic_inc(&u->ref);
 406                         return u;
 407                 }
 408
 409                 if (match < 0)
 410                         p = &parent->rb_left;
 411                 else
 412                         p = &parent->rb_right;
 413
 414         }
 415         u = NULL;
 416         rb_link_node(&uprobe->rb_node, parent, p);
 417         rb_insert_color(&uprobe->rb_node, &uprobes_tree);
 418         /* get access + creation ref */
 419         atomic_set(&uprobe->ref, 2);
 420         return u;
 421 }
 422
 423 /*
 424  * Acquires uprobes_treelock.
 425  * Matching uprobe already exists in rbtree;
 426  *      increment (access refcount) and return the matching uprobe.
 427  *
 428  * No matching uprobe; insert the uprobe in rb_tree;
 429  *      get a double refcount (access + creation) and return NULL.
 430  */
 431 static struct uprobe *insert_uprobe(struct uprobe *uprobe)
 432 {
 433         unsigned long flags;
 434         struct uprobe *u;
 435
 436         spin_lock_irqsave(&uprobes_treelock, flags);
 437         u = __insert_uprobe(uprobe);
 438         spin_unlock_irqrestore(&uprobes_treelock, flags);
 439
 440         /* For now assume that the instruction need not be single-stepped */
 441         uprobe->flags |= UPROBES_SKIP_SSTEP;
 442         return u;
 443 }
 444
 445 static void put_uprobe(struct uprobe *uprobe)
 446 {
 447         if (atomic_dec_and_test(&uprobe->ref))
 448                 kfree(uprobe);
 449 }
 450
 451 static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
 452 {
 453         struct uprobe *uprobe, *cur_uprobe;
 454
 455         uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
 456         if (!uprobe)
 457                 return NULL;
 458
 459         uprobe->inode = igrab(inode);
 460         uprobe->offset = offset;
 461         init_rwsem(&uprobe->consumer_rwsem);
 462         INIT_LIST_HEAD(&uprobe->pending_list);
 463
 464         /* add to uprobes_tree, sorted on inode:offset */
 465         cur_uprobe = insert_uprobe(uprobe);
 466
 467         /* a uprobe exists for this inode:offset combination */
 468         if (cur_uprobe) {
 469                 kfree(uprobe);
 470                 uprobe = cur_uprobe;
 471                 iput(inode);
 472         } else
 473                 atomic_inc(&uprobe_events);
 474         return uprobe;
 475 }
 476
 477 static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
 478 {
 479         struct uprobe_consumer *consumer;
 480
 481         if (!(uprobe->flags & UPROBES_RUN_HANDLER))
 482                 return;
 483
 484         down_read(&uprobe->consumer_rwsem);
 485         consumer = uprobe->consumers;
 486         for (consumer = uprobe->consumers; consumer;
 487                                         consumer = consumer->next) {
 488                 if (!consumer->filter ||
 489                                 consumer->filter(consumer, current))
 490                         consumer->handler(consumer, regs);
 491         }
 492         up_read(&uprobe->consumer_rwsem);
 493 }
 494
 495 /* Returns the previous consumer */
 496 static struct uprobe_consumer *add_consumer(struct uprobe *uprobe,
 497                                 struct uprobe_consumer *consumer)
 498 {
 499         down_write(&uprobe->consumer_rwsem);
 500         consumer->next = uprobe->consumers;
 501         uprobe->consumers = consumer;
 502         up_write(&uprobe->consumer_rwsem);
 503         return consumer->next;
 504 }
 505
 506 /*
 507  * For uprobe @uprobe, delete the consumer @consumer.
 508  * Return true if the @consumer is deleted successfully
 509  * or return false.
 510  */
 511 static bool del_consumer(struct uprobe *uprobe,
 512                                 struct uprobe_consumer *consumer)
 513 {
 514         struct uprobe_consumer **con;
 515         bool ret = false;
 516
 517         down_write(&uprobe->consumer_rwsem);
 518         for (con = &uprobe->consumers; *con; con = &(*con)->next) {
 519                 if (*con == consumer) {
 520                         *con = consumer->next;
 521                         ret = true;
 522                         break;
 523                 }
 524         }
 525         up_write(&uprobe->consumer_rwsem);
 526         return ret;
 527 }
 528
 529 static int __copy_insn(struct address_space *mapping,
 530                         struct vm_area_struct *vma, char *insn,
 531                         unsigned long nbytes, unsigned long offset)
 532 {
 533         struct file *filp = vma->vm_file;
 534         struct page *page;
 535         void *vaddr;
 536         unsigned long off1;
 537         unsigned long idx;
 538
 539         if (!filp)
 540                 return -EINVAL;
 541
 542         idx = (unsigned long)(offset >> PAGE_CACHE_SHIFT);
 543         off1 = offset &= ~PAGE_MASK;
 544
 545         /*
 546          * Ensure that the page that has the original instruction is
 547          * populated and in page-cache.
 548          */
 549         page = read_mapping_page(mapping, idx, filp);
 550         if (IS_ERR(page))
 551                 return -ENOMEM;
 552
 553         vaddr = kmap_atomic(page);
 554         memcpy(insn, vaddr + off1, nbytes);
 555         kunmap_atomic(vaddr);
 556         page_cache_release(page);
 557         return 0;
 558 }
 559
 560 static int copy_insn(struct uprobe *uprobe, struct vm_area_struct *vma,
 561                                         unsigned long addr)
 562 {
 563         struct address_space *mapping;
 564         int bytes;
 565         unsigned long nbytes;
 566
 567         addr &= ~PAGE_MASK;
 568         nbytes = PAGE_SIZE - addr;
 569         mapping = uprobe->inode->i_mapping;
 570
 571         /* Instruction at end of binary; copy only available bytes */
 572         if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
 573                 bytes = uprobe->inode->i_size - uprobe->offset;
 574         else
 575                 bytes = MAX_UINSN_BYTES;
 576
 577         /* Instruction at the page-boundary; copy bytes in second page */
 578         if (nbytes < bytes) {
 579                 if (__copy_insn(mapping, vma, uprobe->insn + nbytes,
 580                                 bytes - nbytes, uprobe->offset + nbytes))
 581                         return -ENOMEM;
 582
 583                 bytes = nbytes;
 584         }
 585         return __copy_insn(mapping, vma, uprobe->insn, bytes, uprobe->offset);
 586 }
 587
 588 static int install_breakpoint(struct mm_struct *mm, struct uprobe *uprobe,
 589                                 struct vm_area_struct *vma, loff_t vaddr)
 590 {
 591         unsigned long addr;
 592         int ret;
 593
 594         /*
 595          * Probe is to be deleted;
 596          * Dont know if somebody already inserted the probe;
 597          * behave as if probe already exists.
 598          */
 599         if (!uprobe->consumers)
 600                 return -EEXIST;
 601
 602         addr = (unsigned long)vaddr;
 603         if (!(uprobe->flags & UPROBES_COPY_INSN)) {
 604                 ret = copy_insn(uprobe, vma, addr);
 605                 if (ret)
 606                         return ret;
 607
 608                 if (is_bkpt_insn(uprobe->insn))
 609                         return -EEXIST;
 610
 611                 ret = analyze_insn(mm, uprobe);
 612                 if (ret)
 613                         return ret;
 614
 615                 uprobe->flags |= UPROBES_COPY_INSN;
 616         }
 617         ret = set_bkpt(mm, uprobe, addr);
 618         if (!ret)
 619                 atomic_inc(&mm->mm_uprobes_count);
 620
 621         return ret;
 622 }
 623
 624 static void remove_breakpoint(struct mm_struct *mm, struct uprobe *uprobe,
 625                                                         loff_t vaddr)
 626 {
 627         if (!set_orig_insn(mm, uprobe, (unsigned long)vaddr, true))
 628                 atomic_dec(&mm->mm_uprobes_count);
 629 }
 630
 631 /*
 632  * There could be threads that have hit the breakpoint and are entering the
 633  * notifier code and trying to acquire the uprobes_treelock. The thread
 634  * calling delete_uprobe() that is removing the uprobe from the rb_tree can
 635  * race with these threads and might acquire the uprobes_treelock compared
 636  * to some of the breakpoint hit threads. In such a case, the breakpoint hit
 637  * threads will not find the uprobe. Hence wait till the current breakpoint
 638  * hit threads acquire the uprobes_treelock before the uprobe is removed
 639  * from the rbtree.
 640  */
 641 static void delete_uprobe(struct uprobe *uprobe)
 642 {
 643         unsigned long flags;
 644
 645         bulkref_wait_old(&uprobes_srcu);
 646         spin_lock_irqsave(&uprobes_treelock, flags);
 647         rb_erase(&uprobe->rb_node, &uprobes_tree);
 648         spin_unlock_irqrestore(&uprobes_treelock, flags);
 649         iput(uprobe->inode);
 650         put_uprobe(uprobe);
 651         atomic_dec(&uprobe_events);
 652 }
 653
 654 static struct vma_info *__find_next_vma_info(struct list_head *head,
 655                         loff_t offset, struct address_space *mapping,
 656                         struct vma_info *vi, bool is_register)
 657 {
 658         struct prio_tree_iter iter;
 659         struct vm_area_struct *vma;
 660         struct vma_info *tmpvi;
 661         loff_t vaddr;
 662         unsigned long pgoff = offset >> PAGE_SHIFT;
 663         int existing_vma;
 664
 665         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 666                 if (!valid_vma(vma, is_register))
 667                         continue;
 668
 669                 existing_vma = 0;
 670                 vaddr = vma->vm_start + offset;
 671                 vaddr -= vma->vm_pgoff << PAGE_SHIFT;
 672                 list_for_each_entry(tmpvi, head, probe_list) {
 673                         if (tmpvi->mm == vma->vm_mm && tmpvi->vaddr == vaddr) {
 674                                 existing_vma = 1;
 675                                 break;
 676                         }
 677                 }
 678
 679                 /*
 680                  * Another vma needs a probe to be installed. However skip
 681                  * installing the probe if the vma is about to be unlinked.
 682                  */
 683                 if (!existing_vma &&
 684                                 atomic_inc_not_zero(&vma->vm_mm->mm_users)) {
 685                         vi->mm = vma->vm_mm;
 686                         vi->vaddr = vaddr;
 687                         list_add(&vi->probe_list, head);
 688                         return vi;
 689                 }
 690         }
 691         return NULL;
 692 }
 693
 694 /*
 695  * Iterate in the rmap prio tree  and find a vma where a probe has not
 696  * yet been inserted.
 697  */
 698 static struct vma_info *find_next_vma_info(struct list_head *head,
 699                         loff_t offset, struct address_space *mapping,
 700                         bool is_register)
 701 {
 702         struct vma_info *vi, *retvi;
 703         vi = kzalloc(sizeof(struct vma_info), GFP_KERNEL);
 704         if (!vi)
 705                 return ERR_PTR(-ENOMEM);
 706
 707         mutex_lock(&mapping->i_mmap_mutex);
 708         retvi = __find_next_vma_info(head, offset, mapping, vi, is_register);
 709         mutex_unlock(&mapping->i_mmap_mutex);
 710
 711         if (!retvi)
 712                 kfree(vi);
 713         return retvi;
 714 }
 715
 716 static int __register_uprobe(struct inode *inode, loff_t offset,
 717                                 struct uprobe *uprobe)
 718 {
 719         struct list_head try_list;
 720         struct vm_area_struct *vma;
 721         struct address_space *mapping;
 722         struct vma_info *vi, *tmpvi;
 723         struct mm_struct *mm;
 724         loff_t vaddr;
 725         int ret = 0;
 726
 727         mapping = inode->i_mapping;
 728         INIT_LIST_HEAD(&try_list);
 729         while ((vi = find_next_vma_info(&try_list, offset,
 730                                                 mapping, true)) != NULL) {
 731                 if (IS_ERR(vi)) {
 732                         ret = -ENOMEM;
 733                         break;
 734                 }
 735                 mm = vi->mm;
 736                 down_read(&mm->mmap_sem);
 737                 vma = find_vma(mm, (unsigned long)vi->vaddr);
 738                 if (!vma || !valid_vma(vma, true)) {
 739                         list_del(&vi->probe_list);
 740                         kfree(vi);
 741                         up_read(&mm->mmap_sem);
 742                         mmput(mm);
 743                         continue;
 744                 }
 745                 vaddr = vma->vm_start + offset;
 746                 vaddr -= vma->vm_pgoff << PAGE_SHIFT;
 747                 if (vma->vm_file->f_mapping->host != inode ||
 748                                                 vaddr != vi->vaddr) {
 749                         list_del(&vi->probe_list);
 750                         kfree(vi);
 751                         up_read(&mm->mmap_sem);
 752                         mmput(mm);
 753                         continue;
 754                 }
 755                 ret = install_breakpoint(mm, uprobe, vma, vi->vaddr);
 756                 up_read(&mm->mmap_sem);
 757                 mmput(mm);
 758                 if (ret && ret == -EEXIST)
 759                         ret = 0;
 760                 if (!ret)
 761                         break;
 762         }
 763         list_for_each_entry_safe(vi, tmpvi, &try_list, probe_list) {
 764                 list_del(&vi->probe_list);
 765                 kfree(vi);
 766         }
 767         return ret;
 768 }
 769
 770 static void __unregister_uprobe(struct inode *inode, loff_t offset,
 771                                                 struct uprobe *uprobe)
 772 {
 773         struct list_head try_list;
 774         struct address_space *mapping;
 775         struct vma_info *vi, *tmpvi;
 776         struct vm_area_struct *vma;
 777         struct mm_struct *mm;
 778         loff_t vaddr;
 779
 780         mapping = inode->i_mapping;
 781         INIT_LIST_HEAD(&try_list);
 782         while ((vi = find_next_vma_info(&try_list, offset,
 783                                                 mapping, false)) != NULL) {
 784                 if (IS_ERR(vi))
 785                         break;
 786                 mm = vi->mm;
 787                 down_read(&mm->mmap_sem);
 788                 vma = find_vma(mm, (unsigned long)vi->vaddr);
 789                 if (!vma || !valid_vma(vma, false)) {
 790                         list_del(&vi->probe_list);
 791                         kfree(vi);
 792                         up_read(&mm->mmap_sem);
 793                         mmput(mm);
 794                         continue;
 795                 }
 796                 vaddr = vma->vm_start + offset;
 797                 vaddr -= vma->vm_pgoff << PAGE_SHIFT;
 798                 if (vma->vm_file->f_mapping->host != inode ||
 799                                                 vaddr != vi->vaddr) {
 800                         list_del(&vi->probe_list);
 801                         kfree(vi);
 802                         up_read(&mm->mmap_sem);
 803                         mmput(mm);
 804                         continue;
 805                 }
 806                 remove_breakpoint(mm, uprobe, vi->vaddr);
 807                 up_read(&mm->mmap_sem);
 808                 mmput(mm);
 809         }
 810
 811         list_for_each_entry_safe(vi, tmpvi, &try_list, probe_list) {
 812                 list_del(&vi->probe_list);
 813                 kfree(vi);
 814         }
 815         delete_uprobe(uprobe);
 816 }
 817
 818 /*
 819  * register_uprobe - register a probe
 820  * @inode: the file in which the probe has to be placed.
 821  * @offset: offset from the start of the file.
 822  * @consumer: information on howto handle the probe..
 823  *
 824  * Apart from the access refcount, register_uprobe() takes a creation
 825  * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
 826  * inserted into the rbtree (i.e first consumer for a @inode:@offset
 827  * tuple).  Creation refcount stops unregister_uprobe from freeing the
 828  * @uprobe even before the register operation is complete. Creation
 829  * refcount is released when the last @consumer for the @uprobe
 830  * unregisters.
 831  *
 832  * Return errno if it cannot successully install probes
 833  * else return 0 (success)
 834  */
 835 int register_uprobe(struct inode *inode, loff_t offset,
 836                                 struct uprobe_consumer *consumer)
 837 {
 838         struct uprobe *uprobe;
 839         int ret = -EINVAL;
 840
 841         if (!consumer || consumer->next)
 842                 return ret;
 843
 844         inode = igrab(inode);
 845         if (!inode)
 846                 return ret;
 847
 848         if (offset > i_size_read(inode))
 849                 goto reg_out;
 850
 851         ret = 0;
 852         mutex_lock(uprobes_hash(inode));
 853         uprobe = alloc_uprobe(inode, offset);
 854         if (uprobe && !add_consumer(uprobe, consumer)) {
 855                 ret = __register_uprobe(inode, offset, uprobe);
 856                 if (ret) {
 857                         uprobe->consumers = NULL;
 858                         __unregister_uprobe(inode, offset, uprobe);
 859                 } else
 860                         uprobe->flags |= UPROBES_RUN_HANDLER;
 861         }
 862
 863         mutex_unlock(uprobes_hash(inode));
 864         put_uprobe(uprobe);
 865
 866 reg_out:
 867         iput(inode);
 868         return ret;
 869 }
 870
 871 /*
 872  * unregister_uprobe - unregister a already registered probe.
 873  * @inode: the file in which the probe has to be removed.
 874  * @offset: offset from the start of the file.
 875  * @consumer: identify which probe if multiple probes are colocated.
 876  */
 877 void unregister_uprobe(struct inode *inode, loff_t offset,
 878                                 struct uprobe_consumer *consumer)
 879 {
 880         struct uprobe *uprobe = NULL;
 881
 882         inode = igrab(inode);
 883         if (!inode || !consumer)
 884                 goto unreg_out;
 885
 886         uprobe = find_uprobe(inode, offset);
 887         if (!uprobe)
 888                 goto unreg_out;
 889
 890         mutex_lock(uprobes_hash(inode));
 891         if (!del_consumer(uprobe, consumer)) {
 892                 mutex_unlock(uprobes_hash(inode));
 893                 goto unreg_out;
 894         }
 895
 896         if (!uprobe->consumers) {
 897                 __unregister_uprobe(inode, offset, uprobe);
 898                 uprobe->flags &= ~UPROBES_RUN_HANDLER;
 899         }
 900         mutex_unlock(uprobes_hash(inode));
 901
 902 unreg_out:
 903         if (uprobe)
 904                 put_uprobe(uprobe);
 905         if (inode)
 906                 iput(inode);
 907 }
 908
 909 /*
 910  * Of all the nodes that correspond to the given inode, return the node
 911  * with the least offset.
 912  */
 913 static struct rb_node *find_least_offset_node(struct inode *inode)
 914 {
 915         struct uprobe u = { .inode = inode, .offset = 0};
 916         struct rb_node *n = uprobes_tree.rb_node;
 917         struct rb_node *close_node = NULL;
 918         struct uprobe *uprobe;
 919         int match;
 920
 921         while (n) {
 922                 uprobe = rb_entry(n, struct uprobe, rb_node);
 923                 match = match_uprobe(&u, uprobe);
 924                 if (uprobe->inode == inode)
 925                         close_node = n;
 926
 927                 if (!match)
 928                         return close_node;
 929
 930                 if (match < 0)
 931                         n = n->rb_left;
 932                 else
 933                         n = n->rb_right;
 934         }
 935         return close_node;
 936 }
 937
 938 /*
 939  * For a given inode, build a list of probes that need to be inserted.
 940  */
 941 static void build_probe_list(struct inode *inode, struct list_head *head)
 942 {
 943         struct uprobe *uprobe;
 944         struct rb_node *n;
 945         unsigned long flags;
 946
 947         spin_lock_irqsave(&uprobes_treelock, flags);
 948         n = find_least_offset_node(inode);
 949         for (; n; n = rb_next(n)) {
 950                 uprobe = rb_entry(n, struct uprobe, rb_node);
 951                 if (uprobe->inode != inode)
 952                         break;
 953
 954                 list_add(&uprobe->pending_list, head);
 955                 atomic_inc(&uprobe->ref);
 956         }
 957         spin_unlock_irqrestore(&uprobes_treelock, flags);
 958 }
 959
 960 /*
 961  * Called from mmap_region.
 962  * called with mm->mmap_sem acquired.
 963  *
 964  * Return -ve no if we fail to insert probes and we cannot
 965  * bail-out.
 966  * Return 0 otherwise. i.e :
 967  *      - successful insertion of probes
 968  *      - (or) no possible probes to be inserted.
 969  *      - (or) insertion of probes failed but we can bail-out.
 970  */
 971 int mmap_uprobe(struct vm_area_struct *vma)
 972 {
 973         struct list_head tmp_list;
 974         struct uprobe *uprobe, *u;
 975         struct inode *inode;
 976         int ret = 0, count = 0;
 977
 978         if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
 979                 return ret;     /* Bail-out */
 980
 981         inode = igrab(vma->vm_file->f_mapping->host);
 982         if (!inode)
 983                 return ret;
 984
 985         INIT_LIST_HEAD(&tmp_list);
 986         mutex_lock(uprobes_mmap_hash(inode));
 987         build_probe_list(inode, &tmp_list);
 988         list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
 989                 loff_t vaddr;
 990
 991                 list_del(&uprobe->pending_list);
 992                 if (!ret) {
 993                         vaddr = vma->vm_start + uprobe->offset;
 994                         vaddr -= vma->vm_pgoff << PAGE_SHIFT;
 995                         if (vaddr < vma->vm_start || vaddr >= vma->vm_end) {
 996                                 put_uprobe(uprobe);
 997                                 continue;
 998                         }
 999                         ret = install_breakpoint(vma->vm_mm, uprobe, vma,
1000                                                                 vaddr);
1001                         if (ret == -EEXIST) {
1002                                 atomic_inc(&vma->vm_mm->mm_uprobes_count);
1003                                 ret = 0;
1004                         }
1005                         if (!ret)
1006                                 count++;
1007                 }
1008                 put_uprobe(uprobe);
1009         }
1010
1011         mutex_unlock(uprobes_mmap_hash(inode));
1012         iput(inode);
1013         if (ret)
1014                 atomic_sub(count, &vma->vm_mm->mm_uprobes_count);
1015
1016         return ret;
1017 }
1018
1019 /*
1020  * Called in context of a munmap of a vma.
1021  */
1022 void munmap_uprobe(struct vm_area_struct *vma)
1023 {
1024         struct list_head tmp_list;
1025         struct uprobe *uprobe, *u;
1026         struct inode *inode;
1027
1028         if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1029                 return;         /* Bail-out */
1030
1031         if (!atomic_read(&vma->vm_mm->mm_uprobes_count))
1032                 return;
1033
1034         inode = igrab(vma->vm_file->f_mapping->host);
1035         if (!inode)
1036                 return;
1037
1038         INIT_LIST_HEAD(&tmp_list);
1039         mutex_lock(uprobes_mmap_hash(inode));
1040         build_probe_list(inode, &tmp_list);
1041         list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1042                 loff_t vaddr;
1043                 uprobe_opcode_t opcode;
1044
1045                 list_del(&uprobe->pending_list);
1046                 vaddr = vma->vm_start + uprobe->offset;
1047                 vaddr -= vma->vm_pgoff << PAGE_SHIFT;
1048                 if (vaddr >= vma->vm_start && vaddr < vma->vm_end) {
1049
1050                         /*
1051                          * An unregister could have removed the probe before
1052                          * unmap. So check before we decrement the count.
1053                          */
1054                         if (!read_opcode(vma->vm_mm, vaddr, &opcode) &&
1055                                                 (opcode == UPROBES_BKPT_INSN))
1056                                 atomic_dec(&vma->vm_mm->mm_uprobes_count);
1057                 }
1058                 put_uprobe(uprobe);
1059         }
1060         mutex_unlock(uprobes_mmap_hash(inode));
1061         iput(inode);
1062         return;
1063 }
1064
1065 /* Slot allocation for XOL */
1066 static int xol_add_vma(struct uprobes_xol_area *area)
1067 {
1068         struct mm_struct *mm;
1069         int ret;
1070
1071         area->page = alloc_page(GFP_HIGHUSER);
1072         if (!area->page)
1073                 return -ENOMEM;
1074
1075         mm = current->mm;
1076         down_write(&mm->mmap_sem);
1077         ret = -EALREADY;
1078         if (mm->uprobes_xol_area)
1079                 goto fail;
1080
1081         ret = -ENOMEM;
1082
1083         /* Try to map as high as possible, this is only a hint. */
1084         area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE,
1085                                                         PAGE_SIZE, 0, 0);
1086         if (area->vaddr & ~PAGE_MASK) {
1087                 ret = area->vaddr;
1088                 goto fail;
1089         }
1090
1091         ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1092                                 VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO,
1093                                 &area->page);
1094         if (ret)
1095                 goto fail;
1096
1097         smp_wmb();      /* pairs with get_uprobes_xol_area() */
1098         mm->uprobes_xol_area = area;
1099         ret = 0;
1100
1101 fail:
1102         up_write(&mm->mmap_sem);
1103         if (ret)
1104                 __free_page(area->page);
1105
1106         return ret;
1107 }
1108
1109 static struct uprobes_xol_area *get_uprobes_xol_area(struct mm_struct *mm)
1110 {
1111         struct uprobes_xol_area *area = mm->uprobes_xol_area;
1112         smp_read_barrier_depends();/* pairs with wmb in xol_add_vma() */
1113         return area;
1114 }
1115
1116 /*
1117  * xol_alloc_area - Allocate process's uprobes_xol_area.
1118  * This area will be used for storing instructions for execution out of
1119  * line.
1120  *
1121  * Returns the allocated area or NULL.
1122  */
1123 static struct uprobes_xol_area *xol_alloc_area(void)
1124 {
1125         struct uprobes_xol_area *area;
1126
1127         area = kzalloc(sizeof(*area), GFP_KERNEL);
1128         if (unlikely(!area))
1129                 return NULL;
1130
1131         area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long),
1132                                                                 GFP_KERNEL);
1133
1134         if (!area->bitmap)
1135                 goto fail;
1136
1137         init_waitqueue_head(&area->wq);
1138         if (!xol_add_vma(area))
1139                 return area;
1140
1141 fail:
1142         kfree(area->bitmap);
1143         kfree(area);
1144         return get_uprobes_xol_area(current->mm);
1145 }
1146
1147 /*
1148  * free_uprobes_xol_area - Free the area allocated for slots.
1149  */
1150 void free_uprobes_xol_area(struct mm_struct *mm)
1151 {
1152         struct uprobes_xol_area *area = mm->uprobes_xol_area;
1153
1154         if (!area)
1155                 return;
1156
1157         put_page(area->page);
1158         kfree(area->bitmap);
1159         kfree(area);
1160 }
1161
1162 /*
1163  *  - search for a free slot.
1164  */
1165 static unsigned long xol_take_insn_slot(struct uprobes_xol_area *area)
1166 {
1167         unsigned long slot_addr;
1168         int slot_nr;
1169
1170         do {
1171                 slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1172                 if (slot_nr < UINSNS_PER_PAGE) {
1173                         if (!test_and_set_bit(slot_nr, area->bitmap))
1174                                 break;
1175
1176                         slot_nr = UINSNS_PER_PAGE;
1177                         continue;
1178                 }
1179                 wait_event(area->wq,
1180                         (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1181         } while (slot_nr >= UINSNS_PER_PAGE);
1182
1183         slot_addr = area->vaddr + (slot_nr * UPROBES_XOL_SLOT_BYTES);
1184         atomic_inc(&area->slot_count);
1185         return slot_addr;
1186 }
1187
1188 /*
1189  * xol_get_insn_slot - If was not allocated a slot, then
1190  * allocate a slot.
1191  * Returns the allocated slot address or 0.
1192  */
1193 static unsigned long xol_get_insn_slot(struct uprobe *uprobe,
1194                                         unsigned long slot_addr)
1195 {
1196         struct uprobes_xol_area *area;
1197         unsigned long offset;
1198         void *vaddr;
1199
1200         area = get_uprobes_xol_area(current->mm);
1201         if (!area) {
1202                 area = xol_alloc_area();
1203                 if (!area)
1204                         return 0;
1205         }
1206         current->utask->xol_vaddr = xol_take_insn_slot(area);
1207
1208         /*
1209          * Initialize the slot if xol_vaddr points to valid
1210          * instruction slot.
1211          */
1212         if (unlikely(!current->utask->xol_vaddr))
1213                 return 0;
1214
1215         current->utask->vaddr = slot_addr;
1216         offset = current->utask->xol_vaddr & ~PAGE_MASK;
1217         vaddr = kmap_atomic(area->page);
1218         memcpy(vaddr + offset, uprobe->insn, MAX_UINSN_BYTES);
1219         kunmap_atomic(vaddr);
1220         return current->utask->xol_vaddr;
1221 }
1222
1223 /*
1224  * xol_free_insn_slot - If slot was earlier allocated by
1225  * @xol_get_insn_slot(), make the slot available for
1226  * subsequent requests.
1227  */
1228 static void xol_free_insn_slot(struct task_struct *tsk)
1229 {
1230         struct uprobes_xol_area *area;
1231         unsigned long vma_end;
1232         unsigned long slot_addr;
1233
1234         if (!tsk->mm || !tsk->mm->uprobes_xol_area || !tsk->utask)
1235                 return;
1236
1237         slot_addr = tsk->utask->xol_vaddr;
1238
1239         if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1240                 return;
1241
1242         area = tsk->mm->uprobes_xol_area;
1243         vma_end = area->vaddr + PAGE_SIZE;
1244         if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1245                 int slot_nr;
1246                 unsigned long offset = slot_addr - area->vaddr;
1247
1248                 slot_nr = offset / UPROBES_XOL_SLOT_BYTES;
1249                 if (slot_nr >= UINSNS_PER_PAGE)
1250                         return;
1251
1252                 clear_bit(slot_nr, area->bitmap);
1253                 atomic_dec(&area->slot_count);
1254                 if (waitqueue_active(&area->wq))
1255                         wake_up(&area->wq);
1256                 tsk->utask->xol_vaddr = 0;
1257         }
1258 }
1259
1260 /**
1261  * get_uprobe_bkpt_addr - compute address of bkpt given post-bkpt regs
1262  * @regs: Reflects the saved state of the task after it has hit a breakpoint
1263  * instruction.
1264  * Return the address of the breakpoint instruction.
1265  */
1266 unsigned long __weak get_uprobe_bkpt_addr(struct pt_regs *regs)
1267 {
1268         return instruction_pointer(regs) - UPROBES_BKPT_INSN_SIZE;
1269 }
1270
1271 /*
1272  * Called with no locks held.
1273  * Called in context of a exiting or a exec-ing thread.
1274  */
1275 void free_uprobe_utask(struct task_struct *tsk)
1276 {
1277         struct uprobe_task *utask = tsk->utask;
1278
1279         if (tsk->uprobes_bulkref_id != -1)
1280                 bulkref_put(&uprobes_srcu, tsk->uprobes_bulkref_id);
1281
1282         if (!utask)
1283                 return;
1284
1285         if (utask->active_uprobe)
1286                 put_uprobe(utask->active_uprobe);
1287
1288         xol_free_insn_slot(tsk);
1289         kfree(utask);
1290         tsk->utask = NULL;
1291 }
1292
1293 /*
1294  * Allocate a uprobe_task object for the task.
1295  * Called when the thread hits a breakpoint for the first time.
1296  *
1297  * Returns:
1298  * - pointer to new uprobe_task on success
1299  * - negative errno otherwise
1300  */
1301 static struct uprobe_task *add_utask(void)
1302 {
1303         struct uprobe_task *utask;
1304
1305         utask = kzalloc(sizeof *utask, GFP_KERNEL);
1306         if (unlikely(utask == NULL))
1307                 return ERR_PTR(-ENOMEM);
1308
1309         utask->active_uprobe = NULL;
1310         current->utask = utask;
1311         return utask;
1312 }
1313
1314 /* Prepare to single-step probed instruction out of line. */
1315 static int pre_ssout(struct uprobe *uprobe, struct pt_regs *regs,
1316                                 unsigned long vaddr)
1317 {
1318         if (xol_get_insn_slot(uprobe, vaddr) && !pre_xol(uprobe, regs))
1319                 return 0;
1320         return -EFAULT;
1321 }
1322
1323 bool uprobe_deny_signal(void)
1324 {
1325         struct task_struct *tsk = current;
1326         struct uprobe_task *utask = tsk->utask;
1327
1328         if (likely(!utask || !utask->active_uprobe))
1329                 return false;
1330
1331         WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1332
1333         if (signal_pending(tsk)) {
1334                 spin_lock_irq(&tsk->sighand->siglock);
1335                 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
1336                 spin_unlock_irq(&tsk->sighand->siglock);
1337
1338                 if (__fatal_signal_pending(tsk) || xol_was_trapped(tsk)) {
1339                         utask->state = UTASK_SSTEP_TRAPPED;
1340                         set_tsk_thread_flag(tsk, TIF_UPROBE);
1341                         set_tsk_thread_flag(tsk, TIF_NOTIFY_RESUME);
1342                 }
1343         }
1344
1345         return true;
1346 }
1347
1348 bool __weak can_skip_xol(struct pt_regs *regs, struct uprobe *u)
1349 {
1350         u->flags &= ~UPROBES_SKIP_SSTEP;
1351         return false;
1352 }
1353
1354 /*
1355  * uprobe_notify_resume gets called in task context just before returning
1356  * to userspace.
1357  *
1358  *  If its the first time the probepoint is hit, slot gets allocated here.
1359  *  If its the first time the thread hit a breakpoint, utask gets
1360  *  allocated here.
1361  */
1362 void uprobe_notify_resume(struct pt_regs *regs)
1363 {
1364         struct vm_area_struct *vma;
1365         struct uprobe_task *utask;
1366         struct mm_struct *mm;
1367         struct uprobe *u = NULL;
1368         unsigned long probept;
1369
1370         utask = current->utask;
1371         mm = current->mm;
1372         if (!utask || utask->state == UTASK_BP_HIT) {
1373                 probept = get_uprobe_bkpt_addr(regs);
1374                 down_read(&mm->mmap_sem);
1375                 vma = find_vma(mm, probept);
1376                 if (vma && valid_vma(vma, false))
1377                         u = find_uprobe(vma->vm_file->f_mapping->host,
1378                                         probept - vma->vm_start +
1379                                         (vma->vm_pgoff << PAGE_SHIFT));
1380
1381                 bulkref_put(&uprobes_srcu, current->uprobes_bulkref_id);
1382                 current->uprobes_bulkref_id = -1;
1383                 up_read(&mm->mmap_sem);
1384                 if (!u)
1385                         /* No matching uprobe; signal SIGTRAP. */
1386                         goto cleanup_ret;
1387                 if (!utask) {
1388                         utask = add_utask();
1389                         /* Cannot Allocate; re-execute the instruction. */
1390                         if (!utask)
1391                                 goto cleanup_ret;
1392                 }
1393                 utask->active_uprobe = u;
1394                 handler_chain(u, regs);
1395
1396                 if (u->flags & UPROBES_SKIP_SSTEP && can_skip_xol(regs, u))
1397                         goto cleanup_ret;
1398
1399                 utask->state = UTASK_SSTEP;
1400                 if (!pre_ssout(u, regs, probept))
1401                         user_enable_single_step(current);
1402                 else
1403                         /* Cannot Singlestep; re-execute the instruction. */
1404                         goto cleanup_ret;
1405         } else {
1406                 u = utask->active_uprobe;
1407                 if (utask->state == UTASK_SSTEP_ACK)
1408                         post_xol(u, regs);
1409                 else if (utask->state == UTASK_SSTEP_TRAPPED)
1410                         abort_xol(regs, u);
1411                 else
1412                         WARN_ON_ONCE(1);
1413
1414                 put_uprobe(u);
1415                 utask->active_uprobe = NULL;
1416                 utask->state = UTASK_RUNNING;
1417                 user_disable_single_step(current);
1418                 xol_free_insn_slot(current);
1419
1420                 spin_lock_irq(&current->sighand->siglock);
1421                 recalc_sigpending(); /* see uprobe_deny_signal() */
1422                 spin_unlock_irq(&current->sighand->siglock);
1423         }
1424         return;
1425
1426 cleanup_ret:
1427         if (utask) {
1428                 utask->active_uprobe = NULL;
1429                 utask->state = UTASK_RUNNING;
1430         }
1431         if (u) {
1432                 if (!(u->flags & UPROBES_SKIP_SSTEP))
1433                         set_instruction_pointer(regs, probept);
1434
1435                 put_uprobe(u);
1436         } else
1437                 send_sig(SIGTRAP, current, 0);
1438 }
1439
1440 /*
1441  * uprobe_bkpt_notifier gets called from interrupt context
1442  * it gets a reference to the ppt and sets TIF_UPROBE flag,
1443  */
1444 int uprobe_bkpt_notifier(struct pt_regs *regs)
1445 {
1446         struct uprobe_task *utask;
1447
1448         if (!current->mm || !atomic_read(&current->mm->mm_uprobes_count))
1449                 /* task is currently not uprobed */
1450                 return 0;
1451
1452         utask = current->utask;
1453         if (utask)
1454                 utask->state = UTASK_BP_HIT;
1455
1456         set_thread_flag(TIF_UPROBE);
1457         current->uprobes_bulkref_id = bulkref_get(&uprobes_srcu);
1458         return 1;
1459 }
1460
1461 /*
1462  * uprobe_post_notifier gets called in interrupt context.
1463  * It completes the single step operation.
1464  */
1465 int uprobe_post_notifier(struct pt_regs *regs)
1466 {
1467         struct uprobe_task *utask = current->utask;
1468
1469         if (!current->mm || !utask || !utask->active_uprobe)
1470                 /* task is currently not uprobed */
1471                 return 0;
1472
1473         utask->state = UTASK_SSTEP_ACK;
1474         set_thread_flag(TIF_UPROBE);
1475         return 1;
1476 }
1477
1478 struct notifier_block uprobe_exception_nb = {
1479         .notifier_call = uprobe_exception_notify,
1480         .priority = INT_MAX - 1,        /* notified after kprobes, kgdb */
1481 };
1482
1483 static int __init init_uprobes(void)
1484 {
1485         int i;
1486
1487         for (i = 0; i < UPROBES_HASH_SZ; i++) {
1488                 mutex_init(&uprobes_mutex[i]);
1489                 mutex_init(&uprobes_mmap_mutex[i]);
1490         }
1491         init_bulkref(&uprobes_srcu);
1492         return register_die_notifier(&uprobe_exception_nb);
1493 }
1494
1495 static void __exit exit_uprobes(void)
1496 {
1497 }
1498
1499 module_init(init_uprobes);
1500 module_exit(exit_uprobes);