linux/kernel/events/uprobes.c
<<
>>
Prefs
   1/*
   2 * User-space 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, 2008-2012
  19 * Authors:
  20 *      Srikar Dronamraju
  21 *      Jim Keniston
  22 * Copyright (C) 2011-2012 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
  23 */
  24
  25#include <linux/kernel.h>
  26#include <linux/highmem.h>
  27#include <linux/pagemap.h>      /* read_mapping_page */
  28#include <linux/slab.h>
  29#include <linux/sched.h>
  30#include <linux/rmap.h>         /* anon_vma_prepare */
  31#include <linux/mmu_notifier.h> /* set_pte_at_notify */
  32#include <linux/swap.h>         /* try_to_free_swap */
  33#include <linux/ptrace.h>       /* user_enable_single_step */
  34#include <linux/kdebug.h>       /* notifier mechanism */
  35#include "../../mm/internal.h"  /* munlock_vma_page */
  36
  37#include <linux/uprobes.h>
  38
  39#define UINSNS_PER_PAGE                 (PAGE_SIZE/UPROBE_XOL_SLOT_BYTES)
  40#define MAX_UPROBE_XOL_SLOTS            UINSNS_PER_PAGE
  41
  42static struct rb_root uprobes_tree = RB_ROOT;
  43
  44static DEFINE_SPINLOCK(uprobes_treelock);       /* serialize rbtree access */
  45
  46#define UPROBES_HASH_SZ 13
  47
  48/*
  49 * We need separate register/unregister and mmap/munmap lock hashes because
  50 * of mmap_sem nesting.
  51 *
  52 * uprobe_register() needs to install probes on (potentially) all processes
  53 * and thus needs to acquire multiple mmap_sems (consequtively, not
  54 * concurrently), whereas uprobe_mmap() is called while holding mmap_sem
  55 * for the particular process doing the mmap.
  56 *
  57 * uprobe_register()->register_for_each_vma() needs to drop/acquire mmap_sem
  58 * because of lock order against i_mmap_mutex. This means there's a hole in
  59 * the register vma iteration where a mmap() can happen.
  60 *
  61 * Thus uprobe_register() can race with uprobe_mmap() and we can try and
  62 * install a probe where one is already installed.
  63 */
  64
  65/* serialize (un)register */
  66static struct mutex uprobes_mutex[UPROBES_HASH_SZ];
  67
  68#define uprobes_hash(v)         (&uprobes_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
  69
  70/* serialize uprobe->pending_list */
  71static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
  72#define uprobes_mmap_hash(v)    (&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
  73
  74/*
  75 * uprobe_events allows us to skip the uprobe_mmap if there are no uprobe
  76 * events active at this time.  Probably a fine grained per inode count is
  77 * better?
  78 */
  79static atomic_t uprobe_events = ATOMIC_INIT(0);
  80
  81struct uprobe {
  82        struct rb_node          rb_node;        /* node in the rb tree */
  83        atomic_t                ref;
  84        struct rw_semaphore     consumer_rwsem;
  85        struct list_head        pending_list;
  86        struct uprobe_consumer  *consumers;
  87        struct inode            *inode;         /* Also hold a ref to inode */
  88        loff_t                  offset;
  89        int                     flags;
  90        struct arch_uprobe      arch;
  91};
  92
  93/*
  94 * valid_vma: Verify if the specified vma is an executable vma
  95 * Relax restrictions while unregistering: vm_flags might have
  96 * changed after breakpoint was inserted.
  97 *      - is_register: indicates if we are in register context.
  98 *      - Return 1 if the specified virtual address is in an
  99 *        executable vma.
 100 */
 101static bool valid_vma(struct vm_area_struct *vma, bool is_register)
 102{
 103        if (!vma->vm_file)
 104                return false;
 105
 106        if (!is_register)
 107                return true;
 108
 109        if ((vma->vm_flags & (VM_HUGETLB|VM_READ|VM_WRITE|VM_EXEC|VM_SHARED))
 110                                == (VM_READ|VM_EXEC))
 111                return true;
 112
 113        return false;
 114}
 115
 116static unsigned long offset_to_vaddr(struct vm_area_struct *vma, loff_t offset)
 117{
 118        return vma->vm_start + offset - ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 119}
 120
 121static loff_t vaddr_to_offset(struct vm_area_struct *vma, unsigned long vaddr)
 122{
 123        return ((loff_t)vma->vm_pgoff << PAGE_SHIFT) + (vaddr - vma->vm_start);
 124}
 125
 126/**
 127 * __replace_page - replace page in vma by new page.
 128 * based on replace_page in mm/ksm.c
 129 *
 130 * @vma:      vma that holds the pte pointing to page
 131 * @addr:     address the old @page is mapped at
 132 * @page:     the cowed page we are replacing by kpage
 133 * @kpage:    the modified page we replace page by
 134 *
 135 * Returns 0 on success, -EFAULT on failure.
 136 */
 137static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
 138                                struct page *page, struct page *kpage)
 139{
 140        struct mm_struct *mm = vma->vm_mm;
 141        spinlock_t *ptl;
 142        pte_t *ptep;
 143        int err;
 144
 145        /* For try_to_free_swap() and munlock_vma_page() below */
 146        lock_page(page);
 147
 148        err = -EAGAIN;
 149        ptep = page_check_address(page, mm, addr, &ptl, 0);
 150        if (!ptep)
 151                goto unlock;
 152
 153        get_page(kpage);
 154        page_add_new_anon_rmap(kpage, vma, addr);
 155
 156        if (!PageAnon(page)) {
 157                dec_mm_counter(mm, MM_FILEPAGES);
 158                inc_mm_counter(mm, MM_ANONPAGES);
 159        }
 160
 161        flush_cache_page(vma, addr, pte_pfn(*ptep));
 162        ptep_clear_flush(vma, addr, ptep);
 163        set_pte_at_notify(mm, addr, ptep, mk_pte(kpage, vma->vm_page_prot));
 164
 165        page_remove_rmap(page);
 166        if (!page_mapped(page))
 167                try_to_free_swap(page);
 168        pte_unmap_unlock(ptep, ptl);
 169
 170        if (vma->vm_flags & VM_LOCKED)
 171                munlock_vma_page(page);
 172        put_page(page);
 173
 174        err = 0;
 175 unlock:
 176        unlock_page(page);
 177        return err;
 178}
 179
 180/**
 181 * is_swbp_insn - check if instruction is breakpoint instruction.
 182 * @insn: instruction to be checked.
 183 * Default implementation of is_swbp_insn
 184 * Returns true if @insn is a breakpoint instruction.
 185 */
 186bool __weak is_swbp_insn(uprobe_opcode_t *insn)
 187{
 188        return *insn == UPROBE_SWBP_INSN;
 189}
 190
 191/*
 192 * NOTE:
 193 * Expect the breakpoint instruction to be the smallest size instruction for
 194 * the architecture. If an arch has variable length instruction and the
 195 * breakpoint instruction is not of the smallest length instruction
 196 * supported by that architecture then we need to modify read_opcode /
 197 * write_opcode accordingly. This would never be a problem for archs that
 198 * have fixed length instructions.
 199 */
 200
 201/*
 202 * write_opcode - write the opcode at a given virtual address.
 203 * @auprobe: arch breakpointing information.
 204 * @mm: the probed process address space.
 205 * @vaddr: the virtual address to store the opcode.
 206 * @opcode: opcode to be written at @vaddr.
 207 *
 208 * Called with mm->mmap_sem held (for read and with a reference to
 209 * mm).
 210 *
 211 * For mm @mm, write the opcode at @vaddr.
 212 * Return 0 (success) or a negative errno.
 213 */
 214static int write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
 215                        unsigned long vaddr, uprobe_opcode_t opcode)
 216{
 217        struct page *old_page, *new_page;
 218        void *vaddr_old, *vaddr_new;
 219        struct vm_area_struct *vma;
 220        int ret;
 221
 222retry:
 223        /* Read the page with vaddr into memory */
 224        ret = get_user_pages(NULL, mm, vaddr, 1, 0, 0, &old_page, &vma);
 225        if (ret <= 0)
 226                return ret;
 227
 228        ret = -ENOMEM;
 229        new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, vaddr);
 230        if (!new_page)
 231                goto put_old;
 232
 233        __SetPageUptodate(new_page);
 234
 235        /* copy the page now that we've got it stable */
 236        vaddr_old = kmap_atomic(old_page);
 237        vaddr_new = kmap_atomic(new_page);
 238
 239        memcpy(vaddr_new, vaddr_old, PAGE_SIZE);
 240        memcpy(vaddr_new + (vaddr & ~PAGE_MASK), &opcode, UPROBE_SWBP_INSN_SIZE);
 241
 242        kunmap_atomic(vaddr_new);
 243        kunmap_atomic(vaddr_old);
 244
 245        ret = anon_vma_prepare(vma);
 246        if (ret)
 247                goto put_new;
 248
 249        ret = __replace_page(vma, vaddr, old_page, new_page);
 250
 251put_new:
 252        page_cache_release(new_page);
 253put_old:
 254        put_page(old_page);
 255
 256        if (unlikely(ret == -EAGAIN))
 257                goto retry;
 258        return ret;
 259}
 260
 261/**
 262 * read_opcode - read the opcode at a given virtual address.
 263 * @mm: the probed process address space.
 264 * @vaddr: the virtual address to read the opcode.
 265 * @opcode: location to store the read opcode.
 266 *
 267 * Called with mm->mmap_sem held (for read and with a reference to
 268 * mm.
 269 *
 270 * For mm @mm, read the opcode at @vaddr and store it in @opcode.
 271 * Return 0 (success) or a negative errno.
 272 */
 273static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_t *opcode)
 274{
 275        struct page *page;
 276        void *vaddr_new;
 277        int ret;
 278
 279        ret = get_user_pages(NULL, mm, vaddr, 1, 0, 1, &page, NULL);
 280        if (ret <= 0)
 281                return ret;
 282
 283        lock_page(page);
 284        vaddr_new = kmap_atomic(page);
 285        vaddr &= ~PAGE_MASK;
 286        memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
 287        kunmap_atomic(vaddr_new);
 288        unlock_page(page);
 289
 290        put_page(page);
 291
 292        return 0;
 293}
 294
 295static int is_swbp_at_addr(struct mm_struct *mm, unsigned long vaddr)
 296{
 297        uprobe_opcode_t opcode;
 298        int result;
 299
 300        if (current->mm == mm) {
 301                pagefault_disable();
 302                result = __copy_from_user_inatomic(&opcode, (void __user*)vaddr,
 303                                                                sizeof(opcode));
 304                pagefault_enable();
 305
 306                if (likely(result == 0))
 307                        goto out;
 308        }
 309
 310        result = read_opcode(mm, vaddr, &opcode);
 311        if (result)
 312                return result;
 313out:
 314        if (is_swbp_insn(&opcode))
 315                return 1;
 316
 317        return 0;
 318}
 319
 320/**
 321 * set_swbp - store breakpoint at a given address.
 322 * @auprobe: arch specific probepoint information.
 323 * @mm: the probed process address space.
 324 * @vaddr: the virtual address to insert the opcode.
 325 *
 326 * For mm @mm, store the breakpoint instruction at @vaddr.
 327 * Return 0 (success) or a negative errno.
 328 */
 329int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
 330{
 331        int result;
 332        /*
 333         * See the comment near uprobes_hash().
 334         */
 335        result = is_swbp_at_addr(mm, vaddr);
 336        if (result == 1)
 337                return -EEXIST;
 338
 339        if (result)
 340                return result;
 341
 342        return write_opcode(auprobe, mm, vaddr, UPROBE_SWBP_INSN);
 343}
 344
 345/**
 346 * set_orig_insn - Restore the original instruction.
 347 * @mm: the probed process address space.
 348 * @auprobe: arch specific probepoint information.
 349 * @vaddr: the virtual address to insert the opcode.
 350 * @verify: if true, verify existance of breakpoint instruction.
 351 *
 352 * For mm @mm, restore the original opcode (opcode) at @vaddr.
 353 * Return 0 (success) or a negative errno.
 354 */
 355int __weak
 356set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify)
 357{
 358        if (verify) {
 359                int result;
 360
 361                result = is_swbp_at_addr(mm, vaddr);
 362                if (!result)
 363                        return -EINVAL;
 364
 365                if (result != 1)
 366                        return result;
 367        }
 368        return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
 369}
 370
 371static int match_uprobe(struct uprobe *l, struct uprobe *r)
 372{
 373        if (l->inode < r->inode)
 374                return -1;
 375
 376        if (l->inode > r->inode)
 377                return 1;
 378
 379        if (l->offset < r->offset)
 380                return -1;
 381
 382        if (l->offset > r->offset)
 383                return 1;
 384
 385        return 0;
 386}
 387
 388static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
 389{
 390        struct uprobe u = { .inode = inode, .offset = offset };
 391        struct rb_node *n = uprobes_tree.rb_node;
 392        struct uprobe *uprobe;
 393        int match;
 394
 395        while (n) {
 396                uprobe = rb_entry(n, struct uprobe, rb_node);
 397                match = match_uprobe(&u, uprobe);
 398                if (!match) {
 399                        atomic_inc(&uprobe->ref);
 400                        return uprobe;
 401                }
 402
 403                if (match < 0)
 404                        n = n->rb_left;
 405                else
 406                        n = n->rb_right;
 407        }
 408        return NULL;
 409}
 410
 411/*
 412 * Find a uprobe corresponding to a given inode:offset
 413 * Acquires uprobes_treelock
 414 */
 415static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
 416{
 417        struct uprobe *uprobe;
 418        unsigned long flags;
 419
 420        spin_lock_irqsave(&uprobes_treelock, flags);
 421        uprobe = __find_uprobe(inode, offset);
 422        spin_unlock_irqrestore(&uprobes_treelock, flags);
 423
 424        return uprobe;
 425}
 426
 427static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
 428{
 429        struct rb_node **p = &uprobes_tree.rb_node;
 430        struct rb_node *parent = NULL;
 431        struct uprobe *u;
 432        int match;
 433
 434        while (*p) {
 435                parent = *p;
 436                u = rb_entry(parent, struct uprobe, rb_node);
 437                match = match_uprobe(uprobe, u);
 438                if (!match) {
 439                        atomic_inc(&u->ref);
 440                        return u;
 441                }
 442
 443                if (match < 0)
 444                        p = &parent->rb_left;
 445                else
 446                        p = &parent->rb_right;
 447
 448        }
 449
 450        u = NULL;
 451        rb_link_node(&uprobe->rb_node, parent, p);
 452        rb_insert_color(&uprobe->rb_node, &uprobes_tree);
 453        /* get access + creation ref */
 454        atomic_set(&uprobe->ref, 2);
 455
 456        return u;
 457}
 458
 459/*
 460 * Acquire uprobes_treelock.
 461 * Matching uprobe already exists in rbtree;
 462 *      increment (access refcount) and return the matching uprobe.
 463 *
 464 * No matching uprobe; insert the uprobe in rb_tree;
 465 *      get a double refcount (access + creation) and return NULL.
 466 */
 467static struct uprobe *insert_uprobe(struct uprobe *uprobe)
 468{
 469        unsigned long flags;
 470        struct uprobe *u;
 471
 472        spin_lock_irqsave(&uprobes_treelock, flags);
 473        u = __insert_uprobe(uprobe);
 474        spin_unlock_irqrestore(&uprobes_treelock, flags);
 475
 476        /* For now assume that the instruction need not be single-stepped */
 477        uprobe->flags |= UPROBE_SKIP_SSTEP;
 478
 479        return u;
 480}
 481
 482static void put_uprobe(struct uprobe *uprobe)
 483{
 484        if (atomic_dec_and_test(&uprobe->ref))
 485                kfree(uprobe);
 486}
 487
 488static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset)
 489{
 490        struct uprobe *uprobe, *cur_uprobe;
 491
 492        uprobe = kzalloc(sizeof(struct uprobe), GFP_KERNEL);
 493        if (!uprobe)
 494                return NULL;
 495
 496        uprobe->inode = igrab(inode);
 497        uprobe->offset = offset;
 498        init_rwsem(&uprobe->consumer_rwsem);
 499
 500        /* add to uprobes_tree, sorted on inode:offset */
 501        cur_uprobe = insert_uprobe(uprobe);
 502
 503        /* a uprobe exists for this inode:offset combination */
 504        if (cur_uprobe) {
 505                kfree(uprobe);
 506                uprobe = cur_uprobe;
 507                iput(inode);
 508        } else {
 509                atomic_inc(&uprobe_events);
 510        }
 511
 512        return uprobe;
 513}
 514
 515static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
 516{
 517        struct uprobe_consumer *uc;
 518
 519        if (!(uprobe->flags & UPROBE_RUN_HANDLER))
 520                return;
 521
 522        down_read(&uprobe->consumer_rwsem);
 523        for (uc = uprobe->consumers; uc; uc = uc->next) {
 524                if (!uc->filter || uc->filter(uc, current))
 525                        uc->handler(uc, regs);
 526        }
 527        up_read(&uprobe->consumer_rwsem);
 528}
 529
 530/* Returns the previous consumer */
 531static struct uprobe_consumer *
 532consumer_add(struct uprobe *uprobe, struct uprobe_consumer *uc)
 533{
 534        down_write(&uprobe->consumer_rwsem);
 535        uc->next = uprobe->consumers;
 536        uprobe->consumers = uc;
 537        up_write(&uprobe->consumer_rwsem);
 538
 539        return uc->next;
 540}
 541
 542/*
 543 * For uprobe @uprobe, delete the consumer @uc.
 544 * Return true if the @uc is deleted successfully
 545 * or return false.
 546 */
 547static bool consumer_del(struct uprobe *uprobe, struct uprobe_consumer *uc)
 548{
 549        struct uprobe_consumer **con;
 550        bool ret = false;
 551
 552        down_write(&uprobe->consumer_rwsem);
 553        for (con = &uprobe->consumers; *con; con = &(*con)->next) {
 554                if (*con == uc) {
 555                        *con = uc->next;
 556                        ret = true;
 557                        break;
 558                }
 559        }
 560        up_write(&uprobe->consumer_rwsem);
 561
 562        return ret;
 563}
 564
 565static int
 566__copy_insn(struct address_space *mapping, struct file *filp, char *insn,
 567                        unsigned long nbytes, loff_t offset)
 568{
 569        struct page *page;
 570        void *vaddr;
 571        unsigned long off;
 572        pgoff_t idx;
 573
 574        if (!filp)
 575                return -EINVAL;
 576
 577        if (!mapping->a_ops->readpage)
 578                return -EIO;
 579
 580        idx = offset >> PAGE_CACHE_SHIFT;
 581        off = offset & ~PAGE_MASK;
 582
 583        /*
 584         * Ensure that the page that has the original instruction is
 585         * populated and in page-cache.
 586         */
 587        page = read_mapping_page(mapping, idx, filp);
 588        if (IS_ERR(page))
 589                return PTR_ERR(page);
 590
 591        vaddr = kmap_atomic(page);
 592        memcpy(insn, vaddr + off, nbytes);
 593        kunmap_atomic(vaddr);
 594        page_cache_release(page);
 595
 596        return 0;
 597}
 598
 599static int copy_insn(struct uprobe *uprobe, struct file *filp)
 600{
 601        struct address_space *mapping;
 602        unsigned long nbytes;
 603        int bytes;
 604
 605        nbytes = PAGE_SIZE - (uprobe->offset & ~PAGE_MASK);
 606        mapping = uprobe->inode->i_mapping;
 607
 608        /* Instruction at end of binary; copy only available bytes */
 609        if (uprobe->offset + MAX_UINSN_BYTES > uprobe->inode->i_size)
 610                bytes = uprobe->inode->i_size - uprobe->offset;
 611        else
 612                bytes = MAX_UINSN_BYTES;
 613
 614        /* Instruction at the page-boundary; copy bytes in second page */
 615        if (nbytes < bytes) {
 616                int err = __copy_insn(mapping, filp, uprobe->arch.insn + nbytes,
 617                                bytes - nbytes, uprobe->offset + nbytes);
 618                if (err)
 619                        return err;
 620                bytes = nbytes;
 621        }
 622        return __copy_insn(mapping, filp, uprobe->arch.insn, bytes, uprobe->offset);
 623}
 624
 625/*
 626 * How mm->uprobes_state.count gets updated
 627 * uprobe_mmap() increments the count if
 628 *      - it successfully adds a breakpoint.
 629 *      - it cannot add a breakpoint, but sees that there is a underlying
 630 *        breakpoint (via a is_swbp_at_addr()).
 631 *
 632 * uprobe_munmap() decrements the count if
 633 *      - it sees a underlying breakpoint, (via is_swbp_at_addr)
 634 *        (Subsequent uprobe_unregister wouldnt find the breakpoint
 635 *        unless a uprobe_mmap kicks in, since the old vma would be
 636 *        dropped just after uprobe_munmap.)
 637 *
 638 * uprobe_register increments the count if:
 639 *      - it successfully adds a breakpoint.
 640 *
 641 * uprobe_unregister decrements the count if:
 642 *      - it sees a underlying breakpoint and removes successfully.
 643 *        (via is_swbp_at_addr)
 644 *        (Subsequent uprobe_munmap wouldnt find the breakpoint
 645 *        since there is no underlying breakpoint after the
 646 *        breakpoint removal.)
 647 */
 648static int
 649install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
 650                        struct vm_area_struct *vma, unsigned long vaddr)
 651{
 652        int ret;
 653
 654        /*
 655         * If probe is being deleted, unregister thread could be done with
 656         * the vma-rmap-walk through. Adding a probe now can be fatal since
 657         * nobody will be able to cleanup. Also we could be from fork or
 658         * mremap path, where the probe might have already been inserted.
 659         * Hence behave as if probe already existed.
 660         */
 661        if (!uprobe->consumers)
 662                return -EEXIST;
 663
 664        if (!(uprobe->flags & UPROBE_COPY_INSN)) {
 665                ret = copy_insn(uprobe, vma->vm_file);
 666                if (ret)
 667                        return ret;
 668
 669                if (is_swbp_insn((uprobe_opcode_t *)uprobe->arch.insn))
 670                        return -ENOTSUPP;
 671
 672                ret = arch_uprobe_analyze_insn(&uprobe->arch, mm, vaddr);
 673                if (ret)
 674                        return ret;
 675
 676                /* write_opcode() assumes we don't cross page boundary */
 677                BUG_ON((uprobe->offset & ~PAGE_MASK) +
 678                                UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
 679
 680                uprobe->flags |= UPROBE_COPY_INSN;
 681        }
 682
 683        /*
 684         * Ideally, should be updating the probe count after the breakpoint
 685         * has been successfully inserted. However a thread could hit the
 686         * breakpoint we just inserted even before the probe count is
 687         * incremented. If this is the first breakpoint placed, breakpoint
 688         * notifier might ignore uprobes and pass the trap to the thread.
 689         * Hence increment before and decrement on failure.
 690         */
 691        atomic_inc(&mm->uprobes_state.count);
 692        ret = set_swbp(&uprobe->arch, mm, vaddr);
 693        if (ret)
 694                atomic_dec(&mm->uprobes_state.count);
 695
 696        return ret;
 697}
 698
 699static void
 700remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
 701{
 702        if (!set_orig_insn(&uprobe->arch, mm, vaddr, true))
 703                atomic_dec(&mm->uprobes_state.count);
 704}
 705
 706/*
 707 * There could be threads that have already hit the breakpoint. They
 708 * will recheck the current insn and restart if find_uprobe() fails.
 709 * See find_active_uprobe().
 710 */
 711static void delete_uprobe(struct uprobe *uprobe)
 712{
 713        unsigned long flags;
 714
 715        spin_lock_irqsave(&uprobes_treelock, flags);
 716        rb_erase(&uprobe->rb_node, &uprobes_tree);
 717        spin_unlock_irqrestore(&uprobes_treelock, flags);
 718        iput(uprobe->inode);
 719        put_uprobe(uprobe);
 720        atomic_dec(&uprobe_events);
 721}
 722
 723struct map_info {
 724        struct map_info *next;
 725        struct mm_struct *mm;
 726        unsigned long vaddr;
 727};
 728
 729static inline struct map_info *free_map_info(struct map_info *info)
 730{
 731        struct map_info *next = info->next;
 732        kfree(info);
 733        return next;
 734}
 735
 736static struct map_info *
 737build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
 738{
 739        unsigned long pgoff = offset >> PAGE_SHIFT;
 740        struct prio_tree_iter iter;
 741        struct vm_area_struct *vma;
 742        struct map_info *curr = NULL;
 743        struct map_info *prev = NULL;
 744        struct map_info *info;
 745        int more = 0;
 746
 747 again:
 748        mutex_lock(&mapping->i_mmap_mutex);
 749        vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 750                if (!valid_vma(vma, is_register))
 751                        continue;
 752
 753                if (!prev && !more) {
 754                        /*
 755                         * Needs GFP_NOWAIT to avoid i_mmap_mutex recursion through
 756                         * reclaim. This is optimistic, no harm done if it fails.
 757                         */
 758                        prev = kmalloc(sizeof(struct map_info),
 759                                        GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);
 760                        if (prev)
 761                                prev->next = NULL;
 762                }
 763                if (!prev) {
 764                        more++;
 765                        continue;
 766                }
 767
 768                if (!atomic_inc_not_zero(&vma->vm_mm->mm_users))
 769                        continue;
 770
 771                info = prev;
 772                prev = prev->next;
 773                info->next = curr;
 774                curr = info;
 775
 776                info->mm = vma->vm_mm;
 777                info->vaddr = offset_to_vaddr(vma, offset);
 778        }
 779        mutex_unlock(&mapping->i_mmap_mutex);
 780
 781        if (!more)
 782                goto out;
 783
 784        prev = curr;
 785        while (curr) {
 786                mmput(curr->mm);
 787                curr = curr->next;
 788        }
 789
 790        do {
 791                info = kmalloc(sizeof(struct map_info), GFP_KERNEL);
 792                if (!info) {
 793                        curr = ERR_PTR(-ENOMEM);
 794                        goto out;
 795                }
 796                info->next = prev;
 797                prev = info;
 798        } while (--more);
 799
 800        goto again;
 801 out:
 802        while (prev)
 803                prev = free_map_info(prev);
 804        return curr;
 805}
 806
 807static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
 808{
 809        struct map_info *info;
 810        int err = 0;
 811
 812        info = build_map_info(uprobe->inode->i_mapping,
 813                                        uprobe->offset, is_register);
 814        if (IS_ERR(info))
 815                return PTR_ERR(info);
 816
 817        while (info) {
 818                struct mm_struct *mm = info->mm;
 819                struct vm_area_struct *vma;
 820
 821                if (err)
 822                        goto free;
 823
 824                down_write(&mm->mmap_sem);
 825                vma = find_vma(mm, info->vaddr);
 826                if (!vma || !valid_vma(vma, is_register) ||
 827                    vma->vm_file->f_mapping->host != uprobe->inode)
 828                        goto unlock;
 829
 830                if (vma->vm_start > info->vaddr ||
 831                    vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
 832                        goto unlock;
 833
 834                if (is_register) {
 835                        err = install_breakpoint(uprobe, mm, vma, info->vaddr);
 836                        /*
 837                         * We can race against uprobe_mmap(), see the
 838                         * comment near uprobe_hash().
 839                         */
 840                        if (err == -EEXIST)
 841                                err = 0;
 842                } else {
 843                        remove_breakpoint(uprobe, mm, info->vaddr);
 844                }
 845 unlock:
 846                up_write(&mm->mmap_sem);
 847 free:
 848                mmput(mm);
 849                info = free_map_info(info);
 850        }
 851
 852        return err;
 853}
 854
 855static int __uprobe_register(struct uprobe *uprobe)
 856{
 857        return register_for_each_vma(uprobe, true);
 858}
 859
 860static void __uprobe_unregister(struct uprobe *uprobe)
 861{
 862        if (!register_for_each_vma(uprobe, false))
 863                delete_uprobe(uprobe);
 864
 865        /* TODO : cant unregister? schedule a worker thread */
 866}
 867
 868/*
 869 * uprobe_register - register a probe
 870 * @inode: the file in which the probe has to be placed.
 871 * @offset: offset from the start of the file.
 872 * @uc: information on howto handle the probe..
 873 *
 874 * Apart from the access refcount, uprobe_register() takes a creation
 875 * refcount (thro alloc_uprobe) if and only if this @uprobe is getting
 876 * inserted into the rbtree (i.e first consumer for a @inode:@offset
 877 * tuple).  Creation refcount stops uprobe_unregister from freeing the
 878 * @uprobe even before the register operation is complete. Creation
 879 * refcount is released when the last @uc for the @uprobe
 880 * unregisters.
 881 *
 882 * Return errno if it cannot successully install probes
 883 * else return 0 (success)
 884 */
 885int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
 886{
 887        struct uprobe *uprobe;
 888        int ret;
 889
 890        if (!inode || !uc || uc->next)
 891                return -EINVAL;
 892
 893        if (offset > i_size_read(inode))
 894                return -EINVAL;
 895
 896        ret = 0;
 897        mutex_lock(uprobes_hash(inode));
 898        uprobe = alloc_uprobe(inode, offset);
 899
 900        if (uprobe && !consumer_add(uprobe, uc)) {
 901                ret = __uprobe_register(uprobe);
 902                if (ret) {
 903                        uprobe->consumers = NULL;
 904                        __uprobe_unregister(uprobe);
 905                } else {
 906                        uprobe->flags |= UPROBE_RUN_HANDLER;
 907                }
 908        }
 909
 910        mutex_unlock(uprobes_hash(inode));
 911        put_uprobe(uprobe);
 912
 913        return ret;
 914}
 915
 916/*
 917 * uprobe_unregister - unregister a already registered probe.
 918 * @inode: the file in which the probe has to be removed.
 919 * @offset: offset from the start of the file.
 920 * @uc: identify which probe if multiple probes are colocated.
 921 */
 922void uprobe_unregister(struct inode *inode, loff_t offset, struct uprobe_consumer *uc)
 923{
 924        struct uprobe *uprobe;
 925
 926        if (!inode || !uc)
 927                return;
 928
 929        uprobe = find_uprobe(inode, offset);
 930        if (!uprobe)
 931                return;
 932
 933        mutex_lock(uprobes_hash(inode));
 934
 935        if (consumer_del(uprobe, uc)) {
 936                if (!uprobe->consumers) {
 937                        __uprobe_unregister(uprobe);
 938                        uprobe->flags &= ~UPROBE_RUN_HANDLER;
 939                }
 940        }
 941
 942        mutex_unlock(uprobes_hash(inode));
 943        if (uprobe)
 944                put_uprobe(uprobe);
 945}
 946
 947static struct rb_node *
 948find_node_in_range(struct inode *inode, loff_t min, loff_t max)
 949{
 950        struct rb_node *n = uprobes_tree.rb_node;
 951
 952        while (n) {
 953                struct uprobe *u = rb_entry(n, struct uprobe, rb_node);
 954
 955                if (inode < u->inode) {
 956                        n = n->rb_left;
 957                } else if (inode > u->inode) {
 958                        n = n->rb_right;
 959                } else {
 960                        if (max < u->offset)
 961                                n = n->rb_left;
 962                        else if (min > u->offset)
 963                                n = n->rb_right;
 964                        else
 965                                break;
 966                }
 967        }
 968
 969        return n;
 970}
 971
 972/*
 973 * For a given range in vma, build a list of probes that need to be inserted.
 974 */
 975static void build_probe_list(struct inode *inode,
 976                                struct vm_area_struct *vma,
 977                                unsigned long start, unsigned long end,
 978                                struct list_head *head)
 979{
 980        loff_t min, max;
 981        unsigned long flags;
 982        struct rb_node *n, *t;
 983        struct uprobe *u;
 984
 985        INIT_LIST_HEAD(head);
 986        min = vaddr_to_offset(vma, start);
 987        max = min + (end - start) - 1;
 988
 989        spin_lock_irqsave(&uprobes_treelock, flags);
 990        n = find_node_in_range(inode, min, max);
 991        if (n) {
 992                for (t = n; t; t = rb_prev(t)) {
 993                        u = rb_entry(t, struct uprobe, rb_node);
 994                        if (u->inode != inode || u->offset < min)
 995                                break;
 996                        list_add(&u->pending_list, head);
 997                        atomic_inc(&u->ref);
 998                }
 999                for (t = n; (t = rb_next(t)); ) {
1000                        u = rb_entry(t, struct uprobe, rb_node);
1001                        if (u->inode != inode || u->offset > max)
1002                                break;
1003                        list_add(&u->pending_list, head);
1004                        atomic_inc(&u->ref);
1005                }
1006        }
1007        spin_unlock_irqrestore(&uprobes_treelock, flags);
1008}
1009
1010/*
1011 * Called from mmap_region.
1012 * called with mm->mmap_sem acquired.
1013 *
1014 * Return -ve no if we fail to insert probes and we cannot
1015 * bail-out.
1016 * Return 0 otherwise. i.e:
1017 *
1018 *      - successful insertion of probes
1019 *      - (or) no possible probes to be inserted.
1020 *      - (or) insertion of probes failed but we can bail-out.
1021 */
1022int uprobe_mmap(struct vm_area_struct *vma)
1023{
1024        struct list_head tmp_list;
1025        struct uprobe *uprobe, *u;
1026        struct inode *inode;
1027        int ret, count;
1028
1029        if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1030                return 0;
1031
1032        inode = vma->vm_file->f_mapping->host;
1033        if (!inode)
1034                return 0;
1035
1036        mutex_lock(uprobes_mmap_hash(inode));
1037        build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1038
1039        ret = 0;
1040        count = 0;
1041
1042        list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1043                if (!ret) {
1044                        unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1045
1046                        ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1047                        /*
1048                         * We can race against uprobe_register(), see the
1049                         * comment near uprobe_hash().
1050                         */
1051                        if (ret == -EEXIST) {
1052                                ret = 0;
1053
1054                                if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1055                                        continue;
1056
1057                                /*
1058                                 * Unable to insert a breakpoint, but
1059                                 * breakpoint lies underneath. Increment the
1060                                 * probe count.
1061                                 */
1062                                atomic_inc(&vma->vm_mm->uprobes_state.count);
1063                        }
1064
1065                        if (!ret)
1066                                count++;
1067                }
1068                put_uprobe(uprobe);
1069        }
1070
1071        mutex_unlock(uprobes_mmap_hash(inode));
1072
1073        if (ret)
1074                atomic_sub(count, &vma->vm_mm->uprobes_state.count);
1075
1076        return ret;
1077}
1078
1079/*
1080 * Called in context of a munmap of a vma.
1081 */
1082void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1083{
1084        struct list_head tmp_list;
1085        struct uprobe *uprobe, *u;
1086        struct inode *inode;
1087
1088        if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1089                return;
1090
1091        if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1092                return;
1093
1094        if (!atomic_read(&vma->vm_mm->uprobes_state.count))
1095                return;
1096
1097        inode = vma->vm_file->f_mapping->host;
1098        if (!inode)
1099                return;
1100
1101        mutex_lock(uprobes_mmap_hash(inode));
1102        build_probe_list(inode, vma, start, end, &tmp_list);
1103
1104        list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1105                unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1106                /*
1107                 * An unregister could have removed the probe before
1108                 * unmap. So check before we decrement the count.
1109                 */
1110                if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1111                        atomic_dec(&vma->vm_mm->uprobes_state.count);
1112                put_uprobe(uprobe);
1113        }
1114        mutex_unlock(uprobes_mmap_hash(inode));
1115}
1116
1117/* Slot allocation for XOL */
1118static int xol_add_vma(struct xol_area *area)
1119{
1120        struct mm_struct *mm;
1121        int ret;
1122
1123        area->page = alloc_page(GFP_HIGHUSER);
1124        if (!area->page)
1125                return -ENOMEM;
1126
1127        ret = -EALREADY;
1128        mm = current->mm;
1129
1130        down_write(&mm->mmap_sem);
1131        if (mm->uprobes_state.xol_area)
1132                goto fail;
1133
1134        ret = -ENOMEM;
1135
1136        /* Try to map as high as possible, this is only a hint. */
1137        area->vaddr = get_unmapped_area(NULL, TASK_SIZE - PAGE_SIZE, PAGE_SIZE, 0, 0);
1138        if (area->vaddr & ~PAGE_MASK) {
1139                ret = area->vaddr;
1140                goto fail;
1141        }
1142
1143        ret = install_special_mapping(mm, area->vaddr, PAGE_SIZE,
1144                                VM_EXEC|VM_MAYEXEC|VM_DONTCOPY|VM_IO, &area->page);
1145        if (ret)
1146                goto fail;
1147
1148        smp_wmb();      /* pairs with get_xol_area() */
1149        mm->uprobes_state.xol_area = area;
1150        ret = 0;
1151
1152fail:
1153        up_write(&mm->mmap_sem);
1154        if (ret)
1155                __free_page(area->page);
1156
1157        return ret;
1158}
1159
1160static struct xol_area *get_xol_area(struct mm_struct *mm)
1161{
1162        struct xol_area *area;
1163
1164        area = mm->uprobes_state.xol_area;
1165        smp_read_barrier_depends();     /* pairs with wmb in xol_add_vma() */
1166
1167        return area;
1168}
1169
1170/*
1171 * xol_alloc_area - Allocate process's xol_area.
1172 * This area will be used for storing instructions for execution out of
1173 * line.
1174 *
1175 * Returns the allocated area or NULL.
1176 */
1177static struct xol_area *xol_alloc_area(void)
1178{
1179        struct xol_area *area;
1180
1181        area = kzalloc(sizeof(*area), GFP_KERNEL);
1182        if (unlikely(!area))
1183                return NULL;
1184
1185        area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long), GFP_KERNEL);
1186
1187        if (!area->bitmap)
1188                goto fail;
1189
1190        init_waitqueue_head(&area->wq);
1191        if (!xol_add_vma(area))
1192                return area;
1193
1194fail:
1195        kfree(area->bitmap);
1196        kfree(area);
1197
1198        return get_xol_area(current->mm);
1199}
1200
1201/*
1202 * uprobe_clear_state - Free the area allocated for slots.
1203 */
1204void uprobe_clear_state(struct mm_struct *mm)
1205{
1206        struct xol_area *area = mm->uprobes_state.xol_area;
1207
1208        if (!area)
1209                return;
1210
1211        put_page(area->page);
1212        kfree(area->bitmap);
1213        kfree(area);
1214}
1215
1216/*
1217 * uprobe_reset_state - Free the area allocated for slots.
1218 */
1219void uprobe_reset_state(struct mm_struct *mm)
1220{
1221        mm->uprobes_state.xol_area = NULL;
1222        atomic_set(&mm->uprobes_state.count, 0);
1223}
1224
1225/*
1226 *  - search for a free slot.
1227 */
1228static unsigned long xol_take_insn_slot(struct xol_area *area)
1229{
1230        unsigned long slot_addr;
1231        int slot_nr;
1232
1233        do {
1234                slot_nr = find_first_zero_bit(area->bitmap, UINSNS_PER_PAGE);
1235                if (slot_nr < UINSNS_PER_PAGE) {
1236                        if (!test_and_set_bit(slot_nr, area->bitmap))
1237                                break;
1238
1239                        slot_nr = UINSNS_PER_PAGE;
1240                        continue;
1241                }
1242                wait_event(area->wq, (atomic_read(&area->slot_count) < UINSNS_PER_PAGE));
1243        } while (slot_nr >= UINSNS_PER_PAGE);
1244
1245        slot_addr = area->vaddr + (slot_nr * UPROBE_XOL_SLOT_BYTES);
1246        atomic_inc(&area->slot_count);
1247
1248        return slot_addr;
1249}
1250
1251/*
1252 * xol_get_insn_slot - If was not allocated a slot, then
1253 * allocate a slot.
1254 * Returns the allocated slot address or 0.
1255 */
1256static unsigned long xol_get_insn_slot(struct uprobe *uprobe, unsigned long slot_addr)
1257{
1258        struct xol_area *area;
1259        unsigned long offset;
1260        void *vaddr;
1261
1262        area = get_xol_area(current->mm);
1263        if (!area) {
1264                area = xol_alloc_area();
1265                if (!area)
1266                        return 0;
1267        }
1268        current->utask->xol_vaddr = xol_take_insn_slot(area);
1269
1270        /*
1271         * Initialize the slot if xol_vaddr points to valid
1272         * instruction slot.
1273         */
1274        if (unlikely(!current->utask->xol_vaddr))
1275                return 0;
1276
1277        current->utask->vaddr = slot_addr;
1278        offset = current->utask->xol_vaddr & ~PAGE_MASK;
1279        vaddr = kmap_atomic(area->page);
1280        memcpy(vaddr + offset, uprobe->arch.insn, MAX_UINSN_BYTES);
1281        kunmap_atomic(vaddr);
1282
1283        return current->utask->xol_vaddr;
1284}
1285
1286/*
1287 * xol_free_insn_slot - If slot was earlier allocated by
1288 * @xol_get_insn_slot(), make the slot available for
1289 * subsequent requests.
1290 */
1291static void xol_free_insn_slot(struct task_struct *tsk)
1292{
1293        struct xol_area *area;
1294        unsigned long vma_end;
1295        unsigned long slot_addr;
1296
1297        if (!tsk->mm || !tsk->mm->uprobes_state.xol_area || !tsk->utask)
1298                return;
1299
1300        slot_addr = tsk->utask->xol_vaddr;
1301
1302        if (unlikely(!slot_addr || IS_ERR_VALUE(slot_addr)))
1303                return;
1304
1305        area = tsk->mm->uprobes_state.xol_area;
1306        vma_end = area->vaddr + PAGE_SIZE;
1307        if (area->vaddr <= slot_addr && slot_addr < vma_end) {
1308                unsigned long offset;
1309                int slot_nr;
1310
1311                offset = slot_addr - area->vaddr;
1312                slot_nr = offset / UPROBE_XOL_SLOT_BYTES;
1313                if (slot_nr >= UINSNS_PER_PAGE)
1314                        return;
1315
1316                clear_bit(slot_nr, area->bitmap);
1317                atomic_dec(&area->slot_count);
1318                if (waitqueue_active(&area->wq))
1319                        wake_up(&area->wq);
1320
1321                tsk->utask->xol_vaddr = 0;
1322        }
1323}
1324
1325/**
1326 * uprobe_get_swbp_addr - compute address of swbp given post-swbp regs
1327 * @regs: Reflects the saved state of the task after it has hit a breakpoint
1328 * instruction.
1329 * Return the address of the breakpoint instruction.
1330 */
1331unsigned long __weak uprobe_get_swbp_addr(struct pt_regs *regs)
1332{
1333        return instruction_pointer(regs) - UPROBE_SWBP_INSN_SIZE;
1334}
1335
1336/*
1337 * Called with no locks held.
1338 * Called in context of a exiting or a exec-ing thread.
1339 */
1340void uprobe_free_utask(struct task_struct *t)
1341{
1342        struct uprobe_task *utask = t->utask;
1343
1344        if (!utask)
1345                return;
1346
1347        if (utask->active_uprobe)
1348                put_uprobe(utask->active_uprobe);
1349
1350        xol_free_insn_slot(t);
1351        kfree(utask);
1352        t->utask = NULL;
1353}
1354
1355/*
1356 * Called in context of a new clone/fork from copy_process.
1357 */
1358void uprobe_copy_process(struct task_struct *t)
1359{
1360        t->utask = NULL;
1361}
1362
1363/*
1364 * Allocate a uprobe_task object for the task.
1365 * Called when the thread hits a breakpoint for the first time.
1366 *
1367 * Returns:
1368 * - pointer to new uprobe_task on success
1369 * - NULL otherwise
1370 */
1371static struct uprobe_task *add_utask(void)
1372{
1373        struct uprobe_task *utask;
1374
1375        utask = kzalloc(sizeof *utask, GFP_KERNEL);
1376        if (unlikely(!utask))
1377                return NULL;
1378
1379        current->utask = utask;
1380        return utask;
1381}
1382
1383/* Prepare to single-step probed instruction out of line. */
1384static int
1385pre_ssout(struct uprobe *uprobe, struct pt_regs *regs, unsigned long vaddr)
1386{
1387        if (xol_get_insn_slot(uprobe, vaddr) && !arch_uprobe_pre_xol(&uprobe->arch, regs))
1388                return 0;
1389
1390        return -EFAULT;
1391}
1392
1393/*
1394 * If we are singlestepping, then ensure this thread is not connected to
1395 * non-fatal signals until completion of singlestep.  When xol insn itself
1396 * triggers the signal,  restart the original insn even if the task is
1397 * already SIGKILL'ed (since coredump should report the correct ip).  This
1398 * is even more important if the task has a handler for SIGSEGV/etc, The
1399 * _same_ instruction should be repeated again after return from the signal
1400 * handler, and SSTEP can never finish in this case.
1401 */
1402bool uprobe_deny_signal(void)
1403{
1404        struct task_struct *t = current;
1405        struct uprobe_task *utask = t->utask;
1406
1407        if (likely(!utask || !utask->active_uprobe))
1408                return false;
1409
1410        WARN_ON_ONCE(utask->state != UTASK_SSTEP);
1411
1412        if (signal_pending(t)) {
1413                spin_lock_irq(&t->sighand->siglock);
1414                clear_tsk_thread_flag(t, TIF_SIGPENDING);
1415                spin_unlock_irq(&t->sighand->siglock);
1416
1417                if (__fatal_signal_pending(t) || arch_uprobe_xol_was_trapped(t)) {
1418                        utask->state = UTASK_SSTEP_TRAPPED;
1419                        set_tsk_thread_flag(t, TIF_UPROBE);
1420                        set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
1421                }
1422        }
1423
1424        return true;
1425}
1426
1427/*
1428 * Avoid singlestepping the original instruction if the original instruction
1429 * is a NOP or can be emulated.
1430 */
1431static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1432{
1433        if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
1434                return true;
1435
1436        uprobe->flags &= ~UPROBE_SKIP_SSTEP;
1437        return false;
1438}
1439
1440static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1441{
1442        struct mm_struct *mm = current->mm;
1443        struct uprobe *uprobe = NULL;
1444        struct vm_area_struct *vma;
1445
1446        down_read(&mm->mmap_sem);
1447        vma = find_vma(mm, bp_vaddr);
1448        if (vma && vma->vm_start <= bp_vaddr) {
1449                if (valid_vma(vma, false)) {
1450                        struct inode *inode = vma->vm_file->f_mapping->host;
1451                        loff_t offset = vaddr_to_offset(vma, bp_vaddr);
1452
1453                        uprobe = find_uprobe(inode, offset);
1454                }
1455
1456                if (!uprobe)
1457                        *is_swbp = is_swbp_at_addr(mm, bp_vaddr);
1458        } else {
1459                *is_swbp = -EFAULT;
1460        }
1461        up_read(&mm->mmap_sem);
1462
1463        return uprobe;
1464}
1465
1466/*
1467 * Run handler and ask thread to singlestep.
1468 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
1469 */
1470static void handle_swbp(struct pt_regs *regs)
1471{
1472        struct uprobe_task *utask;
1473        struct uprobe *uprobe;
1474        unsigned long bp_vaddr;
1475        int uninitialized_var(is_swbp);
1476
1477        bp_vaddr = uprobe_get_swbp_addr(regs);
1478        uprobe = find_active_uprobe(bp_vaddr, &is_swbp);
1479
1480        if (!uprobe) {
1481                if (is_swbp > 0) {
1482                        /* No matching uprobe; signal SIGTRAP. */
1483                        send_sig(SIGTRAP, current, 0);
1484                } else {
1485                        /*
1486                         * Either we raced with uprobe_unregister() or we can't
1487                         * access this memory. The latter is only possible if
1488                         * another thread plays with our ->mm. In both cases
1489                         * we can simply restart. If this vma was unmapped we
1490                         * can pretend this insn was not executed yet and get
1491                         * the (correct) SIGSEGV after restart.
1492                         */
1493                        instruction_pointer_set(regs, bp_vaddr);
1494                }
1495                return;
1496        }
1497
1498        utask = current->utask;
1499        if (!utask) {
1500                utask = add_utask();
1501                /* Cannot allocate; re-execute the instruction. */
1502                if (!utask)
1503                        goto cleanup_ret;
1504        }
1505        utask->active_uprobe = uprobe;
1506        handler_chain(uprobe, regs);
1507        if (uprobe->flags & UPROBE_SKIP_SSTEP && can_skip_sstep(uprobe, regs))
1508                goto cleanup_ret;
1509
1510        utask->state = UTASK_SSTEP;
1511        if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1512                user_enable_single_step(current);
1513                return;
1514        }
1515
1516cleanup_ret:
1517        if (utask) {
1518                utask->active_uprobe = NULL;
1519                utask->state = UTASK_RUNNING;
1520        }
1521        if (uprobe) {
1522                if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1523
1524                        /*
1525                         * cannot singlestep; cannot skip instruction;
1526                         * re-execute the instruction.
1527                         */
1528                        instruction_pointer_set(regs, bp_vaddr);
1529
1530                put_uprobe(uprobe);
1531        }
1532}
1533
1534/*
1535 * Perform required fix-ups and disable singlestep.
1536 * Allow pending signals to take effect.
1537 */
1538static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1539{
1540        struct uprobe *uprobe;
1541
1542        uprobe = utask->active_uprobe;
1543        if (utask->state == UTASK_SSTEP_ACK)
1544                arch_uprobe_post_xol(&uprobe->arch, regs);
1545        else if (utask->state == UTASK_SSTEP_TRAPPED)
1546                arch_uprobe_abort_xol(&uprobe->arch, regs);
1547        else
1548                WARN_ON_ONCE(1);
1549
1550        put_uprobe(uprobe);
1551        utask->active_uprobe = NULL;
1552        utask->state = UTASK_RUNNING;
1553        user_disable_single_step(current);
1554        xol_free_insn_slot(current);
1555
1556        spin_lock_irq(&current->sighand->siglock);
1557        recalc_sigpending(); /* see uprobe_deny_signal() */
1558        spin_unlock_irq(&current->sighand->siglock);
1559}
1560
1561/*
1562 * On breakpoint hit, breakpoint notifier sets the TIF_UPROBE flag.  (and on
1563 * subsequent probe hits on the thread sets the state to UTASK_BP_HIT) and
1564 * allows the thread to return from interrupt.
1565 *
1566 * On singlestep exception, singlestep notifier sets the TIF_UPROBE flag and
1567 * also sets the state to UTASK_SSTEP_ACK and allows the thread to return from
1568 * interrupt.
1569 *
1570 * While returning to userspace, thread notices the TIF_UPROBE flag and calls
1571 * uprobe_notify_resume().
1572 */
1573void uprobe_notify_resume(struct pt_regs *regs)
1574{
1575        struct uprobe_task *utask;
1576
1577        utask = current->utask;
1578        if (!utask || utask->state == UTASK_BP_HIT)
1579                handle_swbp(regs);
1580        else
1581                handle_singlestep(utask, regs);
1582}
1583
1584/*
1585 * uprobe_pre_sstep_notifier gets called from interrupt context as part of
1586 * notifier mechanism. Set TIF_UPROBE flag and indicate breakpoint hit.
1587 */
1588int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1589{
1590        struct uprobe_task *utask;
1591
1592        if (!current->mm || !atomic_read(&current->mm->uprobes_state.count))
1593                /* task is currently not uprobed */
1594                return 0;
1595
1596        utask = current->utask;
1597        if (utask)
1598                utask->state = UTASK_BP_HIT;
1599
1600        set_thread_flag(TIF_UPROBE);
1601
1602        return 1;
1603}
1604
1605/*
1606 * uprobe_post_sstep_notifier gets called in interrupt context as part of notifier
1607 * mechanism. Set TIF_UPROBE flag and indicate completion of singlestep.
1608 */
1609int uprobe_post_sstep_notifier(struct pt_regs *regs)
1610{
1611        struct uprobe_task *utask = current->utask;
1612
1613        if (!current->mm || !utask || !utask->active_uprobe)
1614                /* task is currently not uprobed */
1615                return 0;
1616
1617        utask->state = UTASK_SSTEP_ACK;
1618        set_thread_flag(TIF_UPROBE);
1619        return 1;
1620}
1621
1622static struct notifier_block uprobe_exception_nb = {
1623        .notifier_call          = arch_uprobe_exception_notify,
1624        .priority               = INT_MAX-1,    /* notified after kprobes, kgdb */
1625};
1626
1627static int __init init_uprobes(void)
1628{
1629        int i;
1630
1631        for (i = 0; i < UPROBES_HASH_SZ; i++) {
1632                mutex_init(&uprobes_mutex[i]);
1633                mutex_init(&uprobes_mmap_mutex[i]);
1634        }
1635
1636        return register_die_notifier(&uprobe_exception_nb);
1637}
1638module_init(init_uprobes);
1639
1640static void __exit exit_uprobes(void)
1641{
1642}
1643module_exit(exit_uprobes);
1644
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.