linux/mm/mmap.c
<<
>>
Prefs
   1/*
   2 * mm/mmap.c
   3 *
   4 * Written by obz.
   5 *
   6 * Address space accounting code        <alan@lxorguk.ukuu.org.uk>
   7 */
   8
   9#include <linux/slab.h>
  10#include <linux/backing-dev.h>
  11#include <linux/mm.h>
  12#include <linux/shm.h>
  13#include <linux/mman.h>
  14#include <linux/pagemap.h>
  15#include <linux/swap.h>
  16#include <linux/syscalls.h>
  17#include <linux/capability.h>
  18#include <linux/init.h>
  19#include <linux/file.h>
  20#include <linux/fs.h>
  21#include <linux/personality.h>
  22#include <linux/security.h>
  23#include <linux/hugetlb.h>
  24#include <linux/profile.h>
  25#include <linux/export.h>
  26#include <linux/mount.h>
  27#include <linux/mempolicy.h>
  28#include <linux/rmap.h>
  29#include <linux/mmu_notifier.h>
  30#include <linux/perf_event.h>
  31#include <linux/audit.h>
  32#include <linux/khugepaged.h>
  33#include <linux/uprobes.h>
  34
  35#include <asm/uaccess.h>
  36#include <asm/cacheflush.h>
  37#include <asm/tlb.h>
  38#include <asm/mmu_context.h>
  39
  40#include "internal.h"
  41
  42#ifndef arch_mmap_check
  43#define arch_mmap_check(addr, len, flags)       (0)
  44#endif
  45
  46#ifndef arch_rebalance_pgtables
  47#define arch_rebalance_pgtables(addr, len)              (addr)
  48#endif
  49
  50static void unmap_region(struct mm_struct *mm,
  51                struct vm_area_struct *vma, struct vm_area_struct *prev,
  52                unsigned long start, unsigned long end);
  53
  54/*
  55 * WARNING: the debugging will use recursive algorithms so never enable this
  56 * unless you know what you are doing.
  57 */
  58#undef DEBUG_MM_RB
  59
  60/* description of effects of mapping type and prot in current implementation.
  61 * this is due to the limited x86 page protection hardware.  The expected
  62 * behavior is in parens:
  63 *
  64 * map_type     prot
  65 *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  66 * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  67 *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  68 *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  69 *              
  70 * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  71 *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  72 *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  73 *
  74 */
  75pgprot_t protection_map[16] = {
  76        __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  77        __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  78};
  79
  80pgprot_t vm_get_page_prot(unsigned long vm_flags)
  81{
  82        return __pgprot(pgprot_val(protection_map[vm_flags &
  83                                (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
  84                        pgprot_val(arch_vm_get_page_prot(vm_flags)));
  85}
  86EXPORT_SYMBOL(vm_get_page_prot);
  87
  88int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;  /* heuristic overcommit */
  89int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
  90int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
  91/*
  92 * Make sure vm_committed_as in one cacheline and not cacheline shared with
  93 * other variables. It can be updated by several CPUs frequently.
  94 */
  95struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
  96
  97/*
  98 * Check that a process has enough memory to allocate a new virtual
  99 * mapping. 0 means there is enough memory for the allocation to
 100 * succeed and -ENOMEM implies there is not.
 101 *
 102 * We currently support three overcommit policies, which are set via the
 103 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 104 *
 105 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 106 * Additional code 2002 Jul 20 by Robert Love.
 107 *
 108 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 109 *
 110 * Note this is a helper function intended to be used by LSMs which
 111 * wish to use this logic.
 112 */
 113int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 114{
 115        unsigned long free, allowed;
 116
 117        vm_acct_memory(pages);
 118
 119        /*
 120         * Sometimes we want to use more memory than we have
 121         */
 122        if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 123                return 0;
 124
 125        if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 126                free = global_page_state(NR_FREE_PAGES);
 127                free += global_page_state(NR_FILE_PAGES);
 128
 129                /*
 130                 * shmem pages shouldn't be counted as free in this
 131                 * case, they can't be purged, only swapped out, and
 132                 * that won't affect the overall amount of available
 133                 * memory in the system.
 134                 */
 135                free -= global_page_state(NR_SHMEM);
 136
 137                free += nr_swap_pages;
 138
 139                /*
 140                 * Any slabs which are created with the
 141                 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 142                 * which are reclaimable, under pressure.  The dentry
 143                 * cache and most inode caches should fall into this
 144                 */
 145                free += global_page_state(NR_SLAB_RECLAIMABLE);
 146
 147                /*
 148                 * Leave reserved pages. The pages are not for anonymous pages.
 149                 */
 150                if (free <= totalreserve_pages)
 151                        goto error;
 152                else
 153                        free -= totalreserve_pages;
 154
 155                /*
 156                 * Leave the last 3% for root
 157                 */
 158                if (!cap_sys_admin)
 159                        free -= free / 32;
 160
 161                if (free > pages)
 162                        return 0;
 163
 164                goto error;
 165        }
 166
 167        allowed = (totalram_pages - hugetlb_total_pages())
 168                * sysctl_overcommit_ratio / 100;
 169        /*
 170         * Leave the last 3% for root
 171         */
 172        if (!cap_sys_admin)
 173                allowed -= allowed / 32;
 174        allowed += total_swap_pages;
 175
 176        /* Don't let a single process grow too big:
 177           leave 3% of the size of this process for other processes */
 178        if (mm)
 179                allowed -= mm->total_vm / 32;
 180
 181        if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 182                return 0;
 183error:
 184        vm_unacct_memory(pages);
 185
 186        return -ENOMEM;
 187}
 188
 189/*
 190 * Requires inode->i_mapping->i_mmap_mutex
 191 */
 192static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 193                struct file *file, struct address_space *mapping)
 194{
 195        if (vma->vm_flags & VM_DENYWRITE)
 196                atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
 197        if (vma->vm_flags & VM_SHARED)
 198                mapping->i_mmap_writable--;
 199
 200        flush_dcache_mmap_lock(mapping);
 201        if (unlikely(vma->vm_flags & VM_NONLINEAR))
 202                list_del_init(&vma->shared.vm_set.list);
 203        else
 204                vma_prio_tree_remove(vma, &mapping->i_mmap);
 205        flush_dcache_mmap_unlock(mapping);
 206}
 207
 208/*
 209 * Unlink a file-based vm structure from its prio_tree, to hide
 210 * vma from rmap and vmtruncate before freeing its page tables.
 211 */
 212void unlink_file_vma(struct vm_area_struct *vma)
 213{
 214        struct file *file = vma->vm_file;
 215
 216        if (file) {
 217                struct address_space *mapping = file->f_mapping;
 218                mutex_lock(&mapping->i_mmap_mutex);
 219                __remove_shared_vm_struct(vma, file, mapping);
 220                mutex_unlock(&mapping->i_mmap_mutex);
 221        }
 222}
 223
 224/*
 225 * Close a vm structure and free it, returning the next.
 226 */
 227static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 228{
 229        struct vm_area_struct *next = vma->vm_next;
 230
 231        might_sleep();
 232        if (vma->vm_ops && vma->vm_ops->close)
 233                vma->vm_ops->close(vma);
 234        if (vma->vm_file) {
 235                fput(vma->vm_file);
 236                if (vma->vm_flags & VM_EXECUTABLE)
 237                        removed_exe_file_vma(vma->vm_mm);
 238        }
 239        mpol_put(vma_policy(vma));
 240        kmem_cache_free(vm_area_cachep, vma);
 241        return next;
 242}
 243
 244static unsigned long do_brk(unsigned long addr, unsigned long len);
 245
 246SYSCALL_DEFINE1(brk, unsigned long, brk)
 247{
 248        unsigned long rlim, retval;
 249        unsigned long newbrk, oldbrk;
 250        struct mm_struct *mm = current->mm;
 251        unsigned long min_brk;
 252
 253        down_write(&mm->mmap_sem);
 254
 255#ifdef CONFIG_COMPAT_BRK
 256        /*
 257         * CONFIG_COMPAT_BRK can still be overridden by setting
 258         * randomize_va_space to 2, which will still cause mm->start_brk
 259         * to be arbitrarily shifted
 260         */
 261        if (current->brk_randomized)
 262                min_brk = mm->start_brk;
 263        else
 264                min_brk = mm->end_data;
 265#else
 266        min_brk = mm->start_brk;
 267#endif
 268        if (brk < min_brk)
 269                goto out;
 270
 271        /*
 272         * Check against rlimit here. If this check is done later after the test
 273         * of oldbrk with newbrk then it can escape the test and let the data
 274         * segment grow beyond its set limit the in case where the limit is
 275         * not page aligned -Ram Gupta
 276         */
 277        rlim = rlimit(RLIMIT_DATA);
 278        if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
 279                        (mm->end_data - mm->start_data) > rlim)
 280                goto out;
 281
 282        newbrk = PAGE_ALIGN(brk);
 283        oldbrk = PAGE_ALIGN(mm->brk);
 284        if (oldbrk == newbrk)
 285                goto set_brk;
 286
 287        /* Always allow shrinking brk. */
 288        if (brk <= mm->brk) {
 289                if (!do_munmap(mm, newbrk, oldbrk-newbrk))
 290                        goto set_brk;
 291                goto out;
 292        }
 293
 294        /* Check against existing mmap mappings. */
 295        if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 296                goto out;
 297
 298        /* Ok, looks good - let it rip. */
 299        if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
 300                goto out;
 301set_brk:
 302        mm->brk = brk;
 303out:
 304        retval = mm->brk;
 305        up_write(&mm->mmap_sem);
 306        return retval;
 307}
 308
 309#ifdef DEBUG_MM_RB
 310static int browse_rb(struct rb_root *root)
 311{
 312        int i = 0, j;
 313        struct rb_node *nd, *pn = NULL;
 314        unsigned long prev = 0, pend = 0;
 315
 316        for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 317                struct vm_area_struct *vma;
 318                vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 319                if (vma->vm_start < prev)
 320                        printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
 321                if (vma->vm_start < pend)
 322                        printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
 323                if (vma->vm_start > vma->vm_end)
 324                        printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
 325                i++;
 326                pn = nd;
 327                prev = vma->vm_start;
 328                pend = vma->vm_end;
 329        }
 330        j = 0;
 331        for (nd = pn; nd; nd = rb_prev(nd)) {
 332                j++;
 333        }
 334        if (i != j)
 335                printk("backwards %d, forwards %d\n", j, i), i = 0;
 336        return i;
 337}
 338
 339void validate_mm(struct mm_struct *mm)
 340{
 341        int bug = 0;
 342        int i = 0;
 343        struct vm_area_struct *tmp = mm->mmap;
 344        while (tmp) {
 345                tmp = tmp->vm_next;
 346                i++;
 347        }
 348        if (i != mm->map_count)
 349                printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
 350        i = browse_rb(&mm->mm_rb);
 351        if (i != mm->map_count)
 352                printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
 353        BUG_ON(bug);
 354}
 355#else
 356#define validate_mm(mm) do { } while (0)
 357#endif
 358
 359static struct vm_area_struct *
 360find_vma_prepare(struct mm_struct *mm, unsigned long addr,
 361                struct vm_area_struct **pprev, struct rb_node ***rb_link,
 362                struct rb_node ** rb_parent)
 363{
 364        struct vm_area_struct * vma;
 365        struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
 366
 367        __rb_link = &mm->mm_rb.rb_node;
 368        rb_prev = __rb_parent = NULL;
 369        vma = NULL;
 370
 371        while (*__rb_link) {
 372                struct vm_area_struct *vma_tmp;
 373
 374                __rb_parent = *__rb_link;
 375                vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
 376
 377                if (vma_tmp->vm_end > addr) {
 378                        vma = vma_tmp;
 379                        if (vma_tmp->vm_start <= addr)
 380                                break;
 381                        __rb_link = &__rb_parent->rb_left;
 382                } else {
 383                        rb_prev = __rb_parent;
 384                        __rb_link = &__rb_parent->rb_right;
 385                }
 386        }
 387
 388        *pprev = NULL;
 389        if (rb_prev)
 390                *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 391        *rb_link = __rb_link;
 392        *rb_parent = __rb_parent;
 393        return vma;
 394}
 395
 396void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
 397                struct rb_node **rb_link, struct rb_node *rb_parent)
 398{
 399        rb_link_node(&vma->vm_rb, rb_parent, rb_link);
 400        rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 401}
 402
 403static void __vma_link_file(struct vm_area_struct *vma)
 404{
 405        struct file *file;
 406
 407        file = vma->vm_file;
 408        if (file) {
 409                struct address_space *mapping = file->f_mapping;
 410
 411                if (vma->vm_flags & VM_DENYWRITE)
 412                        atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
 413                if (vma->vm_flags & VM_SHARED)
 414                        mapping->i_mmap_writable++;
 415
 416                flush_dcache_mmap_lock(mapping);
 417                if (unlikely(vma->vm_flags & VM_NONLINEAR))
 418                        vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 419                else
 420                        vma_prio_tree_insert(vma, &mapping->i_mmap);
 421                flush_dcache_mmap_unlock(mapping);
 422        }
 423}
 424
 425static void
 426__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 427        struct vm_area_struct *prev, struct rb_node **rb_link,
 428        struct rb_node *rb_parent)
 429{
 430        __vma_link_list(mm, vma, prev, rb_parent);
 431        __vma_link_rb(mm, vma, rb_link, rb_parent);
 432}
 433
 434static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 435                        struct vm_area_struct *prev, struct rb_node **rb_link,
 436                        struct rb_node *rb_parent)
 437{
 438        struct address_space *mapping = NULL;
 439
 440        if (vma->vm_file)
 441                mapping = vma->vm_file->f_mapping;
 442
 443        if (mapping)
 444                mutex_lock(&mapping->i_mmap_mutex);
 445
 446        __vma_link(mm, vma, prev, rb_link, rb_parent);
 447        __vma_link_file(vma);
 448
 449        if (mapping)
 450                mutex_unlock(&mapping->i_mmap_mutex);
 451
 452        mm->map_count++;
 453        validate_mm(mm);
 454}
 455
 456/*
 457 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
 458 * mm's list and rbtree.  It has already been inserted into the prio_tree.
 459 */
 460static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
 461{
 462        struct vm_area_struct *__vma, *prev;
 463        struct rb_node **rb_link, *rb_parent;
 464
 465        __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
 466        BUG_ON(__vma && __vma->vm_start < vma->vm_end);
 467        __vma_link(mm, vma, prev, rb_link, rb_parent);
 468        mm->map_count++;
 469}
 470
 471static inline void
 472__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
 473                struct vm_area_struct *prev)
 474{
 475        struct vm_area_struct *next = vma->vm_next;
 476
 477        prev->vm_next = next;
 478        if (next)
 479                next->vm_prev = prev;
 480        rb_erase(&vma->vm_rb, &mm->mm_rb);
 481        if (mm->mmap_cache == vma)
 482                mm->mmap_cache = prev;
 483}
 484
 485/*
 486 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
 487 * is already present in an i_mmap tree without adjusting the tree.
 488 * The following helper function should be used when such adjustments
 489 * are necessary.  The "insert" vma (if any) is to be inserted
 490 * before we drop the necessary locks.
 491 */
 492int vma_adjust(struct vm_area_struct *vma, unsigned long start,
 493        unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
 494{
 495        struct mm_struct *mm = vma->vm_mm;
 496        struct vm_area_struct *next = vma->vm_next;
 497        struct vm_area_struct *importer = NULL;
 498        struct address_space *mapping = NULL;
 499        struct prio_tree_root *root = NULL;
 500        struct anon_vma *anon_vma = NULL;
 501        struct file *file = vma->vm_file;
 502        long adjust_next = 0;
 503        int remove_next = 0;
 504
 505        if (next && !insert) {
 506                struct vm_area_struct *exporter = NULL;
 507
 508                if (end >= next->vm_end) {
 509                        /*
 510                         * vma expands, overlapping all the next, and
 511                         * perhaps the one after too (mprotect case 6).
 512                         */
 513again:                  remove_next = 1 + (end > next->vm_end);
 514                        end = next->vm_end;
 515                        exporter = next;
 516                        importer = vma;
 517                } else if (end > next->vm_start) {
 518                        /*
 519                         * vma expands, overlapping part of the next:
 520                         * mprotect case 5 shifting the boundary up.
 521                         */
 522                        adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
 523                        exporter = next;
 524                        importer = vma;
 525                } else if (end < vma->vm_end) {
 526                        /*
 527                         * vma shrinks, and !insert tells it's not
 528                         * split_vma inserting another: so it must be
 529                         * mprotect case 4 shifting the boundary down.
 530                         */
 531                        adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
 532                        exporter = vma;
 533                        importer = next;
 534                }
 535
 536                /*
 537                 * Easily overlooked: when mprotect shifts the boundary,
 538                 * make sure the expanding vma has anon_vma set if the
 539                 * shrinking vma had, to cover any anon pages imported.
 540                 */
 541                if (exporter && exporter->anon_vma && !importer->anon_vma) {
 542                        if (anon_vma_clone(importer, exporter))
 543                                return -ENOMEM;
 544                        importer->anon_vma = exporter->anon_vma;
 545                }
 546        }
 547
 548        if (file) {
 549                mapping = file->f_mapping;
 550                if (!(vma->vm_flags & VM_NONLINEAR)) {
 551                        root = &mapping->i_mmap;
 552                        uprobe_munmap(vma, vma->vm_start, vma->vm_end);
 553
 554                        if (adjust_next)
 555                                uprobe_munmap(next, next->vm_start,
 556                                                        next->vm_end);
 557                }
 558
 559                mutex_lock(&mapping->i_mmap_mutex);
 560                if (insert) {
 561                        /*
 562                         * Put into prio_tree now, so instantiated pages
 563                         * are visible to arm/parisc __flush_dcache_page
 564                         * throughout; but we cannot insert into address
 565                         * space until vma start or end is updated.
 566                         */
 567                        __vma_link_file(insert);
 568                }
 569        }
 570
 571        vma_adjust_trans_huge(vma, start, end, adjust_next);
 572
 573        /*
 574         * When changing only vma->vm_end, we don't really need anon_vma
 575         * lock. This is a fairly rare case by itself, but the anon_vma
 576         * lock may be shared between many sibling processes.  Skipping
 577         * the lock for brk adjustments makes a difference sometimes.
 578         */
 579        if (vma->anon_vma && (importer || start != vma->vm_start)) {
 580                anon_vma = vma->anon_vma;
 581                anon_vma_lock(anon_vma);
 582        }
 583
 584        if (root) {
 585                flush_dcache_mmap_lock(mapping);
 586                vma_prio_tree_remove(vma, root);
 587                if (adjust_next)
 588                        vma_prio_tree_remove(next, root);
 589        }
 590
 591        vma->vm_start = start;
 592        vma->vm_end = end;
 593        vma->vm_pgoff = pgoff;
 594        if (adjust_next) {
 595                next->vm_start += adjust_next << PAGE_SHIFT;
 596                next->vm_pgoff += adjust_next;
 597        }
 598
 599        if (root) {
 600                if (adjust_next)
 601                        vma_prio_tree_insert(next, root);
 602                vma_prio_tree_insert(vma, root);
 603                flush_dcache_mmap_unlock(mapping);
 604        }
 605
 606        if (remove_next) {
 607                /*
 608                 * vma_merge has merged next into vma, and needs
 609                 * us to remove next before dropping the locks.
 610                 */
 611                __vma_unlink(mm, next, vma);
 612                if (file)
 613                        __remove_shared_vm_struct(next, file, mapping);
 614        } else if (insert) {
 615                /*
 616                 * split_vma has split insert from vma, and needs
 617                 * us to insert it before dropping the locks
 618                 * (it may either follow vma or precede it).
 619                 */
 620                __insert_vm_struct(mm, insert);
 621        }
 622
 623        if (anon_vma)
 624                anon_vma_unlock(anon_vma);
 625        if (mapping)
 626                mutex_unlock(&mapping->i_mmap_mutex);
 627
 628        if (root) {
 629                uprobe_mmap(vma);
 630
 631                if (adjust_next)
 632                        uprobe_mmap(next);
 633        }
 634
 635        if (remove_next) {
 636                if (file) {
 637                        uprobe_munmap(next, next->vm_start, next->vm_end);
 638                        fput(file);
 639                        if (next->vm_flags & VM_EXECUTABLE)
 640                                removed_exe_file_vma(mm);
 641                }
 642                if (next->anon_vma)
 643                        anon_vma_merge(vma, next);
 644                mm->map_count--;
 645                mpol_put(vma_policy(next));
 646                kmem_cache_free(vm_area_cachep, next);
 647                /*
 648                 * In mprotect's case 6 (see comments on vma_merge),
 649                 * we must remove another next too. It would clutter
 650                 * up the code too much to do both in one go.
 651                 */
 652                if (remove_next == 2) {
 653                        next = vma->vm_next;
 654                        goto again;
 655                }
 656        }
 657        if (insert && file)
 658                uprobe_mmap(insert);
 659
 660        validate_mm(mm);
 661
 662        return 0;
 663}
 664
 665/*
 666 * If the vma has a ->close operation then the driver probably needs to release
 667 * per-vma resources, so we don't attempt to merge those.
 668 */
 669static inline int is_mergeable_vma(struct vm_area_struct *vma,
 670                        struct file *file, unsigned long vm_flags)
 671{
 672        /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
 673        if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
 674                return 0;
 675        if (vma->vm_file != file)
 676                return 0;
 677        if (vma->vm_ops && vma->vm_ops->close)
 678                return 0;
 679        return 1;
 680}
 681
 682static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
 683                                        struct anon_vma *anon_vma2,
 684                                        struct vm_area_struct *vma)
 685{
 686        /*
 687         * The list_is_singular() test is to avoid merging VMA cloned from
 688         * parents. This can improve scalability caused by anon_vma lock.
 689         */
 690        if ((!anon_vma1 || !anon_vma2) && (!vma ||
 691                list_is_singular(&vma->anon_vma_chain)))
 692                return 1;
 693        return anon_vma1 == anon_vma2;
 694}
 695
 696/*
 697 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 698 * in front of (at a lower virtual address and file offset than) the vma.
 699 *
 700 * We cannot merge two vmas if they have differently assigned (non-NULL)
 701 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 702 *
 703 * We don't check here for the merged mmap wrapping around the end of pagecache
 704 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
 705 * wrap, nor mmaps which cover the final page at index -1UL.
 706 */
 707static int
 708can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
 709        struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 710{
 711        if (is_mergeable_vma(vma, file, vm_flags) &&
 712            is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 713                if (vma->vm_pgoff == vm_pgoff)
 714                        return 1;
 715        }
 716        return 0;
 717}
 718
 719/*
 720 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 721 * beyond (at a higher virtual address and file offset than) the vma.
 722 *
 723 * We cannot merge two vmas if they have differently assigned (non-NULL)
 724 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 725 */
 726static int
 727can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 728        struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 729{
 730        if (is_mergeable_vma(vma, file, vm_flags) &&
 731            is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 732                pgoff_t vm_pglen;
 733                vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 734                if (vma->vm_pgoff + vm_pglen == vm_pgoff)
 735                        return 1;
 736        }
 737        return 0;
 738}
 739
 740/*
 741 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
 742 * whether that can be merged with its predecessor or its successor.
 743 * Or both (it neatly fills a hole).
 744 *
 745 * In most cases - when called for mmap, brk or mremap - [addr,end) is
 746 * certain not to be mapped by the time vma_merge is called; but when
 747 * called for mprotect, it is certain to be already mapped (either at
 748 * an offset within prev, or at the start of next), and the flags of
 749 * this area are about to be changed to vm_flags - and the no-change
 750 * case has already been eliminated.
 751 *
 752 * The following mprotect cases have to be considered, where AAAA is
 753 * the area passed down from mprotect_fixup, never extending beyond one
 754 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
 755 *
 756 *     AAAA             AAAA                AAAA          AAAA
 757 *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
 758 *    cannot merge    might become    might become    might become
 759 *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
 760 *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
 761 *    mremap move:                                    PPPPNNNNNNNN 8
 762 *        AAAA
 763 *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
 764 *    might become    case 1 below    case 2 below    case 3 below
 765 *
 766 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
 767 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
 768 */
 769struct vm_area_struct *vma_merge(struct mm_struct *mm,
 770                        struct vm_area_struct *prev, unsigned long addr,
 771                        unsigned long end, unsigned long vm_flags,
 772                        struct anon_vma *anon_vma, struct file *file,
 773                        pgoff_t pgoff, struct mempolicy *policy)
 774{
 775        pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
 776        struct vm_area_struct *area, *next;
 777        int err;
 778
 779        /*
 780         * We later require that vma->vm_flags == vm_flags,
 781         * so this tests vma->vm_flags & VM_SPECIAL, too.
 782         */
 783        if (vm_flags & VM_SPECIAL)
 784                return NULL;
 785
 786        if (prev)
 787                next = prev->vm_next;
 788        else
 789                next = mm->mmap;
 790        area = next;
 791        if (next && next->vm_end == end)                /* cases 6, 7, 8 */
 792                next = next->vm_next;
 793
 794        /*
 795         * Can it merge with the predecessor?
 796         */
 797        if (prev && prev->vm_end == addr &&
 798                        mpol_equal(vma_policy(prev), policy) &&
 799                        can_vma_merge_after(prev, vm_flags,
 800                                                anon_vma, file, pgoff)) {
 801                /*
 802                 * OK, it can.  Can we now merge in the successor as well?
 803                 */
 804                if (next && end == next->vm_start &&
 805                                mpol_equal(policy, vma_policy(next)) &&
 806                                can_vma_merge_before(next, vm_flags,
 807                                        anon_vma, file, pgoff+pglen) &&
 808                                is_mergeable_anon_vma(prev->anon_vma,
 809                                                      next->anon_vma, NULL)) {
 810                                                        /* cases 1, 6 */
 811                        err = vma_adjust(prev, prev->vm_start,
 812                                next->vm_end, prev->vm_pgoff, NULL);
 813                } else                                  /* cases 2, 5, 7 */
 814                        err = vma_adjust(prev, prev->vm_start,
 815                                end, prev->vm_pgoff, NULL);
 816                if (err)
 817                        return NULL;
 818                khugepaged_enter_vma_merge(prev);
 819                return prev;
 820        }
 821
 822        /*
 823         * Can this new request be merged in front of next?
 824         */
 825        if (next && end == next->vm_start &&
 826                        mpol_equal(policy, vma_policy(next)) &&
 827                        can_vma_merge_before(next, vm_flags,
 828                                        anon_vma, file, pgoff+pglen)) {
 829                if (prev && addr < prev->vm_end)        /* case 4 */
 830                        err = vma_adjust(prev, prev->vm_start,
 831                                addr, prev->vm_pgoff, NULL);
 832                else                                    /* cases 3, 8 */
 833                        err = vma_adjust(area, addr, next->vm_end,
 834                                next->vm_pgoff - pglen, NULL);
 835                if (err)
 836                        return NULL;
 837                khugepaged_enter_vma_merge(area);
 838                return area;
 839        }
 840
 841        return NULL;
 842}
 843
 844/*
 845 * Rough compatbility check to quickly see if it's even worth looking
 846 * at sharing an anon_vma.
 847 *
 848 * They need to have the same vm_file, and the flags can only differ
 849 * in things that mprotect may change.
 850 *
 851 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
 852 * we can merge the two vma's. For example, we refuse to merge a vma if
 853 * there is a vm_ops->close() function, because that indicates that the
 854 * driver is doing some kind of reference counting. But that doesn't
 855 * really matter for the anon_vma sharing case.
 856 */
 857static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
 858{
 859        return a->vm_end == b->vm_start &&
 860                mpol_equal(vma_policy(a), vma_policy(b)) &&
 861                a->vm_file == b->vm_file &&
 862                !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
 863                b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
 864}
 865
 866/*
 867 * Do some basic sanity checking to see if we can re-use the anon_vma
 868 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
 869 * the same as 'old', the other will be the new one that is trying
 870 * to share the anon_vma.
 871 *
 872 * NOTE! This runs with mm_sem held for reading, so it is possible that
 873 * the anon_vma of 'old' is concurrently in the process of being set up
 874 * by another page fault trying to merge _that_. But that's ok: if it
 875 * is being set up, that automatically means that it will be a singleton
 876 * acceptable for merging, so we can do all of this optimistically. But
 877 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
 878 *
 879 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
 880 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
 881 * is to return an anon_vma that is "complex" due to having gone through
 882 * a fork).
 883 *
 884 * We also make sure that the two vma's are compatible (adjacent,
 885 * and with the same memory policies). That's all stable, even with just
 886 * a read lock on the mm_sem.
 887 */
 888static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
 889{
 890        if (anon_vma_compatible(a, b)) {
 891                struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
 892
 893                if (anon_vma && list_is_singular(&old->anon_vma_chain))
 894                        return anon_vma;
 895        }
 896        return NULL;
 897}
 898
 899/*
 900 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
 901 * neighbouring vmas for a suitable anon_vma, before it goes off
 902 * to allocate a new anon_vma.  It checks because a repetitive
 903 * sequence of mprotects and faults may otherwise lead to distinct
 904 * anon_vmas being allocated, preventing vma merge in subsequent
 905 * mprotect.
 906 */
 907struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
 908{
 909        struct anon_vma *anon_vma;
 910        struct vm_area_struct *near;
 911
 912        near = vma->vm_next;
 913        if (!near)
 914                goto try_prev;
 915
 916        anon_vma = reusable_anon_vma(near, vma, near);
 917        if (anon_vma)
 918                return anon_vma;
 919try_prev:
 920        near = vma->vm_prev;
 921        if (!near)
 922                goto none;
 923
 924        anon_vma = reusable_anon_vma(near, near, vma);
 925        if (anon_vma)
 926                return anon_vma;
 927none:
 928        /*
 929         * There's no absolute need to look only at touching neighbours:
 930         * we could search further afield for "compatible" anon_vmas.
 931         * But it would probably just be a waste of time searching,
 932         * or lead to too many vmas hanging off the same anon_vma.
 933         * We're trying to allow mprotect remerging later on,
 934         * not trying to minimize memory used for anon_vmas.
 935         */
 936        return NULL;
 937}
 938
 939#ifdef CONFIG_PROC_FS
 940void vm_stat_account(struct mm_struct *mm, unsigned long flags,
 941                                                struct file *file, long pages)
 942{
 943        const unsigned long stack_flags
 944                = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
 945
 946        mm->total_vm += pages;
 947
 948        if (file) {
 949                mm->shared_vm += pages;
 950                if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
 951                        mm->exec_vm += pages;
 952        } else if (flags & stack_flags)
 953                mm->stack_vm += pages;
 954        if (flags & (VM_RESERVED|VM_IO))
 955                mm->reserved_vm += pages;
 956}
 957#endif /* CONFIG_PROC_FS */
 958
 959/*
 960 * If a hint addr is less than mmap_min_addr change hint to be as
 961 * low as possible but still greater than mmap_min_addr
 962 */
 963static inline unsigned long round_hint_to_min(unsigned long hint)
 964{
 965        hint &= PAGE_MASK;
 966        if (((void *)hint != NULL) &&
 967            (hint < mmap_min_addr))
 968                return PAGE_ALIGN(mmap_min_addr);
 969        return hint;
 970}
 971
 972/*
 973 * The caller must hold down_write(&current->mm->mmap_sem).
 974 */
 975
 976unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 977                        unsigned long len, unsigned long prot,
 978                        unsigned long flags, unsigned long pgoff)
 979{
 980        struct mm_struct * mm = current->mm;
 981        struct inode *inode;
 982        vm_flags_t vm_flags;
 983
 984        /*
 985         * Does the application expect PROT_READ to imply PROT_EXEC?
 986         *
 987         * (the exception is when the underlying filesystem is noexec
 988         *  mounted, in which case we dont add PROT_EXEC.)
 989         */
 990        if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 991                if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
 992                        prot |= PROT_EXEC;
 993
 994        if (!len)
 995                return -EINVAL;
 996
 997        if (!(flags & MAP_FIXED))
 998                addr = round_hint_to_min(addr);
 999
1000        /* Careful about overflows.. */
1001        len = PAGE_ALIGN(len);
1002        if (!len)
1003                return -ENOMEM;
1004
1005        /* offset overflow? */
1006        if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1007               return -EOVERFLOW;
1008
1009        /* Too many mappings? */
1010        if (mm->map_count > sysctl_max_map_count)
1011                return -ENOMEM;
1012
1013        /* Obtain the address to map to. we verify (or select) it and ensure
1014         * that it represents a valid section of the address space.
1015         */
1016        addr = get_unmapped_area(file, addr, len, pgoff, flags);
1017        if (addr & ~PAGE_MASK)
1018                return addr;
1019
1020        /* Do simple checking here so the lower-level routines won't have
1021         * to. we assume access permissions have been handled by the open
1022         * of the memory object, so we don't do any here.
1023         */
1024        vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1025                        mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1026
1027        if (flags & MAP_LOCKED)
1028                if (!can_do_mlock())
1029                        return -EPERM;
1030
1031        /* mlock MCL_FUTURE? */
1032        if (vm_flags & VM_LOCKED) {
1033                unsigned long locked, lock_limit;
1034                locked = len >> PAGE_SHIFT;
1035                locked += mm->locked_vm;
1036                lock_limit = rlimit(RLIMIT_MEMLOCK);
1037                lock_limit >>= PAGE_SHIFT;
1038                if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1039                        return -EAGAIN;
1040        }
1041
1042        inode = file ? file->f_path.dentry->d_inode : NULL;
1043
1044        if (file) {
1045                switch (flags & MAP_TYPE) {
1046                case MAP_SHARED:
1047                        if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1048                                return -EACCES;
1049
1050                        /*
1051                         * Make sure we don't allow writing to an append-only
1052                         * file..
1053                         */
1054                        if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1055                                return -EACCES;
1056
1057                        /*
1058                         * Make sure there are no mandatory locks on the file.
1059                         */
1060                        if (locks_verify_locked(inode))
1061                                return -EAGAIN;
1062
1063                        vm_flags |= VM_SHARED | VM_MAYSHARE;
1064                        if (!(file->f_mode & FMODE_WRITE))
1065                                vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1066
1067                        /* fall through */
1068                case MAP_PRIVATE:
1069                        if (!(file->f_mode & FMODE_READ))
1070                                return -EACCES;
1071                        if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1072                                if (vm_flags & VM_EXEC)
1073                                        return -EPERM;
1074                                vm_flags &= ~VM_MAYEXEC;
1075                        }
1076
1077                        if (!file->f_op || !file->f_op->mmap)
1078                                return -ENODEV;
1079                        break;
1080
1081                default:
1082                        return -EINVAL;
1083                }
1084        } else {
1085                switch (flags & MAP_TYPE) {
1086                case MAP_SHARED:
1087                        /*
1088                         * Ignore pgoff.
1089                         */
1090                        pgoff = 0;
1091                        vm_flags |= VM_SHARED | VM_MAYSHARE;
1092                        break;
1093                case MAP_PRIVATE:
1094                        /*
1095                         * Set pgoff according to addr for anon_vma.
1096                         */
1097                        pgoff = addr >> PAGE_SHIFT;
1098                        break;
1099                default:
1100                        return -EINVAL;
1101                }
1102        }
1103
1104        return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1105}
1106
1107SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1108                unsigned long, prot, unsigned long, flags,
1109                unsigned long, fd, unsigned long, pgoff)
1110{
1111        struct file *file = NULL;
1112        unsigned long retval = -EBADF;
1113
1114        if (!(flags & MAP_ANONYMOUS)) {
1115                audit_mmap_fd(fd, flags);
1116                if (unlikely(flags & MAP_HUGETLB))
1117                        return -EINVAL;
1118                file = fget(fd);
1119                if (!file)
1120                        goto out;
1121        } else if (flags & MAP_HUGETLB) {
1122                struct user_struct *user = NULL;
1123                /*
1124                 * VM_NORESERVE is used because the reservations will be
1125                 * taken when vm_ops->mmap() is called
1126                 * A dummy user value is used because we are not locking
1127                 * memory so no accounting is necessary
1128                 */
1129                file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1130                                                VM_NORESERVE, &user,
1131                                                HUGETLB_ANONHUGE_INODE);
1132                if (IS_ERR(file))
1133                        return PTR_ERR(file);
1134        }
1135
1136        flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1137
1138        retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1139        if (file)
1140                fput(file);
1141out:
1142        return retval;
1143}
1144
1145#ifdef __ARCH_WANT_SYS_OLD_MMAP
1146struct mmap_arg_struct {
1147        unsigned long addr;
1148        unsigned long len;
1149        unsigned long prot;
1150        unsigned long flags;
1151        unsigned long fd;
1152        unsigned long offset;
1153};
1154
1155SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1156{
1157        struct mmap_arg_struct a;
1158
1159        if (copy_from_user(&a, arg, sizeof(a)))
1160                return -EFAULT;
1161        if (a.offset & ~PAGE_MASK)
1162                return -EINVAL;
1163
1164        return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1165                              a.offset >> PAGE_SHIFT);
1166}
1167#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1168
1169/*
1170 * Some shared mappigns will want the pages marked read-only
1171 * to track write events. If so, we'll downgrade vm_page_prot
1172 * to the private version (using protection_map[] without the
1173 * VM_SHARED bit).
1174 */
1175int vma_wants_writenotify(struct vm_area_struct *vma)
1176{
1177        vm_flags_t vm_flags = vma->vm_flags;
1178
1179        /* If it was private or non-writable, the write bit is already clear */
1180        if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1181                return 0;
1182
1183        /* The backer wishes to know when pages are first written to? */
1184        if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1185                return 1;
1186
1187        /* The open routine did something to the protections already? */
1188        if (pgprot_val(vma->vm_page_prot) !=
1189            pgprot_val(vm_get_page_prot(vm_flags)))
1190                return 0;
1191
1192        /* Specialty mapping? */
1193        if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1194                return 0;
1195
1196        /* Can the mapping track the dirty pages? */
1197        return vma->vm_file && vma->vm_file->f_mapping &&
1198                mapping_cap_account_dirty(vma->vm_file->f_mapping);
1199}
1200
1201/*
1202 * We account for memory if it's a private writeable mapping,
1203 * not hugepages and VM_NORESERVE wasn't set.
1204 */
1205static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1206{
1207        /*
1208         * hugetlb has its own accounting separate from the core VM
1209         * VM_HUGETLB may not be set yet so we cannot check for that flag.
1210         */
1211        if (file && is_file_hugepages(file))
1212                return 0;
1213
1214        return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1215}
1216
1217unsigned long mmap_region(struct file *file, unsigned long addr,
1218                          unsigned long len, unsigned long flags,
1219                          vm_flags_t vm_flags, unsigned long pgoff)
1220{
1221        struct mm_struct *mm = current->mm;
1222        struct vm_area_struct *vma, *prev;
1223        int correct_wcount = 0;
1224        int error;
1225        struct rb_node **rb_link, *rb_parent;
1226        unsigned long charged = 0;
1227        struct inode *inode =  file ? file->f_path.dentry->d_inode : NULL;
1228
1229        /* Clear old maps */
1230        error = -ENOMEM;
1231munmap_back:
1232        vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1233        if (vma && vma->vm_start < addr + len) {
1234                if (do_munmap(mm, addr, len))
1235                        return -ENOMEM;
1236                goto munmap_back;
1237        }
1238
1239        /* Check against address space limit. */
1240        if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1241                return -ENOMEM;
1242
1243        /*
1244         * Set 'VM_NORESERVE' if we should not account for the
1245         * memory use of this mapping.
1246         */
1247        if ((flags & MAP_NORESERVE)) {
1248                /* We honor MAP_NORESERVE if allowed to overcommit */
1249                if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1250                        vm_flags |= VM_NORESERVE;
1251
1252                /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1253                if (file && is_file_hugepages(file))
1254                        vm_flags |= VM_NORESERVE;
1255        }
1256
1257        /*
1258         * Private writable mapping: check memory availability
1259         */
1260        if (accountable_mapping(file, vm_flags)) {
1261                charged = len >> PAGE_SHIFT;
1262                if (security_vm_enough_memory_mm(mm, charged))
1263                        return -ENOMEM;
1264                vm_flags |= VM_ACCOUNT;
1265        }
1266
1267        /*
1268         * Can we just expand an old mapping?
1269         */
1270        vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1271        if (vma)
1272                goto out;
1273
1274        /*
1275         * Determine the object being mapped and call the appropriate
1276         * specific mapper. the address has already been validated, but
1277         * not unmapped, but the maps are removed from the list.
1278         */
1279        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1280        if (!vma) {
1281                error = -ENOMEM;
1282                goto unacct_error;
1283        }
1284
1285        vma->vm_mm = mm;
1286        vma->vm_start = addr;
1287        vma->vm_end = addr + len;
1288        vma->vm_flags = vm_flags;
1289        vma->vm_page_prot = vm_get_page_prot(vm_flags);
1290        vma->vm_pgoff = pgoff;
1291        INIT_LIST_HEAD(&vma->anon_vma_chain);
1292
1293        error = -EINVAL;        /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1294
1295        if (file) {
1296                if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1297                        goto free_vma;
1298                if (vm_flags & VM_DENYWRITE) {
1299                        error = deny_write_access(file);
1300                        if (error)
1301                                goto free_vma;
1302                        correct_wcount = 1;
1303                }
1304                vma->vm_file = file;
1305                get_file(file);
1306                error = file->f_op->mmap(file, vma);
1307                if (error)
1308                        goto unmap_and_free_vma;
1309                if (vm_flags & VM_EXECUTABLE)
1310                        added_exe_file_vma(mm);
1311
1312                /* Can addr have changed??
1313                 *
1314                 * Answer: Yes, several device drivers can do it in their
1315                 *         f_op->mmap method. -DaveM
1316                 */
1317                addr = vma->vm_start;
1318                pgoff = vma->vm_pgoff;
1319                vm_flags = vma->vm_flags;
1320        } else if (vm_flags & VM_SHARED) {
1321                if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1322                        goto free_vma;
1323                error = shmem_zero_setup(vma);
1324                if (error)
1325                        goto free_vma;
1326        }
1327
1328        if (vma_wants_writenotify(vma)) {
1329                pgprot_t pprot = vma->vm_page_prot;
1330
1331                /* Can vma->vm_page_prot have changed??
1332                 *
1333                 * Answer: Yes, drivers may have changed it in their
1334                 *         f_op->mmap method.
1335                 *
1336                 * Ensures that vmas marked as uncached stay that way.
1337                 */
1338                vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1339                if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1340                        vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1341        }
1342
1343        vma_link(mm, vma, prev, rb_link, rb_parent);
1344        file = vma->vm_file;
1345
1346        /* Once vma denies write, undo our temporary denial count */
1347        if (correct_wcount)
1348                atomic_inc(&inode->i_writecount);
1349out:
1350        perf_event_mmap(vma);
1351
1352        vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1353        if (vm_flags & VM_LOCKED) {
1354                if (!mlock_vma_pages_range(vma, addr, addr + len))
1355                        mm->locked_vm += (len >> PAGE_SHIFT);
1356        } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1357                make_pages_present(addr, addr + len);
1358
1359        if (file)
1360                uprobe_mmap(vma);
1361
1362        return addr;
1363
1364unmap_and_free_vma:
1365        if (correct_wcount)
1366                atomic_inc(&inode->i_writecount);
1367        vma->vm_file = NULL;
1368        fput(file);
1369
1370        /* Undo any partial mapping done by a device driver. */
1371        unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1372        charged = 0;
1373free_vma:
1374        kmem_cache_free(vm_area_cachep, vma);
1375unacct_error:
1376        if (charged)
1377                vm_unacct_memory(charged);
1378        return error;
1379}
1380
1381/* Get an address range which is currently unmapped.
1382 * For shmat() with addr=0.
1383 *
1384 * Ugly calling convention alert:
1385 * Return value with the low bits set means error value,
1386 * ie
1387 *      if (ret & ~PAGE_MASK)
1388 *              error = ret;
1389 *
1390 * This function "knows" that -ENOMEM has the bits set.
1391 */
1392#ifndef HAVE_ARCH_UNMAPPED_AREA
1393unsigned long
1394arch_get_unmapped_area(struct file *filp, unsigned long addr,
1395                unsigned long len, unsigned long pgoff, unsigned long flags)
1396{
1397        struct mm_struct *mm = current->mm;
1398        struct vm_area_struct *vma;
1399        unsigned long start_addr;
1400
1401        if (len > TASK_SIZE)
1402                return -ENOMEM;
1403
1404        if (flags & MAP_FIXED)
1405                return addr;
1406
1407        if (addr) {
1408                addr = PAGE_ALIGN(addr);
1409                vma = find_vma(mm, addr);
1410                if (TASK_SIZE - len >= addr &&
1411                    (!vma || addr + len <= vma->vm_start))
1412                        return addr;
1413        }
1414        if (len > mm->cached_hole_size) {
1415                start_addr = addr = mm->free_area_cache;
1416        } else {
1417                start_addr = addr = TASK_UNMAPPED_BASE;
1418                mm->cached_hole_size = 0;
1419        }
1420
1421full_search:
1422        for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1423                /* At this point:  (!vma || addr < vma->vm_end). */
1424                if (TASK_SIZE - len < addr) {
1425                        /*
1426                         * Start a new search - just in case we missed
1427                         * some holes.
1428                         */
1429                        if (start_addr != TASK_UNMAPPED_BASE) {
1430                                addr = TASK_UNMAPPED_BASE;
1431                                start_addr = addr;
1432                                mm->cached_hole_size = 0;
1433                                goto full_search;
1434                        }
1435                        return -ENOMEM;
1436                }
1437                if (!vma || addr + len <= vma->vm_start) {
1438                        /*
1439                         * Remember the place where we stopped the search:
1440                         */
1441                        mm->free_area_cache = addr + len;
1442                        return addr;
1443                }
1444                if (addr + mm->cached_hole_size < vma->vm_start)
1445                        mm->cached_hole_size = vma->vm_start - addr;
1446                addr = vma->vm_end;
1447        }
1448}
1449#endif  
1450
1451void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1452{
1453        /*
1454         * Is this a new hole at the lowest possible address?
1455         */
1456        if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1457                mm->free_area_cache = addr;
1458}
1459
1460/*
1461 * This mmap-allocator allocates new areas top-down from below the
1462 * stack's low limit (the base):
1463 */
1464#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1465unsigned long
1466arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1467                          const unsigned long len, const unsigned long pgoff,
1468                          const unsigned long flags)
1469{
1470        struct vm_area_struct *vma;
1471        struct mm_struct *mm = current->mm;
1472        unsigned long addr = addr0, start_addr;
1473
1474        /* requested length too big for entire address space */
1475        if (len > TASK_SIZE)
1476                return -ENOMEM;
1477
1478        if (flags & MAP_FIXED)
1479                return addr;
1480
1481        /* requesting a specific address */
1482        if (addr) {
1483                addr = PAGE_ALIGN(addr);
1484                vma = find_vma(mm, addr);
1485                if (TASK_SIZE - len >= addr &&
1486                                (!vma || addr + len <= vma->vm_start))
1487                        return addr;
1488        }
1489
1490        /* check if free_area_cache is useful for us */
1491        if (len <= mm->cached_hole_size) {
1492                mm->cached_hole_size = 0;
1493                mm->free_area_cache = mm->mmap_base;
1494        }
1495
1496try_again:
1497        /* either no address requested or can't fit in requested address hole */
1498        start_addr = addr = mm->free_area_cache;
1499
1500        if (addr < len)
1501                goto fail;
1502
1503        addr -= len;
1504        do {
1505                /*
1506                 * Lookup failure means no vma is above this address,
1507                 * else if new region fits below vma->vm_start,
1508                 * return with success:
1509                 */
1510                vma = find_vma(mm, addr);
1511                if (!vma || addr+len <= vma->vm_start)
1512                        /* remember the address as a hint for next time */
1513                        return (mm->free_area_cache = addr);
1514
1515                /* remember the largest hole we saw so far */
1516                if (addr + mm->cached_hole_size < vma->vm_start)
1517                        mm->cached_hole_size = vma->vm_start - addr;
1518
1519                /* try just below the current vma->vm_start */
1520                addr = vma->vm_start-len;
1521        } while (len < vma->vm_start);
1522
1523fail:
1524        /*
1525         * if hint left us with no space for the requested
1526         * mapping then try again:
1527         *
1528         * Note: this is different with the case of bottomup
1529         * which does the fully line-search, but we use find_vma
1530         * here that causes some holes skipped.
1531         */
1532        if (start_addr != mm->mmap_base) {
1533                mm->free_area_cache = mm->mmap_base;
1534                mm->cached_hole_size = 0;
1535                goto try_again;
1536        }
1537
1538        /*
1539         * A failed mmap() very likely causes application failure,
1540         * so fall back to the bottom-up function here. This scenario
1541         * can happen with large stack limits and large mmap()
1542         * allocations.
1543         */
1544        mm->cached_hole_size = ~0UL;
1545        mm->free_area_cache = TASK_UNMAPPED_BASE;
1546        addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1547        /*
1548         * Restore the topdown base:
1549         */
1550        mm->free_area_cache = mm->mmap_base;
1551        mm->cached_hole_size = ~0UL;
1552
1553        return addr;
1554}
1555#endif
1556
1557void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1558{
1559        /*
1560         * Is this a new hole at the highest possible address?
1561         */
1562        if (addr > mm->free_area_cache)
1563                mm->free_area_cache = addr;
1564
1565        /* dont allow allocations above current base */
1566        if (mm->free_area_cache > mm->mmap_base)
1567                mm->free_area_cache = mm->mmap_base;
1568}
1569
1570unsigned long
1571get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1572                unsigned long pgoff, unsigned long flags)
1573{
1574        unsigned long (*get_area)(struct file *, unsigned long,
1575                                  unsigned long, unsigned long, unsigned long);
1576
1577        unsigned long error = arch_mmap_check(addr, len, flags);
1578        if (error)
1579                return error;
1580
1581        /* Careful about overflows.. */
1582        if (len > TASK_SIZE)
1583                return -ENOMEM;
1584
1585        get_area = current->mm->get_unmapped_area;
1586        if (file && file->f_op && file->f_op->get_unmapped_area)
1587                get_area = file->f_op->get_unmapped_area;
1588        addr = get_area(file, addr, len, pgoff, flags);
1589        if (IS_ERR_VALUE(addr))
1590                return addr;
1591
1592        if (addr > TASK_SIZE - len)
1593                return -ENOMEM;
1594        if (addr & ~PAGE_MASK)
1595                return -EINVAL;
1596
1597        addr = arch_rebalance_pgtables(addr, len);
1598        error = security_mmap_addr(addr);
1599        return error ? error : addr;
1600}
1601
1602EXPORT_SYMBOL(get_unmapped_area);
1603
1604/* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
1605struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1606{
1607        struct vm_area_struct *vma = NULL;
1608
1609        if (WARN_ON_ONCE(!mm))          /* Remove this in linux-3.6 */
1610                return NULL;
1611
1612        /* Check the cache first. */
1613        /* (Cache hit rate is typically around 35%.) */
1614        vma = mm->mmap_cache;
1615        if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1616                struct rb_node *rb_node;
1617
1618                rb_node = mm->mm_rb.rb_node;
1619                vma = NULL;
1620
1621                while (rb_node) {
1622                        struct vm_area_struct *vma_tmp;
1623
1624                        vma_tmp = rb_entry(rb_node,
1625                                           struct vm_area_struct, vm_rb);
1626
1627                        if (vma_tmp->vm_end > addr) {
1628                                vma = vma_tmp;
1629                                if (vma_tmp->vm_start <= addr)
1630                                        break;
1631                                rb_node = rb_node->rb_left;
1632                        } else
1633                                rb_node = rb_node->rb_right;
1634                }
1635                if (vma)
1636                        mm->mmap_cache = vma;
1637        }
1638        return vma;
1639}
1640
1641EXPORT_SYMBOL(find_vma);
1642
1643/*
1644 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1645 */
1646struct vm_area_struct *
1647find_vma_prev(struct mm_struct *mm, unsigned long addr,
1648                        struct vm_area_struct **pprev)
1649{
1650        struct vm_area_struct *vma;
1651
1652        vma = find_vma(mm, addr);
1653        if (vma) {
1654                *pprev = vma->vm_prev;
1655        } else {
1656                struct rb_node *rb_node = mm->mm_rb.rb_node;
1657                *pprev = NULL;
1658                while (rb_node) {
1659                        *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1660                        rb_node = rb_node->rb_right;
1661                }
1662        }
1663        return vma;
1664}
1665
1666/*
1667 * Verify that the stack growth is acceptable and
1668 * update accounting. This is shared with both the
1669 * grow-up and grow-down cases.
1670 */
1671static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1672{
1673        struct mm_struct *mm = vma->vm_mm;
1674        struct rlimit *rlim = current->signal->rlim;
1675        unsigned long new_start;
1676
1677        /* address space limit tests */
1678        if (!may_expand_vm(mm, grow))
1679                return -ENOMEM;
1680
1681        /* Stack limit test */
1682        if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1683                return -ENOMEM;
1684
1685        /* mlock limit tests */
1686        if (vma->vm_flags & VM_LOCKED) {
1687                unsigned long locked;
1688                unsigned long limit;
1689                locked = mm->locked_vm + grow;
1690                limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1691                limit >>= PAGE_SHIFT;
1692                if (locked > limit && !capable(CAP_IPC_LOCK))
1693                        return -ENOMEM;
1694        }
1695
1696        /* Check to ensure the stack will not grow into a hugetlb-only region */
1697        new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1698                        vma->vm_end - size;
1699        if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1700                return -EFAULT;
1701
1702        /*
1703         * Overcommit..  This must be the final test, as it will
1704         * update security statistics.
1705         */
1706        if (security_vm_enough_memory_mm(mm, grow))
1707                return -ENOMEM;
1708
1709        /* Ok, everything looks good - let it rip */
1710        if (vma->vm_flags & VM_LOCKED)
1711                mm->locked_vm += grow;
1712        vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1713        return 0;
1714}
1715
1716#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1717/*
1718 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1719 * vma is the last one with address > vma->vm_end.  Have to extend vma.
1720 */
1721int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1722{
1723        int error;
1724
1725        if (!(vma->vm_flags & VM_GROWSUP))
1726                return -EFAULT;
1727
1728        /*
1729         * We must make sure the anon_vma is allocated
1730         * so that the anon_vma locking is not a noop.
1731         */
1732        if (unlikely(anon_vma_prepare(vma)))
1733                return -ENOMEM;
1734        vma_lock_anon_vma(vma);
1735
1736        /*
1737         * vma->vm_start/vm_end cannot change under us because the caller
1738         * is required to hold the mmap_sem in read mode.  We need the
1739         * anon_vma lock to serialize against concurrent expand_stacks.
1740         * Also guard against wrapping around to address 0.
1741         */
1742        if (address < PAGE_ALIGN(address+4))
1743                address = PAGE_ALIGN(address+4);
1744        else {
1745                vma_unlock_anon_vma(vma);
1746                return -ENOMEM;
1747        }
1748        error = 0;
1749
1750        /* Somebody else might have raced and expanded it already */
1751        if (address > vma->vm_end) {
1752                unsigned long size, grow;
1753
1754                size = address - vma->vm_start;
1755                grow = (address - vma->vm_end) >> PAGE_SHIFT;
1756
1757                error = -ENOMEM;
1758                if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1759                        error = acct_stack_growth(vma, size, grow);
1760                        if (!error) {
1761                                vma->vm_end = address;
1762                                perf_event_mmap(vma);
1763                        }
1764                }
1765        }
1766        vma_unlock_anon_vma(vma);
1767        khugepaged_enter_vma_merge(vma);
1768        return error;
1769}
1770#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1771
1772/*
1773 * vma is the first one with address < vma->vm_start.  Have to extend vma.
1774 */
1775int expand_downwards(struct vm_area_struct *vma,
1776                                   unsigned long address)
1777{
1778        int error;
1779
1780        /*
1781         * We must make sure the anon_vma is allocated
1782         * so that the anon_vma locking is not a noop.
1783         */
1784        if (unlikely(anon_vma_prepare(vma)))
1785                return -ENOMEM;
1786
1787        address &= PAGE_MASK;
1788        error = security_mmap_addr(address);
1789        if (error)
1790                return error;
1791
1792        vma_lock_anon_vma(vma);
1793
1794        /*
1795         * vma->vm_start/vm_end cannot change under us because the caller
1796         * is required to hold the mmap_sem in read mode.  We need the
1797         * anon_vma lock to serialize against concurrent expand_stacks.
1798         */
1799
1800        /* Somebody else might have raced and expanded it already */
1801        if (address < vma->vm_start) {
1802                unsigned long size, grow;
1803
1804                size = vma->vm_end - address;
1805                grow = (vma->vm_start - address) >> PAGE_SHIFT;
1806
1807                error = -ENOMEM;
1808                if (grow <= vma->vm_pgoff) {
1809                        error = acct_stack_growth(vma, size, grow);
1810                        if (!error) {
1811                                vma->vm_start = address;
1812                                vma->vm_pgoff -= grow;
1813                                perf_event_mmap(vma);
1814                        }
1815                }
1816        }
1817        vma_unlock_anon_vma(vma);
1818        khugepaged_enter_vma_merge(vma);
1819        return error;
1820}
1821
1822#ifdef CONFIG_STACK_GROWSUP
1823int expand_stack(struct vm_area_struct *vma, unsigned long address)
1824{
1825        return expand_upwards(vma, address);
1826}
1827
1828struct vm_area_struct *
1829find_extend_vma(struct mm_struct *mm, unsigned long addr)
1830{
1831        struct vm_area_struct *vma, *prev;
1832
1833        addr &= PAGE_MASK;
1834        vma = find_vma_prev(mm, addr, &prev);
1835        if (vma && (vma->vm_start <= addr))
1836                return vma;
1837        if (!prev || expand_stack(prev, addr))
1838                return NULL;
1839        if (prev->vm_flags & VM_LOCKED) {
1840                mlock_vma_pages_range(prev, addr, prev->vm_end);
1841        }
1842        return prev;
1843}
1844#else
1845int expand_stack(struct vm_area_struct *vma, unsigned long address)
1846{
1847        return expand_downwards(vma, address);
1848}
1849
1850struct vm_area_struct *
1851find_extend_vma(struct mm_struct * mm, unsigned long addr)
1852{
1853        struct vm_area_struct * vma;
1854        unsigned long start;
1855
1856        addr &= PAGE_MASK;
1857        vma = find_vma(mm,addr);
1858        if (!vma)
1859                return NULL;
1860        if (vma->vm_start <= addr)
1861                return vma;
1862        if (!(vma->vm_flags & VM_GROWSDOWN))
1863                return NULL;
1864        start = vma->vm_start;
1865        if (expand_stack(vma, addr))
1866                return NULL;
1867        if (vma->vm_flags & VM_LOCKED) {
1868                mlock_vma_pages_range(vma, addr, start);
1869        }
1870        return vma;
1871}
1872#endif
1873
1874/*
1875 * Ok - we have the memory areas we should free on the vma list,
1876 * so release them, and do the vma updates.
1877 *
1878 * Called with the mm semaphore held.
1879 */
1880static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1881{
1882        unsigned long nr_accounted = 0;
1883
1884        /* Update high watermark before we lower total_vm */
1885        update_hiwater_vm(mm);
1886        do {
1887                long nrpages = vma_pages(vma);
1888
1889                if (vma->vm_flags & VM_ACCOUNT)
1890                        nr_accounted += nrpages;
1891                vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1892                vma = remove_vma(vma);
1893        } while (vma);
1894        vm_unacct_memory(nr_accounted);
1895        validate_mm(mm);
1896}
1897
1898/*
1899 * Get rid of page table information in the indicated region.
1900 *
1901 * Called with the mm semaphore held.
1902 */
1903static void unmap_region(struct mm_struct *mm,
1904                struct vm_area_struct *vma, struct vm_area_struct *prev,
1905                unsigned long start, unsigned long end)
1906{
1907        struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1908        struct mmu_gather tlb;
1909
1910        lru_add_drain();
1911        tlb_gather_mmu(&tlb, mm, 0);
1912        update_hiwater_rss(mm);
1913        unmap_vmas(&tlb, vma, start, end);
1914        free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1915                                 next ? next->vm_start : 0);
1916        tlb_finish_mmu(&tlb, start, end);
1917}
1918
1919/*
1920 * Create a list of vma's touched by the unmap, removing them from the mm's
1921 * vma list as we go..
1922 */
1923static void
1924detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1925        struct vm_area_struct *prev, unsigned long end)
1926{
1927        struct vm_area_struct **insertion_point;
1928        struct vm_area_struct *tail_vma = NULL;
1929        unsigned long addr;
1930
1931        insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1932        vma->vm_prev = NULL;
1933        do {
1934                rb_erase(&vma->vm_rb, &mm->mm_rb);
1935                mm->map_count--;
1936                tail_vma = vma;
1937                vma = vma->vm_next;
1938        } while (vma && vma->vm_start < end);
1939        *insertion_point = vma;
1940        if (vma)
1941                vma->vm_prev = prev;
1942        tail_vma->vm_next = NULL;
1943        if (mm->unmap_area == arch_unmap_area)
1944                addr = prev ? prev->vm_end : mm->mmap_base;
1945        else
1946                addr = vma ?  vma->vm_start : mm->mmap_base;
1947        mm->unmap_area(mm, addr);
1948        mm->mmap_cache = NULL;          /* Kill the cache. */
1949}
1950
1951/*
1952 * __split_vma() bypasses sysctl_max_map_count checking.  We use this on the
1953 * munmap path where it doesn't make sense to fail.
1954 */
1955static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1956              unsigned long addr, int new_below)
1957{
1958        struct mempolicy *pol;
1959        struct vm_area_struct *new;
1960        int err = -ENOMEM;
1961
1962        if (is_vm_hugetlb_page(vma) && (addr &
1963                                        ~(huge_page_mask(hstate_vma(vma)))))
1964                return -EINVAL;
1965
1966        new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1967        if (!new)
1968                goto out_err;
1969
1970        /* most fields are the same, copy all, and then fixup */
1971        *new = *vma;
1972
1973        INIT_LIST_HEAD(&new->anon_vma_chain);
1974
1975        if (new_below)
1976                new->vm_end = addr;
1977        else {
1978                new->vm_start = addr;
1979                new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1980        }
1981
1982        pol = mpol_dup(vma_policy(vma));
1983        if (IS_ERR(pol)) {
1984                err = PTR_ERR(pol);
1985                goto out_free_vma;
1986        }
1987        vma_set_policy(new, pol);
1988
1989        if (anon_vma_clone(new, vma))
1990                goto out_free_mpol;
1991
1992        if (new->vm_file) {
1993                get_file(new->vm_file);
1994                if (vma->vm_flags & VM_EXECUTABLE)
1995                        added_exe_file_vma(mm);
1996        }
1997
1998        if (new->vm_ops && new->vm_ops->open)
1999                new->vm_ops->open(new);
2000
2001        if (new_below)
2002                err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2003                        ((addr - new->vm_start) >> PAGE_SHIFT), new);
2004        else
2005                err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2006
2007        /* Success. */
2008        if (!err)
2009                return 0;
2010
2011        /* Clean everything up if vma_adjust failed. */
2012        if (new->vm_ops && new->vm_ops->close)
2013                new->vm_ops->close(new);
2014        if (new->vm_file) {
2015                if (vma->vm_flags & VM_EXECUTABLE)
2016                        removed_exe_file_vma(mm);
2017                fput(new->vm_file);
2018        }
2019        unlink_anon_vmas(new);
2020 out_free_mpol:
2021        mpol_put(pol);
2022 out_free_vma:
2023        kmem_cache_free(vm_area_cachep, new);
2024 out_err:
2025        return err;
2026}
2027
2028/*
2029 * Split a vma into two pieces at address 'addr', a new vma is allocated
2030 * either for the first part or the tail.
2031 */
2032int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2033              unsigned long addr, int new_below)
2034{
2035        if (mm->map_count >= sysctl_max_map_count)
2036                return -ENOMEM;
2037
2038        return __split_vma(mm, vma, addr, new_below);
2039}
2040
2041/* Munmap is split into 2 main parts -- this part which finds
2042 * what needs doing, and the areas themselves, which do the
2043 * work.  This now handles partial unmappings.
2044 * Jeremy Fitzhardinge <jeremy@goop.org>
2045 */
2046int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2047{
2048        unsigned long end;
2049        struct vm_area_struct *vma, *prev, *last;
2050
2051        if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2052                return -EINVAL;
2053
2054        if ((len = PAGE_ALIGN(len)) == 0)
2055                return -EINVAL;
2056
2057        /* Find the first overlapping VMA */
2058        vma = find_vma(mm, start);
2059        if (!vma)
2060                return 0;
2061        prev = vma->vm_prev;
2062        /* we have  start < vma->vm_end  */
2063
2064        /* if it doesn't overlap, we have nothing.. */
2065        end = start + len;
2066        if (vma->vm_start >= end)
2067                return 0;
2068
2069        /*
2070         * If we need to split any vma, do it now to save pain later.
2071         *
2072         * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2073         * unmapped vm_area_struct will remain in use: so lower split_vma
2074         * places tmp vma above, and higher split_vma places tmp vma below.
2075         */
2076        if (start > vma->vm_start) {
2077                int error;
2078
2079                /*
2080                 * Make sure that map_count on return from munmap() will
2081                 * not exceed its limit; but let map_count go just above
2082                 * its limit temporarily, to help free resources as expected.
2083                 */
2084                if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2085                        return -ENOMEM;
2086
2087                error = __split_vma(mm, vma, start, 0);
2088                if (error)
2089                        return error;
2090                prev = vma;
2091        }
2092
2093        /* Does it split the last one? */
2094        last = find_vma(mm, end);
2095        if (last && end > last->vm_start) {
2096                int error = __split_vma(mm, last, end, 1);
2097                if (error)
2098                        return error;
2099        }
2100        vma = prev? prev->vm_next: mm->mmap;
2101
2102        /*
2103         * unlock any mlock()ed ranges before detaching vmas
2104         */
2105        if (mm->locked_vm) {
2106                struct vm_area_struct *tmp = vma;
2107                while (tmp && tmp->vm_start < end) {
2108                        if (tmp->vm_flags & VM_LOCKED) {
2109                                mm->locked_vm -= vma_pages(tmp);
2110                                munlock_vma_pages_all(tmp);
2111                        }
2112                        tmp = tmp->vm_next;
2113                }
2114        }
2115
2116        /*
2117         * Remove the vma's, and unmap the actual pages
2118         */
2119        detach_vmas_to_be_unmapped(mm, vma, prev, end);
2120        unmap_region(mm, vma, prev, start, end);
2121
2122        /* Fix up all other VM information */
2123        remove_vma_list(mm, vma);
2124
2125        return 0;
2126}
2127
2128int vm_munmap(unsigned long start, size_t len)
2129{
2130        int ret;
2131        struct mm_struct *mm = current->mm;
2132
2133        down_write(&mm->mmap_sem);
2134        ret = do_munmap(mm, start, len);
2135        up_write(&mm->mmap_sem);
2136        return ret;
2137}
2138EXPORT_SYMBOL(vm_munmap);
2139
2140SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2141{
2142        profile_munmap(addr);
2143        return vm_munmap(addr, len);
2144}
2145
2146static inline void verify_mm_writelocked(struct mm_struct *mm)
2147{
2148#ifdef CONFIG_DEBUG_VM
2149        if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2150                WARN_ON(1);
2151                up_read(&mm->mmap_sem);
2152        }
2153#endif
2154}
2155
2156/*
2157 *  this is really a simplified "do_mmap".  it only handles
2158 *  anonymous maps.  eventually we may be able to do some
2159 *  brk-specific accounting here.
2160 */
2161static unsigned long do_brk(unsigned long addr, unsigned long len)
2162{
2163        struct mm_struct * mm = current->mm;
2164        struct vm_area_struct * vma, * prev;
2165        unsigned long flags;
2166        struct rb_node ** rb_link, * rb_parent;
2167        pgoff_t pgoff = addr >> PAGE_SHIFT;
2168        int error;
2169
2170        len = PAGE_ALIGN(len);
2171        if (!len)
2172                return addr;
2173
2174        flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2175
2176        error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2177        if (error & ~PAGE_MASK)
2178                return error;
2179
2180        /*
2181         * mlock MCL_FUTURE?
2182         */
2183        if (mm->def_flags & VM_LOCKED) {
2184                unsigned long locked, lock_limit;
2185                locked = len >> PAGE_SHIFT;
2186                locked += mm->locked_vm;
2187                lock_limit = rlimit(RLIMIT_MEMLOCK);
2188                lock_limit >>= PAGE_SHIFT;
2189                if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2190                        return -EAGAIN;
2191        }
2192
2193        /*
2194         * mm->mmap_sem is required to protect against another thread
2195         * changing the mappings in case we sleep.
2196         */
2197        verify_mm_writelocked(mm);
2198
2199        /*
2200         * Clear old maps.  this also does some error checking for us
2201         */
2202 munmap_back:
2203        vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2204        if (vma && vma->vm_start < addr + len) {
2205                if (do_munmap(mm, addr, len))
2206                        return -ENOMEM;
2207                goto munmap_back;
2208        }
2209
2210        /* Check against address space limits *after* clearing old maps... */
2211        if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2212                return -ENOMEM;
2213
2214        if (mm->map_count > sysctl_max_map_count)
2215                return -ENOMEM;
2216
2217        if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2218                return -ENOMEM;
2219
2220        /* Can we just expand an old private anonymous mapping? */
2221        vma = vma_merge(mm, prev, addr, addr + len, flags,
2222                                        NULL, NULL, pgoff, NULL);
2223        if (vma)
2224                goto out;
2225
2226        /*
2227         * create a vma struct for an anonymous mapping
2228         */
2229        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2230        if (!vma) {
2231                vm_unacct_memory(len >> PAGE_SHIFT);
2232                return -ENOMEM;
2233        }
2234
2235        INIT_LIST_HEAD(&vma->anon_vma_chain);
2236        vma->vm_mm = mm;
2237        vma->vm_start = addr;
2238        vma->vm_end = addr + len;
2239        vma->vm_pgoff = pgoff;
2240        vma->vm_flags = flags;
2241        vma->vm_page_prot = vm_get_page_prot(flags);
2242        vma_link(mm, vma, prev, rb_link, rb_parent);
2243out:
2244        perf_event_mmap(vma);
2245        mm->total_vm += len >> PAGE_SHIFT;
2246        if (flags & VM_LOCKED) {
2247                if (!mlock_vma_pages_range(vma, addr, addr + len))
2248                        mm->locked_vm += (len >> PAGE_SHIFT);
2249        }
2250        return addr;
2251}
2252
2253unsigned long vm_brk(unsigned long addr, unsigned long len)
2254{
2255        struct mm_struct *mm = current->mm;
2256        unsigned long ret;
2257
2258        down_write(&mm->mmap_sem);
2259        ret = do_brk(addr, len);
2260        up_write(&mm->mmap_sem);
2261        return ret;
2262}
2263EXPORT_SYMBOL(vm_brk);
2264
2265/* Release all mmaps. */
2266void exit_mmap(struct mm_struct *mm)
2267{
2268        struct mmu_gather tlb;
2269        struct vm_area_struct *vma;
2270        unsigned long nr_accounted = 0;
2271
2272        /* mm's last user has gone, and its about to be pulled down */
2273        mmu_notifier_release(mm);
2274
2275        if (mm->locked_vm) {
2276                vma = mm->mmap;
2277                while (vma) {
2278                        if (vma->vm_flags & VM_LOCKED)
2279                                munlock_vma_pages_all(vma);
2280                        vma = vma->vm_next;
2281                }
2282        }
2283
2284        arch_exit_mmap(mm);
2285
2286        vma = mm->mmap;
2287        if (!vma)       /* Can happen if dup_mmap() received an OOM */
2288                return;
2289
2290        lru_add_drain();
2291        flush_cache_mm(mm);
2292        tlb_gather_mmu(&tlb, mm, 1);
2293        /* update_hiwater_rss(mm) here? but nobody should be looking */
2294        /* Use -1 here to ensure all VMAs in the mm are unmapped */
2295        unmap_vmas(&tlb, vma, 0, -1);
2296
2297        free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2298        tlb_finish_mmu(&tlb, 0, -1);
2299
2300        /*
2301         * Walk the list again, actually closing and freeing it,
2302         * with preemption enabled, without holding any MM locks.
2303         */
2304        while (vma) {
2305                if (vma->vm_flags & VM_ACCOUNT)
2306                        nr_accounted += vma_pages(vma);
2307                vma = remove_vma(vma);
2308        }
2309        vm_unacct_memory(nr_accounted);
2310
2311        WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2312}
2313
2314/* Insert vm structure into process list sorted by address
2315 * and into the inode's i_mmap tree.  If vm_file is non-NULL
2316 * then i_mmap_mutex is taken here.
2317 */
2318int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2319{
2320        struct vm_area_struct * __vma, * prev;
2321        struct rb_node ** rb_link, * rb_parent;
2322
2323        /*
2324         * The vm_pgoff of a purely anonymous vma should be irrelevant
2325         * until its first write fault, when page's anon_vma and index
2326         * are set.  But now set the vm_pgoff it will almost certainly
2327         * end up with (unless mremap moves it elsewhere before that
2328         * first wfault), so /proc/pid/maps tells a consistent story.
2329         *
2330         * By setting it to reflect the virtual start address of the
2331         * vma, merges and splits can happen in a seamless way, just
2332         * using the existing file pgoff checks and manipulations.
2333         * Similarly in do_mmap_pgoff and in do_brk.
2334         */
2335        if (!vma->vm_file) {
2336                BUG_ON(vma->anon_vma);
2337                vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2338        }
2339        __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2340        if (__vma && __vma->vm_start < vma->vm_end)
2341                return -ENOMEM;
2342        if ((vma->vm_flags & VM_ACCOUNT) &&
2343             security_vm_enough_memory_mm(mm, vma_pages(vma)))
2344                return -ENOMEM;
2345
2346        vma_link(mm, vma, prev, rb_link, rb_parent);
2347        return 0;
2348}
2349
2350/*
2351 * Copy the vma structure to a new location in the same mm,
2352 * prior to moving page table entries, to effect an mremap move.
2353 */
2354struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2355        unsigned long addr, unsigned long len, pgoff_t pgoff)
2356{
2357        struct vm_area_struct *vma = *vmap;
2358        unsigned long vma_start = vma->vm_start;
2359        struct mm_struct *mm = vma->vm_mm;
2360        struct vm_area_struct *new_vma, *prev;
2361        struct rb_node **rb_link, *rb_parent;
2362        struct mempolicy *pol;
2363        bool faulted_in_anon_vma = true;
2364
2365        /*
2366         * If anonymous vma has not yet been faulted, update new pgoff
2367         * to match new location, to increase its chance of merging.
2368         */
2369        if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2370                pgoff = addr >> PAGE_SHIFT;
2371                faulted_in_anon_vma = false;
2372        }
2373
2374        find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2375        new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2376                        vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2377        if (new_vma) {
2378                /*
2379                 * Source vma may have been merged into new_vma
2380                 */
2381                if (unlikely(vma_start >= new_vma->vm_start &&
2382                             vma_start < new_vma->vm_end)) {
2383                        /*
2384                         * The only way we can get a vma_merge with
2385                         * self during an mremap is if the vma hasn't
2386                         * been faulted in yet and we were allowed to
2387                         * reset the dst vma->vm_pgoff to the
2388                         * destination address of the mremap to allow
2389                         * the merge to happen. mremap must change the
2390                         * vm_pgoff linearity between src and dst vmas
2391                         * (in turn preventing a vma_merge) to be
2392                         * safe. It is only safe to keep the vm_pgoff
2393                         * linear if there are no pages mapped yet.
2394                         */
2395                        VM_BUG_ON(faulted_in_anon_vma);
2396                        *vmap = new_vma;
2397                } else
2398                        anon_vma_moveto_tail(new_vma);
2399        } else {
2400                new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2401                if (new_vma) {
2402                        *new_vma = *vma;
2403                        pol = mpol_dup(vma_policy(vma));
2404                        if (IS_ERR(pol))
2405                                goto out_free_vma;
2406                        INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2407                        if (anon_vma_clone(new_vma, vma))
2408                                goto out_free_mempol;
2409                        vma_set_policy(new_vma, pol);
2410                        new_vma->vm_start = addr;
2411                        new_vma->vm_end = addr + len;
2412                        new_vma->vm_pgoff = pgoff;
2413                        if (new_vma->vm_file) {
2414                                get_file(new_vma->vm_file);
2415
2416                                if (vma->vm_flags & VM_EXECUTABLE)
2417                                        added_exe_file_vma(mm);
2418                        }
2419                        if (new_vma->vm_ops && new_vma->vm_ops->open)
2420                                new_vma->vm_ops->open(new_vma);
2421                        vma_link(mm, new_vma, prev, rb_link, rb_parent);
2422                }
2423        }
2424        return new_vma;
2425
2426 out_free_mempol:
2427        mpol_put(pol);
2428 out_free_vma:
2429        kmem_cache_free(vm_area_cachep, new_vma);
2430        return NULL;
2431}
2432
2433/*
2434 * Return true if the calling process may expand its vm space by the passed
2435 * number of pages
2436 */
2437int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2438{
2439        unsigned long cur = mm->total_vm;       /* pages */
2440        unsigned long lim;
2441
2442        lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2443
2444        if (cur + npages > lim)
2445                return 0;
2446        return 1;
2447}
2448
2449
2450static int special_mapping_fault(struct vm_area_struct *vma,
2451                                struct vm_fault *vmf)
2452{
2453        pgoff_t pgoff;
2454        struct page **pages;
2455
2456        /*
2457         * special mappings have no vm_file, and in that case, the mm
2458         * uses vm_pgoff internally. So we have to subtract it from here.
2459         * We are allowed to do this because we are the mm; do not copy
2460         * this code into drivers!
2461         */
2462        pgoff = vmf->pgoff - vma->vm_pgoff;
2463
2464        for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2465                pgoff--;
2466
2467        if (*pages) {
2468                struct page *page = *pages;
2469                get_page(page);
2470                vmf->page = page;
2471                return 0;
2472        }
2473
2474        return VM_FAULT_SIGBUS;
2475}
2476
2477/*
2478 * Having a close hook prevents vma merging regardless of flags.
2479 */
2480static void special_mapping_close(struct vm_area_struct *vma)
2481{
2482}
2483
2484static const struct vm_operations_struct special_mapping_vmops = {
2485        .close = special_mapping_close,
2486        .fault = special_mapping_fault,
2487};
2488
2489/*
2490 * Called with mm->mmap_sem held for writing.
2491 * Insert a new vma covering the given region, with the given flags.
2492 * Its pages are supplied by the given array of struct page *.
2493 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2494 * The region past the last page supplied will always produce SIGBUS.
2495 * The array pointer and the pages it points to are assumed to stay alive
2496 * for as long as this mapping might exist.
2497 */
2498int install_special_mapping(struct mm_struct *mm,
2499                            unsigned long addr, unsigned long len,
2500                            unsigned long vm_flags, struct page **pages)
2501{
2502        int ret;
2503        struct vm_area_struct *vma;
2504
2505        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2506        if (unlikely(vma == NULL))
2507                return -ENOMEM;
2508
2509        INIT_LIST_HEAD(&vma->anon_vma_chain);
2510        vma->vm_mm = mm;
2511        vma->vm_start = addr;
2512        vma->vm_end = addr + len;
2513
2514        vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2515        vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2516
2517        vma->vm_ops = &special_mapping_vmops;
2518        vma->vm_private_data = pages;
2519
2520        ret = insert_vm_struct(mm, vma);
2521        if (ret)
2522                goto out;
2523
2524        mm->total_vm += len >> PAGE_SHIFT;
2525
2526        perf_event_mmap(vma);
2527
2528        return 0;
2529
2530out:
2531        kmem_cache_free(vm_area_cachep, vma);
2532        return ret;
2533}
2534
2535static DEFINE_MUTEX(mm_all_locks_mutex);
2536
2537static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2538{
2539        if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2540                /*
2541                 * The LSB of head.next can't change from under us
2542                 * because we hold the mm_all_locks_mutex.
2543                 */
2544                mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2545                /*
2546                 * We can safely modify head.next after taking the
2547                 * anon_vma->root->mutex. If some other vma in this mm shares
2548                 * the same anon_vma we won't take it again.
2549                 *
2550                 * No need of atomic instructions here, head.next
2551                 * can't change from under us thanks to the
2552                 * anon_vma->root->mutex.
2553                 */
2554                if (__test_and_set_bit(0, (unsigned long *)
2555                                       &anon_vma->root->head.next))
2556                        BUG();
2557        }
2558}
2559
2560static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2561{
2562        if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2563                /*
2564                 * AS_MM_ALL_LOCKS can't change from under us because
2565                 * we hold the mm_all_locks_mutex.
2566                 *
2567                 * Operations on ->flags have to be atomic because
2568                 * even if AS_MM_ALL_LOCKS is stable thanks to the
2569                 * mm_all_locks_mutex, there may be other cpus
2570                 * changing other bitflags in parallel to us.
2571                 */
2572                if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2573                        BUG();
2574                mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2575        }
2576}
2577
2578/*
2579 * This operation locks against the VM for all pte/vma/mm related
2580 * operations that could ever happen on a certain mm. This includes
2581 * vmtruncate, try_to_unmap, and all page faults.
2582 *
2583 * The caller must take the mmap_sem in write mode before calling
2584 * mm_take_all_locks(). The caller isn't allowed to release the
2585 * mmap_sem until mm_drop_all_locks() returns.
2586 *
2587 * mmap_sem in write mode is required in order to block all operations
2588 * that could modify pagetables and free pages without need of
2589 * altering the vma layout (for example populate_range() with
2590 * nonlinear vmas). It's also needed in write mode to avoid new
2591 * anon_vmas to be associated with existing vmas.
2592 *
2593 * A single task can't take more than one mm_take_all_locks() in a row
2594 * or it would deadlock.
2595 *
2596 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2597 * mapping->flags avoid to take the same lock twice, if more than one
2598 * vma in this mm is backed by the same anon_vma or address_space.
2599 *
2600 * We can take all the locks in random order because the VM code
2601 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2602 * takes more than one of them in a row. Secondly we're protected
2603 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2604 *
2605 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2606 * that may have to take thousand of locks.
2607 *
2608 * mm_take_all_locks() can fail if it's interrupted by signals.
2609 */
2610int mm_take_all_locks(struct mm_struct *mm)
2611{
2612        struct vm_area_struct *vma;
2613        struct anon_vma_chain *avc;
2614
2615        BUG_ON(down_read_trylock(&mm->mmap_sem));
2616
2617        mutex_lock(&mm_all_locks_mutex);
2618
2619        for (vma = mm->mmap; vma; vma = vma->vm_next) {
2620                if (signal_pending(current))
2621                        goto out_unlock;
2622                if (vma->vm_file && vma->vm_file->f_mapping)
2623                        vm_lock_mapping(mm, vma->vm_file->f_mapping);
2624        }
2625
2626        for (vma = mm->mmap; vma; vma = vma->vm_next) {
2627                if (signal_pending(current))
2628                        goto out_unlock;
2629                if (vma->anon_vma)
2630                        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2631                                vm_lock_anon_vma(mm, avc->anon_vma);
2632        }
2633
2634        return 0;
2635
2636out_unlock:
2637        mm_drop_all_locks(mm);
2638        return -EINTR;
2639}
2640
2641static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2642{
2643        if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2644                /*
2645                 * The LSB of head.next can't change to 0 from under
2646                 * us because we hold the mm_all_locks_mutex.
2647                 *
2648                 * We must however clear the bitflag before unlocking
2649                 * the vma so the users using the anon_vma->head will
2650                 * never see our bitflag.
2651                 *
2652                 * No need of atomic instructions here, head.next
2653                 * can't change from under us until we release the
2654                 * anon_vma->root->mutex.
2655                 */
2656                if (!__test_and_clear_bit(0, (unsigned long *)
2657                                          &anon_vma->root->head.next))
2658                        BUG();
2659                anon_vma_unlock(anon_vma);
2660        }
2661}
2662
2663static void vm_unlock_mapping(struct address_space *mapping)
2664{
2665        if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2666                /*
2667                 * AS_MM_ALL_LOCKS can't change to 0 from under us
2668                 * because we hold the mm_all_locks_mutex.
2669                 */
2670                mutex_unlock(&mapping->i_mmap_mutex);
2671                if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2672                                        &mapping->flags))
2673                        BUG();
2674        }
2675}
2676
2677/*
2678 * The mmap_sem cannot be released by the caller until
2679 * mm_drop_all_locks() returns.
2680 */
2681void mm_drop_all_locks(struct mm_struct *mm)
2682{
2683        struct vm_area_struct *vma;
2684        struct anon_vma_chain *avc;
2685
2686        BUG_ON(down_read_trylock(&mm->mmap_sem));
2687        BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2688
2689        for (vma = mm->mmap; vma; vma = vma->vm_next) {
2690                if (vma->anon_vma)
2691                        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2692                                vm_unlock_anon_vma(avc->anon_vma);
2693                if (vma->vm_file && vma->vm_file->f_mapping)
2694                        vm_unlock_mapping(vma->vm_file->f_mapping);
2695        }
2696
2697        mutex_unlock(&mm_all_locks_mutex);
2698}
2699
2700/*
2701 * initialise the VMA slab
2702 */
2703void __init mmap_init(void)
2704{
2705        int ret;
2706
2707        ret = percpu_counter_init(&vm_committed_as, 0);
2708        VM_BUG_ON(ret);
2709}
2710
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.