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#include <linux/rbtree_augmented.h>
  35#include <linux/sched/sysctl.h>
  36
  37#include <asm/uaccess.h>
  38#include <asm/cacheflush.h>
  39#include <asm/tlb.h>
  40#include <asm/mmu_context.h>
  41
  42#include "internal.h"
  43
  44#ifndef arch_mmap_check
  45#define arch_mmap_check(addr, len, flags)       (0)
  46#endif
  47
  48#ifndef arch_rebalance_pgtables
  49#define arch_rebalance_pgtables(addr, len)              (addr)
  50#endif
  51
  52static void unmap_region(struct mm_struct *mm,
  53                struct vm_area_struct *vma, struct vm_area_struct *prev,
  54                unsigned long start, unsigned long end);
  55
  56/* description of effects of mapping type and prot in current implementation.
  57 * this is due to the limited x86 page protection hardware.  The expected
  58 * behavior is in parens:
  59 *
  60 * map_type     prot
  61 *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  62 * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  63 *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  64 *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  65 *              
  66 * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  67 *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  68 *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  69 *
  70 */
  71pgprot_t protection_map[16] = {
  72        __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  73        __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  74};
  75
  76pgprot_t vm_get_page_prot(unsigned long vm_flags)
  77{
  78        return __pgprot(pgprot_val(protection_map[vm_flags &
  79                                (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
  80                        pgprot_val(arch_vm_get_page_prot(vm_flags)));
  81}
  82EXPORT_SYMBOL(vm_get_page_prot);
  83
  84int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;  /* heuristic overcommit */
  85int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
  86int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
  87/*
  88 * Make sure vm_committed_as in one cacheline and not cacheline shared with
  89 * other variables. It can be updated by several CPUs frequently.
  90 */
  91struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
  92
  93/*
  94 * The global memory commitment made in the system can be a metric
  95 * that can be used to drive ballooning decisions when Linux is hosted
  96 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
  97 * balancing memory across competing virtual machines that are hosted.
  98 * Several metrics drive this policy engine including the guest reported
  99 * memory commitment.
 100 */
 101unsigned long vm_memory_committed(void)
 102{
 103        return percpu_counter_read_positive(&vm_committed_as);
 104}
 105EXPORT_SYMBOL_GPL(vm_memory_committed);
 106
 107/*
 108 * Check that a process has enough memory to allocate a new virtual
 109 * mapping. 0 means there is enough memory for the allocation to
 110 * succeed and -ENOMEM implies there is not.
 111 *
 112 * We currently support three overcommit policies, which are set via the
 113 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 114 *
 115 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 116 * Additional code 2002 Jul 20 by Robert Love.
 117 *
 118 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 119 *
 120 * Note this is a helper function intended to be used by LSMs which
 121 * wish to use this logic.
 122 */
 123int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 124{
 125        unsigned long free, allowed;
 126
 127        vm_acct_memory(pages);
 128
 129        /*
 130         * Sometimes we want to use more memory than we have
 131         */
 132        if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 133                return 0;
 134
 135        if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 136                free = global_page_state(NR_FREE_PAGES);
 137                free += global_page_state(NR_FILE_PAGES);
 138
 139                /*
 140                 * shmem pages shouldn't be counted as free in this
 141                 * case, they can't be purged, only swapped out, and
 142                 * that won't affect the overall amount of available
 143                 * memory in the system.
 144                 */
 145                free -= global_page_state(NR_SHMEM);
 146
 147                free += get_nr_swap_pages();
 148
 149                /*
 150                 * Any slabs which are created with the
 151                 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 152                 * which are reclaimable, under pressure.  The dentry
 153                 * cache and most inode caches should fall into this
 154                 */
 155                free += global_page_state(NR_SLAB_RECLAIMABLE);
 156
 157                /*
 158                 * Leave reserved pages. The pages are not for anonymous pages.
 159                 */
 160                if (free <= totalreserve_pages)
 161                        goto error;
 162                else
 163                        free -= totalreserve_pages;
 164
 165                /*
 166                 * Leave the last 3% for root
 167                 */
 168                if (!cap_sys_admin)
 169                        free -= free / 32;
 170
 171                if (free > pages)
 172                        return 0;
 173
 174                goto error;
 175        }
 176
 177        allowed = (totalram_pages - hugetlb_total_pages())
 178                * sysctl_overcommit_ratio / 100;
 179        /*
 180         * Leave the last 3% for root
 181         */
 182        if (!cap_sys_admin)
 183                allowed -= allowed / 32;
 184        allowed += total_swap_pages;
 185
 186        /* Don't let a single process grow too big:
 187           leave 3% of the size of this process for other processes */
 188        if (mm)
 189                allowed -= mm->total_vm / 32;
 190
 191        if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 192                return 0;
 193error:
 194        vm_unacct_memory(pages);
 195
 196        return -ENOMEM;
 197}
 198
 199/*
 200 * Requires inode->i_mapping->i_mmap_mutex
 201 */
 202static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 203                struct file *file, struct address_space *mapping)
 204{
 205        if (vma->vm_flags & VM_DENYWRITE)
 206                atomic_inc(&file_inode(file)->i_writecount);
 207        if (vma->vm_flags & VM_SHARED)
 208                mapping->i_mmap_writable--;
 209
 210        flush_dcache_mmap_lock(mapping);
 211        if (unlikely(vma->vm_flags & VM_NONLINEAR))
 212                list_del_init(&vma->shared.nonlinear);
 213        else
 214                vma_interval_tree_remove(vma, &mapping->i_mmap);
 215        flush_dcache_mmap_unlock(mapping);
 216}
 217
 218/*
 219 * Unlink a file-based vm structure from its interval tree, to hide
 220 * vma from rmap and vmtruncate before freeing its page tables.
 221 */
 222void unlink_file_vma(struct vm_area_struct *vma)
 223{
 224        struct file *file = vma->vm_file;
 225
 226        if (file) {
 227                struct address_space *mapping = file->f_mapping;
 228                mutex_lock(&mapping->i_mmap_mutex);
 229                __remove_shared_vm_struct(vma, file, mapping);
 230                mutex_unlock(&mapping->i_mmap_mutex);
 231        }
 232}
 233
 234/*
 235 * Close a vm structure and free it, returning the next.
 236 */
 237static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 238{
 239        struct vm_area_struct *next = vma->vm_next;
 240
 241        might_sleep();
 242        if (vma->vm_ops && vma->vm_ops->close)
 243                vma->vm_ops->close(vma);
 244        if (vma->vm_file)
 245                fput(vma->vm_file);
 246        mpol_put(vma_policy(vma));
 247        kmem_cache_free(vm_area_cachep, vma);
 248        return next;
 249}
 250
 251static unsigned long do_brk(unsigned long addr, unsigned long len);
 252
 253SYSCALL_DEFINE1(brk, unsigned long, brk)
 254{
 255        unsigned long rlim, retval;
 256        unsigned long newbrk, oldbrk;
 257        struct mm_struct *mm = current->mm;
 258        unsigned long min_brk;
 259        bool populate;
 260
 261        down_write(&mm->mmap_sem);
 262
 263#ifdef CONFIG_COMPAT_BRK
 264        /*
 265         * CONFIG_COMPAT_BRK can still be overridden by setting
 266         * randomize_va_space to 2, which will still cause mm->start_brk
 267         * to be arbitrarily shifted
 268         */
 269        if (current->brk_randomized)
 270                min_brk = mm->start_brk;
 271        else
 272                min_brk = mm->end_data;
 273#else
 274        min_brk = mm->start_brk;
 275#endif
 276        if (brk < min_brk)
 277                goto out;
 278
 279        /*
 280         * Check against rlimit here. If this check is done later after the test
 281         * of oldbrk with newbrk then it can escape the test and let the data
 282         * segment grow beyond its set limit the in case where the limit is
 283         * not page aligned -Ram Gupta
 284         */
 285        rlim = rlimit(RLIMIT_DATA);
 286        if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
 287                        (mm->end_data - mm->start_data) > rlim)
 288                goto out;
 289
 290        newbrk = PAGE_ALIGN(brk);
 291        oldbrk = PAGE_ALIGN(mm->brk);
 292        if (oldbrk == newbrk)
 293                goto set_brk;
 294
 295        /* Always allow shrinking brk. */
 296        if (brk <= mm->brk) {
 297                if (!do_munmap(mm, newbrk, oldbrk-newbrk))
 298                        goto set_brk;
 299                goto out;
 300        }
 301
 302        /* Check against existing mmap mappings. */
 303        if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 304                goto out;
 305
 306        /* Ok, looks good - let it rip. */
 307        if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
 308                goto out;
 309
 310set_brk:
 311        mm->brk = brk;
 312        populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
 313        up_write(&mm->mmap_sem);
 314        if (populate)
 315                mm_populate(oldbrk, newbrk - oldbrk);
 316        return brk;
 317
 318out:
 319        retval = mm->brk;
 320        up_write(&mm->mmap_sem);
 321        return retval;
 322}
 323
 324static long vma_compute_subtree_gap(struct vm_area_struct *vma)
 325{
 326        unsigned long max, subtree_gap;
 327        max = vma->vm_start;
 328        if (vma->vm_prev)
 329                max -= vma->vm_prev->vm_end;
 330        if (vma->vm_rb.rb_left) {
 331                subtree_gap = rb_entry(vma->vm_rb.rb_left,
 332                                struct vm_area_struct, vm_rb)->rb_subtree_gap;
 333                if (subtree_gap > max)
 334                        max = subtree_gap;
 335        }
 336        if (vma->vm_rb.rb_right) {
 337                subtree_gap = rb_entry(vma->vm_rb.rb_right,
 338                                struct vm_area_struct, vm_rb)->rb_subtree_gap;
 339                if (subtree_gap > max)
 340                        max = subtree_gap;
 341        }
 342        return max;
 343}
 344
 345#ifdef CONFIG_DEBUG_VM_RB
 346static int browse_rb(struct rb_root *root)
 347{
 348        int i = 0, j, bug = 0;
 349        struct rb_node *nd, *pn = NULL;
 350        unsigned long prev = 0, pend = 0;
 351
 352        for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 353                struct vm_area_struct *vma;
 354                vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 355                if (vma->vm_start < prev) {
 356                        printk("vm_start %lx prev %lx\n", vma->vm_start, prev);
 357                        bug = 1;
 358                }
 359                if (vma->vm_start < pend) {
 360                        printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
 361                        bug = 1;
 362                }
 363                if (vma->vm_start > vma->vm_end) {
 364                        printk("vm_end %lx < vm_start %lx\n",
 365                                vma->vm_end, vma->vm_start);
 366                        bug = 1;
 367                }
 368                if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
 369                        printk("free gap %lx, correct %lx\n",
 370                               vma->rb_subtree_gap,
 371                               vma_compute_subtree_gap(vma));
 372                        bug = 1;
 373                }
 374                i++;
 375                pn = nd;
 376                prev = vma->vm_start;
 377                pend = vma->vm_end;
 378        }
 379        j = 0;
 380        for (nd = pn; nd; nd = rb_prev(nd))
 381                j++;
 382        if (i != j) {
 383                printk("backwards %d, forwards %d\n", j, i);
 384                bug = 1;
 385        }
 386        return bug ? -1 : i;
 387}
 388
 389static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
 390{
 391        struct rb_node *nd;
 392
 393        for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 394                struct vm_area_struct *vma;
 395                vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 396                BUG_ON(vma != ignore &&
 397                       vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
 398        }
 399}
 400
 401void validate_mm(struct mm_struct *mm)
 402{
 403        int bug = 0;
 404        int i = 0;
 405        unsigned long highest_address = 0;
 406        struct vm_area_struct *vma = mm->mmap;
 407        while (vma) {
 408                struct anon_vma_chain *avc;
 409                vma_lock_anon_vma(vma);
 410                list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 411                        anon_vma_interval_tree_verify(avc);
 412                vma_unlock_anon_vma(vma);
 413                highest_address = vma->vm_end;
 414                vma = vma->vm_next;
 415                i++;
 416        }
 417        if (i != mm->map_count) {
 418                printk("map_count %d vm_next %d\n", mm->map_count, i);
 419                bug = 1;
 420        }
 421        if (highest_address != mm->highest_vm_end) {
 422                printk("mm->highest_vm_end %lx, found %lx\n",
 423                       mm->highest_vm_end, highest_address);
 424                bug = 1;
 425        }
 426        i = browse_rb(&mm->mm_rb);
 427        if (i != mm->map_count) {
 428                printk("map_count %d rb %d\n", mm->map_count, i);
 429                bug = 1;
 430        }
 431        BUG_ON(bug);
 432}
 433#else
 434#define validate_mm_rb(root, ignore) do { } while (0)
 435#define validate_mm(mm) do { } while (0)
 436#endif
 437
 438RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
 439                     unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
 440
 441/*
 442 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
 443 * vma->vm_prev->vm_end values changed, without modifying the vma's position
 444 * in the rbtree.
 445 */
 446static void vma_gap_update(struct vm_area_struct *vma)
 447{
 448        /*
 449         * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
 450         * function that does exacltly what we want.
 451         */
 452        vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
 453}
 454
 455static inline void vma_rb_insert(struct vm_area_struct *vma,
 456                                 struct rb_root *root)
 457{
 458        /* All rb_subtree_gap values must be consistent prior to insertion */
 459        validate_mm_rb(root, NULL);
 460
 461        rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
 462}
 463
 464static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
 465{
 466        /*
 467         * All rb_subtree_gap values must be consistent prior to erase,
 468         * with the possible exception of the vma being erased.
 469         */
 470        validate_mm_rb(root, vma);
 471
 472        /*
 473         * Note rb_erase_augmented is a fairly large inline function,
 474         * so make sure we instantiate it only once with our desired
 475         * augmented rbtree callbacks.
 476         */
 477        rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
 478}
 479
 480/*
 481 * vma has some anon_vma assigned, and is already inserted on that
 482 * anon_vma's interval trees.
 483 *
 484 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
 485 * vma must be removed from the anon_vma's interval trees using
 486 * anon_vma_interval_tree_pre_update_vma().
 487 *
 488 * After the update, the vma will be reinserted using
 489 * anon_vma_interval_tree_post_update_vma().
 490 *
 491 * The entire update must be protected by exclusive mmap_sem and by
 492 * the root anon_vma's mutex.
 493 */
 494static inline void
 495anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
 496{
 497        struct anon_vma_chain *avc;
 498
 499        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 500                anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
 501}
 502
 503static inline void
 504anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
 505{
 506        struct anon_vma_chain *avc;
 507
 508        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 509                anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
 510}
 511
 512static int find_vma_links(struct mm_struct *mm, unsigned long addr,
 513                unsigned long end, struct vm_area_struct **pprev,
 514                struct rb_node ***rb_link, struct rb_node **rb_parent)
 515{
 516        struct rb_node **__rb_link, *__rb_parent, *rb_prev;
 517
 518        __rb_link = &mm->mm_rb.rb_node;
 519        rb_prev = __rb_parent = NULL;
 520
 521        while (*__rb_link) {
 522                struct vm_area_struct *vma_tmp;
 523
 524                __rb_parent = *__rb_link;
 525                vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
 526
 527                if (vma_tmp->vm_end > addr) {
 528                        /* Fail if an existing vma overlaps the area */
 529                        if (vma_tmp->vm_start < end)
 530                                return -ENOMEM;
 531                        __rb_link = &__rb_parent->rb_left;
 532                } else {
 533                        rb_prev = __rb_parent;
 534                        __rb_link = &__rb_parent->rb_right;
 535                }
 536        }
 537
 538        *pprev = NULL;
 539        if (rb_prev)
 540                *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 541        *rb_link = __rb_link;
 542        *rb_parent = __rb_parent;
 543        return 0;
 544}
 545
 546void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
 547                struct rb_node **rb_link, struct rb_node *rb_parent)
 548{
 549        /* Update tracking information for the gap following the new vma. */
 550        if (vma->vm_next)
 551                vma_gap_update(vma->vm_next);
 552        else
 553                mm->highest_vm_end = vma->vm_end;
 554
 555        /*
 556         * vma->vm_prev wasn't known when we followed the rbtree to find the
 557         * correct insertion point for that vma. As a result, we could not
 558         * update the vma vm_rb parents rb_subtree_gap values on the way down.
 559         * So, we first insert the vma with a zero rb_subtree_gap value
 560         * (to be consistent with what we did on the way down), and then
 561         * immediately update the gap to the correct value. Finally we
 562         * rebalance the rbtree after all augmented values have been set.
 563         */
 564        rb_link_node(&vma->vm_rb, rb_parent, rb_link);
 565        vma->rb_subtree_gap = 0;
 566        vma_gap_update(vma);
 567        vma_rb_insert(vma, &mm->mm_rb);
 568}
 569
 570static void __vma_link_file(struct vm_area_struct *vma)
 571{
 572        struct file *file;
 573
 574        file = vma->vm_file;
 575        if (file) {
 576                struct address_space *mapping = file->f_mapping;
 577
 578                if (vma->vm_flags & VM_DENYWRITE)
 579                        atomic_dec(&file_inode(file)->i_writecount);
 580                if (vma->vm_flags & VM_SHARED)
 581                        mapping->i_mmap_writable++;
 582
 583                flush_dcache_mmap_lock(mapping);
 584                if (unlikely(vma->vm_flags & VM_NONLINEAR))
 585                        vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 586                else
 587                        vma_interval_tree_insert(vma, &mapping->i_mmap);
 588                flush_dcache_mmap_unlock(mapping);
 589        }
 590}
 591
 592static void
 593__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 594        struct vm_area_struct *prev, struct rb_node **rb_link,
 595        struct rb_node *rb_parent)
 596{
 597        __vma_link_list(mm, vma, prev, rb_parent);
 598        __vma_link_rb(mm, vma, rb_link, rb_parent);
 599}
 600
 601static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 602                        struct vm_area_struct *prev, struct rb_node **rb_link,
 603                        struct rb_node *rb_parent)
 604{
 605        struct address_space *mapping = NULL;
 606
 607        if (vma->vm_file)
 608                mapping = vma->vm_file->f_mapping;
 609
 610        if (mapping)
 611                mutex_lock(&mapping->i_mmap_mutex);
 612
 613        __vma_link(mm, vma, prev, rb_link, rb_parent);
 614        __vma_link_file(vma);
 615
 616        if (mapping)
 617                mutex_unlock(&mapping->i_mmap_mutex);
 618
 619        mm->map_count++;
 620        validate_mm(mm);
 621}
 622
 623/*
 624 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
 625 * mm's list and rbtree.  It has already been inserted into the interval tree.
 626 */
 627static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
 628{
 629        struct vm_area_struct *prev;
 630        struct rb_node **rb_link, *rb_parent;
 631
 632        if (find_vma_links(mm, vma->vm_start, vma->vm_end,
 633                           &prev, &rb_link, &rb_parent))
 634                BUG();
 635        __vma_link(mm, vma, prev, rb_link, rb_parent);
 636        mm->map_count++;
 637}
 638
 639static inline void
 640__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
 641                struct vm_area_struct *prev)
 642{
 643        struct vm_area_struct *next;
 644
 645        vma_rb_erase(vma, &mm->mm_rb);
 646        prev->vm_next = next = vma->vm_next;
 647        if (next)
 648                next->vm_prev = prev;
 649        if (mm->mmap_cache == vma)
 650                mm->mmap_cache = prev;
 651}
 652
 653/*
 654 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
 655 * is already present in an i_mmap tree without adjusting the tree.
 656 * The following helper function should be used when such adjustments
 657 * are necessary.  The "insert" vma (if any) is to be inserted
 658 * before we drop the necessary locks.
 659 */
 660int vma_adjust(struct vm_area_struct *vma, unsigned long start,
 661        unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
 662{
 663        struct mm_struct *mm = vma->vm_mm;
 664        struct vm_area_struct *next = vma->vm_next;
 665        struct vm_area_struct *importer = NULL;
 666        struct address_space *mapping = NULL;
 667        struct rb_root *root = NULL;
 668        struct anon_vma *anon_vma = NULL;
 669        struct file *file = vma->vm_file;
 670        bool start_changed = false, end_changed = false;
 671        long adjust_next = 0;
 672        int remove_next = 0;
 673
 674        if (next && !insert) {
 675                struct vm_area_struct *exporter = NULL;
 676
 677                if (end >= next->vm_end) {
 678                        /*
 679                         * vma expands, overlapping all the next, and
 680                         * perhaps the one after too (mprotect case 6).
 681                         */
 682again:                  remove_next = 1 + (end > next->vm_end);
 683                        end = next->vm_end;
 684                        exporter = next;
 685                        importer = vma;
 686                } else if (end > next->vm_start) {
 687                        /*
 688                         * vma expands, overlapping part of the next:
 689                         * mprotect case 5 shifting the boundary up.
 690                         */
 691                        adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
 692                        exporter = next;
 693                        importer = vma;
 694                } else if (end < vma->vm_end) {
 695                        /*
 696                         * vma shrinks, and !insert tells it's not
 697                         * split_vma inserting another: so it must be
 698                         * mprotect case 4 shifting the boundary down.
 699                         */
 700                        adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
 701                        exporter = vma;
 702                        importer = next;
 703                }
 704
 705                /*
 706                 * Easily overlooked: when mprotect shifts the boundary,
 707                 * make sure the expanding vma has anon_vma set if the
 708                 * shrinking vma had, to cover any anon pages imported.
 709                 */
 710                if (exporter && exporter->anon_vma && !importer->anon_vma) {
 711                        if (anon_vma_clone(importer, exporter))
 712                                return -ENOMEM;
 713                        importer->anon_vma = exporter->anon_vma;
 714                }
 715        }
 716
 717        if (file) {
 718                mapping = file->f_mapping;
 719                if (!(vma->vm_flags & VM_NONLINEAR)) {
 720                        root = &mapping->i_mmap;
 721                        uprobe_munmap(vma, vma->vm_start, vma->vm_end);
 722
 723                        if (adjust_next)
 724                                uprobe_munmap(next, next->vm_start,
 725                                                        next->vm_end);
 726                }
 727
 728                mutex_lock(&mapping->i_mmap_mutex);
 729                if (insert) {
 730                        /*
 731                         * Put into interval tree now, so instantiated pages
 732                         * are visible to arm/parisc __flush_dcache_page
 733                         * throughout; but we cannot insert into address
 734                         * space until vma start or end is updated.
 735                         */
 736                        __vma_link_file(insert);
 737                }
 738        }
 739
 740        vma_adjust_trans_huge(vma, start, end, adjust_next);
 741
 742        anon_vma = vma->anon_vma;
 743        if (!anon_vma && adjust_next)
 744                anon_vma = next->anon_vma;
 745        if (anon_vma) {
 746                VM_BUG_ON(adjust_next && next->anon_vma &&
 747                          anon_vma != next->anon_vma);
 748                anon_vma_lock_write(anon_vma);
 749                anon_vma_interval_tree_pre_update_vma(vma);
 750                if (adjust_next)
 751                        anon_vma_interval_tree_pre_update_vma(next);
 752        }
 753
 754        if (root) {
 755                flush_dcache_mmap_lock(mapping);
 756                vma_interval_tree_remove(vma, root);
 757                if (adjust_next)
 758                        vma_interval_tree_remove(next, root);
 759        }
 760
 761        if (start != vma->vm_start) {
 762                vma->vm_start = start;
 763                start_changed = true;
 764        }
 765        if (end != vma->vm_end) {
 766                vma->vm_end = end;
 767                end_changed = true;
 768        }
 769        vma->vm_pgoff = pgoff;
 770        if (adjust_next) {
 771                next->vm_start += adjust_next << PAGE_SHIFT;
 772                next->vm_pgoff += adjust_next;
 773        }
 774
 775        if (root) {
 776                if (adjust_next)
 777                        vma_interval_tree_insert(next, root);
 778                vma_interval_tree_insert(vma, root);
 779                flush_dcache_mmap_unlock(mapping);
 780        }
 781
 782        if (remove_next) {
 783                /*
 784                 * vma_merge has merged next into vma, and needs
 785                 * us to remove next before dropping the locks.
 786                 */
 787                __vma_unlink(mm, next, vma);
 788                if (file)
 789                        __remove_shared_vm_struct(next, file, mapping);
 790        } else if (insert) {
 791                /*
 792                 * split_vma has split insert from vma, and needs
 793                 * us to insert it before dropping the locks
 794                 * (it may either follow vma or precede it).
 795                 */
 796                __insert_vm_struct(mm, insert);
 797        } else {
 798                if (start_changed)
 799                        vma_gap_update(vma);
 800                if (end_changed) {
 801                        if (!next)
 802                                mm->highest_vm_end = end;
 803                        else if (!adjust_next)
 804                                vma_gap_update(next);
 805                }
 806        }
 807
 808        if (anon_vma) {
 809                anon_vma_interval_tree_post_update_vma(vma);
 810                if (adjust_next)
 811                        anon_vma_interval_tree_post_update_vma(next);
 812                anon_vma_unlock_write(anon_vma);
 813        }
 814        if (mapping)
 815                mutex_unlock(&mapping->i_mmap_mutex);
 816
 817        if (root) {
 818                uprobe_mmap(vma);
 819
 820                if (adjust_next)
 821                        uprobe_mmap(next);
 822        }
 823
 824        if (remove_next) {
 825                if (file) {
 826                        uprobe_munmap(next, next->vm_start, next->vm_end);
 827                        fput(file);
 828                }
 829                if (next->anon_vma)
 830                        anon_vma_merge(vma, next);
 831                mm->map_count--;
 832                mpol_put(vma_policy(next));
 833                kmem_cache_free(vm_area_cachep, next);
 834                /*
 835                 * In mprotect's case 6 (see comments on vma_merge),
 836                 * we must remove another next too. It would clutter
 837                 * up the code too much to do both in one go.
 838                 */
 839                next = vma->vm_next;
 840                if (remove_next == 2)
 841                        goto again;
 842                else if (next)
 843                        vma_gap_update(next);
 844                else
 845                        mm->highest_vm_end = end;
 846        }
 847        if (insert && file)
 848                uprobe_mmap(insert);
 849
 850        validate_mm(mm);
 851
 852        return 0;
 853}
 854
 855/*
 856 * If the vma has a ->close operation then the driver probably needs to release
 857 * per-vma resources, so we don't attempt to merge those.
 858 */
 859static inline int is_mergeable_vma(struct vm_area_struct *vma,
 860                        struct file *file, unsigned long vm_flags)
 861{
 862        if (vma->vm_flags ^ vm_flags)
 863                return 0;
 864        if (vma->vm_file != file)
 865                return 0;
 866        if (vma->vm_ops && vma->vm_ops->close)
 867                return 0;
 868        return 1;
 869}
 870
 871static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
 872                                        struct anon_vma *anon_vma2,
 873                                        struct vm_area_struct *vma)
 874{
 875        /*
 876         * The list_is_singular() test is to avoid merging VMA cloned from
 877         * parents. This can improve scalability caused by anon_vma lock.
 878         */
 879        if ((!anon_vma1 || !anon_vma2) && (!vma ||
 880                list_is_singular(&vma->anon_vma_chain)))
 881                return 1;
 882        return anon_vma1 == anon_vma2;
 883}
 884
 885/*
 886 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 887 * in front of (at a lower virtual address and file offset than) the vma.
 888 *
 889 * We cannot merge two vmas if they have differently assigned (non-NULL)
 890 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 891 *
 892 * We don't check here for the merged mmap wrapping around the end of pagecache
 893 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
 894 * wrap, nor mmaps which cover the final page at index -1UL.
 895 */
 896static int
 897can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
 898        struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 899{
 900        if (is_mergeable_vma(vma, file, vm_flags) &&
 901            is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 902                if (vma->vm_pgoff == vm_pgoff)
 903                        return 1;
 904        }
 905        return 0;
 906}
 907
 908/*
 909 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 910 * beyond (at a higher virtual address and file offset than) the vma.
 911 *
 912 * We cannot merge two vmas if they have differently assigned (non-NULL)
 913 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 914 */
 915static int
 916can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 917        struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 918{
 919        if (is_mergeable_vma(vma, file, vm_flags) &&
 920            is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 921                pgoff_t vm_pglen;
 922                vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 923                if (vma->vm_pgoff + vm_pglen == vm_pgoff)
 924                        return 1;
 925        }
 926        return 0;
 927}
 928
 929/*
 930 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
 931 * whether that can be merged with its predecessor or its successor.
 932 * Or both (it neatly fills a hole).
 933 *
 934 * In most cases - when called for mmap, brk or mremap - [addr,end) is
 935 * certain not to be mapped by the time vma_merge is called; but when
 936 * called for mprotect, it is certain to be already mapped (either at
 937 * an offset within prev, or at the start of next), and the flags of
 938 * this area are about to be changed to vm_flags - and the no-change
 939 * case has already been eliminated.
 940 *
 941 * The following mprotect cases have to be considered, where AAAA is
 942 * the area passed down from mprotect_fixup, never extending beyond one
 943 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
 944 *
 945 *     AAAA             AAAA                AAAA          AAAA
 946 *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
 947 *    cannot merge    might become    might become    might become
 948 *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
 949 *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
 950 *    mremap move:                                    PPPPNNNNNNNN 8
 951 *        AAAA
 952 *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
 953 *    might become    case 1 below    case 2 below    case 3 below
 954 *
 955 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
 956 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
 957 */
 958struct vm_area_struct *vma_merge(struct mm_struct *mm,
 959                        struct vm_area_struct *prev, unsigned long addr,
 960                        unsigned long end, unsigned long vm_flags,
 961                        struct anon_vma *anon_vma, struct file *file,
 962                        pgoff_t pgoff, struct mempolicy *policy)
 963{
 964        pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
 965        struct vm_area_struct *area, *next;
 966        int err;
 967
 968        /*
 969         * We later require that vma->vm_flags == vm_flags,
 970         * so this tests vma->vm_flags & VM_SPECIAL, too.
 971         */
 972        if (vm_flags & VM_SPECIAL)
 973                return NULL;
 974
 975        if (prev)
 976                next = prev->vm_next;
 977        else
 978                next = mm->mmap;
 979        area = next;
 980        if (next && next->vm_end == end)                /* cases 6, 7, 8 */
 981                next = next->vm_next;
 982
 983        /*
 984         * Can it merge with the predecessor?
 985         */
 986        if (prev && prev->vm_end == addr &&
 987                        mpol_equal(vma_policy(prev), policy) &&
 988                        can_vma_merge_after(prev, vm_flags,
 989                                                anon_vma, file, pgoff)) {
 990                /*
 991                 * OK, it can.  Can we now merge in the successor as well?
 992                 */
 993                if (next && end == next->vm_start &&
 994                                mpol_equal(policy, vma_policy(next)) &&
 995                                can_vma_merge_before(next, vm_flags,
 996                                        anon_vma, file, pgoff+pglen) &&
 997                                is_mergeable_anon_vma(prev->anon_vma,
 998                                                      next->anon_vma, NULL)) {
 999                                                        /* cases 1, 6 */
1000                        err = vma_adjust(prev, prev->vm_start,
1001                                next->vm_end, prev->vm_pgoff, NULL);
1002                } else                                  /* cases 2, 5, 7 */
1003                        err = vma_adjust(prev, prev->vm_start,
1004                                end, prev->vm_pgoff, NULL);
1005                if (err)
1006                        return NULL;
1007                khugepaged_enter_vma_merge(prev);
1008                return prev;
1009        }
1010
1011        /*
1012         * Can this new request be merged in front of next?
1013         */
1014        if (next && end == next->vm_start &&
1015                        mpol_equal(policy, vma_policy(next)) &&
1016                        can_vma_merge_before(next, vm_flags,
1017                                        anon_vma, file, pgoff+pglen)) {
1018                if (prev && addr < prev->vm_end)        /* case 4 */
1019                        err = vma_adjust(prev, prev->vm_start,
1020                                addr, prev->vm_pgoff, NULL);
1021                else                                    /* cases 3, 8 */
1022                        err = vma_adjust(area, addr, next->vm_end,
1023                                next->vm_pgoff - pglen, NULL);
1024                if (err)
1025                        return NULL;
1026                khugepaged_enter_vma_merge(area);
1027                return area;
1028        }
1029
1030        return NULL;
1031}
1032
1033/*
1034 * Rough compatbility check to quickly see if it's even worth looking
1035 * at sharing an anon_vma.
1036 *
1037 * They need to have the same vm_file, and the flags can only differ
1038 * in things that mprotect may change.
1039 *
1040 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1041 * we can merge the two vma's. For example, we refuse to merge a vma if
1042 * there is a vm_ops->close() function, because that indicates that the
1043 * driver is doing some kind of reference counting. But that doesn't
1044 * really matter for the anon_vma sharing case.
1045 */
1046static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1047{
1048        return a->vm_end == b->vm_start &&
1049                mpol_equal(vma_policy(a), vma_policy(b)) &&
1050                a->vm_file == b->vm_file &&
1051                !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
1052                b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1053}
1054
1055/*
1056 * Do some basic sanity checking to see if we can re-use the anon_vma
1057 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1058 * the same as 'old', the other will be the new one that is trying
1059 * to share the anon_vma.
1060 *
1061 * NOTE! This runs with mm_sem held for reading, so it is possible that
1062 * the anon_vma of 'old' is concurrently in the process of being set up
1063 * by another page fault trying to merge _that_. But that's ok: if it
1064 * is being set up, that automatically means that it will be a singleton
1065 * acceptable for merging, so we can do all of this optimistically. But
1066 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1067 *
1068 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1069 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1070 * is to return an anon_vma that is "complex" due to having gone through
1071 * a fork).
1072 *
1073 * We also make sure that the two vma's are compatible (adjacent,
1074 * and with the same memory policies). That's all stable, even with just
1075 * a read lock on the mm_sem.
1076 */
1077static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1078{
1079        if (anon_vma_compatible(a, b)) {
1080                struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1081
1082                if (anon_vma && list_is_singular(&old->anon_vma_chain))
1083                        return anon_vma;
1084        }
1085        return NULL;
1086}
1087
1088/*
1089 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1090 * neighbouring vmas for a suitable anon_vma, before it goes off
1091 * to allocate a new anon_vma.  It checks because a repetitive
1092 * sequence of mprotects and faults may otherwise lead to distinct
1093 * anon_vmas being allocated, preventing vma merge in subsequent
1094 * mprotect.
1095 */
1096struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1097{
1098        struct anon_vma *anon_vma;
1099        struct vm_area_struct *near;
1100
1101        near = vma->vm_next;
1102        if (!near)
1103                goto try_prev;
1104
1105        anon_vma = reusable_anon_vma(near, vma, near);
1106        if (anon_vma)
1107                return anon_vma;
1108try_prev:
1109        near = vma->vm_prev;
1110        if (!near)
1111                goto none;
1112
1113        anon_vma = reusable_anon_vma(near, near, vma);
1114        if (anon_vma)
1115                return anon_vma;
1116none:
1117        /*
1118         * There's no absolute need to look only at touching neighbours:
1119         * we could search further afield for "compatible" anon_vmas.
1120         * But it would probably just be a waste of time searching,
1121         * or lead to too many vmas hanging off the same anon_vma.
1122         * We're trying to allow mprotect remerging later on,
1123         * not trying to minimize memory used for anon_vmas.
1124         */
1125        return NULL;
1126}
1127
1128#ifdef CONFIG_PROC_FS
1129void vm_stat_account(struct mm_struct *mm, unsigned long flags,
1130                                                struct file *file, long pages)
1131{
1132        const unsigned long stack_flags
1133                = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1134
1135        mm->total_vm += pages;
1136
1137        if (file) {
1138                mm->shared_vm += pages;
1139                if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1140                        mm->exec_vm += pages;
1141        } else if (flags & stack_flags)
1142                mm->stack_vm += pages;
1143}
1144#endif /* CONFIG_PROC_FS */
1145
1146/*
1147 * If a hint addr is less than mmap_min_addr change hint to be as
1148 * low as possible but still greater than mmap_min_addr
1149 */
1150static inline unsigned long round_hint_to_min(unsigned long hint)
1151{
1152        hint &= PAGE_MASK;
1153        if (((void *)hint != NULL) &&
1154            (hint < mmap_min_addr))
1155                return PAGE_ALIGN(mmap_min_addr);
1156        return hint;
1157}
1158
1159/*
1160 * The caller must hold down_write(&current->mm->mmap_sem).
1161 */
1162
1163unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
1164                        unsigned long len, unsigned long prot,
1165                        unsigned long flags, unsigned long pgoff,
1166                        unsigned long *populate)
1167{
1168        struct mm_struct * mm = current->mm;
1169        struct inode *inode;
1170        vm_flags_t vm_flags;
1171
1172        *populate = 0;
1173
1174        /*
1175         * Does the application expect PROT_READ to imply PROT_EXEC?
1176         *
1177         * (the exception is when the underlying filesystem is noexec
1178         *  mounted, in which case we dont add PROT_EXEC.)
1179         */
1180        if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1181                if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1182                        prot |= PROT_EXEC;
1183
1184        if (!len)
1185                return -EINVAL;
1186
1187        if (!(flags & MAP_FIXED))
1188                addr = round_hint_to_min(addr);
1189
1190        /* Careful about overflows.. */
1191        len = PAGE_ALIGN(len);
1192        if (!len)
1193                return -ENOMEM;
1194
1195        /* offset overflow? */
1196        if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1197               return -EOVERFLOW;
1198
1199        /* Too many mappings? */
1200        if (mm->map_count > sysctl_max_map_count)
1201                return -ENOMEM;
1202
1203        /* Obtain the address to map to. we verify (or select) it and ensure
1204         * that it represents a valid section of the address space.
1205         */
1206        addr = get_unmapped_area(file, addr, len, pgoff, flags);
1207        if (addr & ~PAGE_MASK)
1208                return addr;
1209
1210        /* Do simple checking here so the lower-level routines won't have
1211         * to. we assume access permissions have been handled by the open
1212         * of the memory object, so we don't do any here.
1213         */
1214        vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1215                        mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1216
1217        if (flags & MAP_LOCKED)
1218                if (!can_do_mlock())
1219                        return -EPERM;
1220
1221        /* mlock MCL_FUTURE? */
1222        if (vm_flags & VM_LOCKED) {
1223                unsigned long locked, lock_limit;
1224                locked = len >> PAGE_SHIFT;
1225                locked += mm->locked_vm;
1226                lock_limit = rlimit(RLIMIT_MEMLOCK);
1227                lock_limit >>= PAGE_SHIFT;
1228                if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1229                        return -EAGAIN;
1230        }
1231
1232        inode = file ? file_inode(file) : NULL;
1233
1234        if (file) {
1235                switch (flags & MAP_TYPE) {
1236                case MAP_SHARED:
1237                        if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1238                                return -EACCES;
1239
1240                        /*
1241                         * Make sure we don't allow writing to an append-only
1242                         * file..
1243                         */
1244                        if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1245                                return -EACCES;
1246
1247                        /*
1248                         * Make sure there are no mandatory locks on the file.
1249                         */
1250                        if (locks_verify_locked(inode))
1251                                return -EAGAIN;
1252
1253                        vm_flags |= VM_SHARED | VM_MAYSHARE;
1254                        if (!(file->f_mode & FMODE_WRITE))
1255                                vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1256
1257                        /* fall through */
1258                case MAP_PRIVATE:
1259                        if (!(file->f_mode & FMODE_READ))
1260                                return -EACCES;
1261                        if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1262                                if (vm_flags & VM_EXEC)
1263                                        return -EPERM;
1264                                vm_flags &= ~VM_MAYEXEC;
1265                        }
1266
1267                        if (!file->f_op || !file->f_op->mmap)
1268                                return -ENODEV;
1269                        break;
1270
1271                default:
1272                        return -EINVAL;
1273                }
1274        } else {
1275                switch (flags & MAP_TYPE) {
1276                case MAP_SHARED:
1277                        /*
1278                         * Ignore pgoff.
1279                         */
1280                        pgoff = 0;
1281                        vm_flags |= VM_SHARED | VM_MAYSHARE;
1282                        break;
1283                case MAP_PRIVATE:
1284                        /*
1285                         * Set pgoff according to addr for anon_vma.
1286                         */
1287                        pgoff = addr >> PAGE_SHIFT;
1288                        break;
1289                default:
1290                        return -EINVAL;
1291                }
1292        }
1293
1294        /*
1295         * Set 'VM_NORESERVE' if we should not account for the
1296         * memory use of this mapping.
1297         */
1298        if (flags & MAP_NORESERVE) {
1299                /* We honor MAP_NORESERVE if allowed to overcommit */
1300                if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1301                        vm_flags |= VM_NORESERVE;
1302
1303                /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1304                if (file && is_file_hugepages(file))
1305                        vm_flags |= VM_NORESERVE;
1306        }
1307
1308        addr = mmap_region(file, addr, len, vm_flags, pgoff);
1309        if (!IS_ERR_VALUE(addr) &&
1310            ((vm_flags & VM_LOCKED) ||
1311             (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1312                *populate = len;
1313        return addr;
1314}
1315
1316SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1317                unsigned long, prot, unsigned long, flags,
1318                unsigned long, fd, unsigned long, pgoff)
1319{
1320        struct file *file = NULL;
1321        unsigned long retval = -EBADF;
1322
1323        if (!(flags & MAP_ANONYMOUS)) {
1324                audit_mmap_fd(fd, flags);
1325                if (unlikely(flags & MAP_HUGETLB))
1326                        return -EINVAL;
1327                file = fget(fd);
1328                if (!file)
1329                        goto out;
1330                if (is_file_hugepages(file))
1331                        len = ALIGN(len, huge_page_size(hstate_file(file)));
1332        } else if (flags & MAP_HUGETLB) {
1333                struct user_struct *user = NULL;
1334                struct hstate *hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) &
1335                                                   SHM_HUGE_MASK);
1336
1337                if (!hs)
1338                        return -EINVAL;
1339
1340                len = ALIGN(len, huge_page_size(hs));
1341                /*
1342                 * VM_NORESERVE is used because the reservations will be
1343                 * taken when vm_ops->mmap() is called
1344                 * A dummy user value is used because we are not locking
1345                 * memory so no accounting is necessary
1346                 */
1347                file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1348                                VM_NORESERVE,
1349                                &user, HUGETLB_ANONHUGE_INODE,
1350                                (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1351                if (IS_ERR(file))
1352                        return PTR_ERR(file);
1353        }
1354
1355        flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1356
1357        retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1358        if (file)
1359                fput(file);
1360out:
1361        return retval;
1362}
1363
1364#ifdef __ARCH_WANT_SYS_OLD_MMAP
1365struct mmap_arg_struct {
1366        unsigned long addr;
1367        unsigned long len;
1368        unsigned long prot;
1369        unsigned long flags;
1370        unsigned long fd;
1371        unsigned long offset;
1372};
1373
1374SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1375{
1376        struct mmap_arg_struct a;
1377
1378        if (copy_from_user(&a, arg, sizeof(a)))
1379                return -EFAULT;
1380        if (a.offset & ~PAGE_MASK)
1381                return -EINVAL;
1382
1383        return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1384                              a.offset >> PAGE_SHIFT);
1385}
1386#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1387
1388/*
1389 * Some shared mappigns will want the pages marked read-only
1390 * to track write events. If so, we'll downgrade vm_page_prot
1391 * to the private version (using protection_map[] without the
1392 * VM_SHARED bit).
1393 */
1394int vma_wants_writenotify(struct vm_area_struct *vma)
1395{
1396        vm_flags_t vm_flags = vma->vm_flags;
1397
1398        /* If it was private or non-writable, the write bit is already clear */
1399        if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1400                return 0;
1401
1402        /* The backer wishes to know when pages are first written to? */
1403        if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1404                return 1;
1405
1406        /* The open routine did something to the protections already? */
1407        if (pgprot_val(vma->vm_page_prot) !=
1408            pgprot_val(vm_get_page_prot(vm_flags)))
1409                return 0;
1410
1411        /* Specialty mapping? */
1412        if (vm_flags & VM_PFNMAP)
1413                return 0;
1414
1415        /* Can the mapping track the dirty pages? */
1416        return vma->vm_file && vma->vm_file->f_mapping &&
1417                mapping_cap_account_dirty(vma->vm_file->f_mapping);
1418}
1419
1420/*
1421 * We account for memory if it's a private writeable mapping,
1422 * not hugepages and VM_NORESERVE wasn't set.
1423 */
1424static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1425{
1426        /*
1427         * hugetlb has its own accounting separate from the core VM
1428         * VM_HUGETLB may not be set yet so we cannot check for that flag.
1429         */
1430        if (file && is_file_hugepages(file))
1431                return 0;
1432
1433        return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1434}
1435
1436unsigned long mmap_region(struct file *file, unsigned long addr,
1437                unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
1438{
1439        struct mm_struct *mm = current->mm;
1440        struct vm_area_struct *vma, *prev;
1441        int correct_wcount = 0;
1442        int error;
1443        struct rb_node **rb_link, *rb_parent;
1444        unsigned long charged = 0;
1445        struct inode *inode =  file ? file_inode(file) : NULL;
1446
1447        /* Clear old maps */
1448        error = -ENOMEM;
1449munmap_back:
1450        if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
1451                if (do_munmap(mm, addr, len))
1452                        return -ENOMEM;
1453                goto munmap_back;
1454        }
1455
1456        /* Check against address space limit. */
1457        if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1458                return -ENOMEM;
1459
1460        /*
1461         * Private writable mapping: check memory availability
1462         */
1463        if (accountable_mapping(file, vm_flags)) {
1464                charged = len >> PAGE_SHIFT;
1465                if (security_vm_enough_memory_mm(mm, charged))
1466                        return -ENOMEM;
1467                vm_flags |= VM_ACCOUNT;
1468        }
1469
1470        /*
1471         * Can we just expand an old mapping?
1472         */
1473        vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1474        if (vma)
1475                goto out;
1476
1477        /*
1478         * Determine the object being mapped and call the appropriate
1479         * specific mapper. the address has already been validated, but
1480         * not unmapped, but the maps are removed from the list.
1481         */
1482        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1483        if (!vma) {
1484                error = -ENOMEM;
1485                goto unacct_error;
1486        }
1487
1488        vma->vm_mm = mm;
1489        vma->vm_start = addr;
1490        vma->vm_end = addr + len;
1491        vma->vm_flags = vm_flags;
1492        vma->vm_page_prot = vm_get_page_prot(vm_flags);
1493        vma->vm_pgoff = pgoff;
1494        INIT_LIST_HEAD(&vma->anon_vma_chain);
1495
1496        error = -EINVAL;        /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1497
1498        if (file) {
1499                if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1500                        goto free_vma;
1501                if (vm_flags & VM_DENYWRITE) {
1502                        error = deny_write_access(file);
1503                        if (error)
1504                                goto free_vma;
1505                        correct_wcount = 1;
1506                }
1507                vma->vm_file = get_file(file);
1508                error = file->f_op->mmap(file, vma);
1509                if (error)
1510                        goto unmap_and_free_vma;
1511
1512                /* Can addr have changed??
1513                 *
1514                 * Answer: Yes, several device drivers can do it in their
1515                 *         f_op->mmap method. -DaveM
1516                 * Bug: If addr is changed, prev, rb_link, rb_parent should
1517                 *      be updated for vma_link()
1518                 */
1519                WARN_ON_ONCE(addr != vma->vm_start);
1520
1521                addr = vma->vm_start;
1522                pgoff = vma->vm_pgoff;
1523                vm_flags = vma->vm_flags;
1524        } else if (vm_flags & VM_SHARED) {
1525                if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1526                        goto free_vma;
1527                error = shmem_zero_setup(vma);
1528                if (error)
1529                        goto free_vma;
1530        }
1531
1532        if (vma_wants_writenotify(vma)) {
1533                pgprot_t pprot = vma->vm_page_prot;
1534
1535                /* Can vma->vm_page_prot have changed??
1536                 *
1537                 * Answer: Yes, drivers may have changed it in their
1538                 *         f_op->mmap method.
1539                 *
1540                 * Ensures that vmas marked as uncached stay that way.
1541                 */
1542                vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1543                if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1544                        vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1545        }
1546
1547        vma_link(mm, vma, prev, rb_link, rb_parent);
1548        file = vma->vm_file;
1549
1550        /* Once vma denies write, undo our temporary denial count */
1551        if (correct_wcount)
1552                atomic_inc(&inode->i_writecount);
1553out:
1554        perf_event_mmap(vma);
1555
1556        vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1557        if (vm_flags & VM_LOCKED) {
1558                if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1559                                        vma == get_gate_vma(current->mm)))
1560                        mm->locked_vm += (len >> PAGE_SHIFT);
1561                else
1562                        vma->vm_flags &= ~VM_LOCKED;
1563        }
1564
1565        if (file)
1566                uprobe_mmap(vma);
1567
1568        return addr;
1569
1570unmap_and_free_vma:
1571        if (correct_wcount)
1572                atomic_inc(&inode->i_writecount);
1573        vma->vm_file = NULL;
1574        fput(file);
1575
1576        /* Undo any partial mapping done by a device driver. */
1577        unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1578        charged = 0;
1579free_vma:
1580        kmem_cache_free(vm_area_cachep, vma);
1581unacct_error:
1582        if (charged)
1583                vm_unacct_memory(charged);
1584        return error;
1585}
1586
1587unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1588{
1589        /*
1590         * We implement the search by looking for an rbtree node that
1591         * immediately follows a suitable gap. That is,
1592         * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1593         * - gap_end   = vma->vm_start        >= info->low_limit  + length;
1594         * - gap_end - gap_start >= length
1595         */
1596
1597        struct mm_struct *mm = current->mm;
1598        struct vm_area_struct *vma;
1599        unsigned long length, low_limit, high_limit, gap_start, gap_end;
1600
1601        /* Adjust search length to account for worst case alignment overhead */
1602        length = info->length + info->align_mask;
1603        if (length < info->length)
1604                return -ENOMEM;
1605
1606        /* Adjust search limits by the desired length */
1607        if (info->high_limit < length)
1608                return -ENOMEM;
1609        high_limit = info->high_limit - length;
1610
1611        if (info->low_limit > high_limit)
1612                return -ENOMEM;
1613        low_limit = info->low_limit + length;
1614
1615        /* Check if rbtree root looks promising */
1616        if (RB_EMPTY_ROOT(&mm->mm_rb))
1617                goto check_highest;
1618        vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1619        if (vma->rb_subtree_gap < length)
1620                goto check_highest;
1621
1622        while (true) {
1623                /* Visit left subtree if it looks promising */
1624                gap_end = vma->vm_start;
1625                if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1626                        struct vm_area_struct *left =
1627                                rb_entry(vma->vm_rb.rb_left,
1628                                         struct vm_area_struct, vm_rb);
1629                        if (left->rb_subtree_gap >= length) {
1630                                vma = left;
1631                                continue;
1632                        }
1633                }
1634
1635                gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1636check_current:
1637                /* Check if current node has a suitable gap */
1638                if (gap_start > high_limit)
1639                        return -ENOMEM;
1640                if (gap_end >= low_limit && gap_end - gap_start >= length)
1641                        goto found;
1642
1643                /* Visit right subtree if it looks promising */
1644                if (vma->vm_rb.rb_right) {
1645                        struct vm_area_struct *right =
1646                                rb_entry(vma->vm_rb.rb_right,
1647                                         struct vm_area_struct, vm_rb);
1648                        if (right->rb_subtree_gap >= length) {
1649                                vma = right;
1650                                continue;
1651                        }
1652                }
1653
1654                /* Go back up the rbtree to find next candidate node */
1655                while (true) {
1656                        struct rb_node *prev = &vma->vm_rb;
1657                        if (!rb_parent(prev))
1658                                goto check_highest;
1659                        vma = rb_entry(rb_parent(prev),
1660                                       struct vm_area_struct, vm_rb);
1661                        if (prev == vma->vm_rb.rb_left) {
1662                                gap_start = vma->vm_prev->vm_end;
1663                                gap_end = vma->vm_start;
1664                                goto check_current;
1665                        }
1666                }
1667        }
1668
1669check_highest:
1670        /* Check highest gap, which does not precede any rbtree node */
1671        gap_start = mm->highest_vm_end;
1672        gap_end = ULONG_MAX;  /* Only for VM_BUG_ON below */
1673        if (gap_start > high_limit)
1674                return -ENOMEM;
1675
1676found:
1677        /* We found a suitable gap. Clip it with the original low_limit. */
1678        if (gap_start < info->low_limit)
1679                gap_start = info->low_limit;
1680
1681        /* Adjust gap address to the desired alignment */
1682        gap_start += (info->align_offset - gap_start) & info->align_mask;
1683
1684        VM_BUG_ON(gap_start + info->length > info->high_limit);
1685        VM_BUG_ON(gap_start + info->length > gap_end);
1686        return gap_start;
1687}
1688
1689unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1690{
1691        struct mm_struct *mm = current->mm;
1692        struct vm_area_struct *vma;
1693        unsigned long length, low_limit, high_limit, gap_start, gap_end;
1694
1695        /* Adjust search length to account for worst case alignment overhead */
1696        length = info->length + info->align_mask;
1697        if (length < info->length)
1698                return -ENOMEM;
1699
1700        /*
1701         * Adjust search limits by the desired length.
1702         * See implementation comment at top of unmapped_area().
1703         */
1704        gap_end = info->high_limit;
1705        if (gap_end < length)
1706                return -ENOMEM;
1707        high_limit = gap_end - length;
1708
1709        if (info->low_limit > high_limit)
1710                return -ENOMEM;
1711        low_limit = info->low_limit + length;
1712
1713        /* Check highest gap, which does not precede any rbtree node */
1714        gap_start = mm->highest_vm_end;
1715        if (gap_start <= high_limit)
1716                goto found_highest;
1717
1718        /* Check if rbtree root looks promising */
1719        if (RB_EMPTY_ROOT(&mm->mm_rb))
1720                return -ENOMEM;
1721        vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1722        if (vma->rb_subtree_gap < length)
1723                return -ENOMEM;
1724
1725        while (true) {
1726                /* Visit right subtree if it looks promising */
1727                gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1728                if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1729                        struct vm_area_struct *right =
1730                                rb_entry(vma->vm_rb.rb_right,
1731                                         struct vm_area_struct, vm_rb);
1732                        if (right->rb_subtree_gap >= length) {
1733                                vma = right;
1734                                continue;
1735                        }
1736                }
1737
1738check_current:
1739                /* Check if current node has a suitable gap */
1740                gap_end = vma->vm_start;
1741                if (gap_end < low_limit)
1742                        return -ENOMEM;
1743                if (gap_start <= high_limit && gap_end - gap_start >= length)
1744                        goto found;
1745
1746                /* Visit left subtree if it looks promising */
1747                if (vma->vm_rb.rb_left) {
1748                        struct vm_area_struct *left =
1749                                rb_entry(vma->vm_rb.rb_left,
1750                                         struct vm_area_struct, vm_rb);
1751                        if (left->rb_subtree_gap >= length) {
1752                                vma = left;
1753                                continue;
1754                        }
1755                }
1756
1757                /* Go back up the rbtree to find next candidate node */
1758                while (true) {
1759                        struct rb_node *prev = &vma->vm_rb;
1760                        if (!rb_parent(prev))
1761                                return -ENOMEM;
1762                        vma = rb_entry(rb_parent(prev),
1763                                       struct vm_area_struct, vm_rb);
1764                        if (prev == vma->vm_rb.rb_right) {
1765                                gap_start = vma->vm_prev ?
1766                                        vma->vm_prev->vm_end : 0;
1767                                goto check_current;
1768                        }
1769                }
1770        }
1771
1772found:
1773        /* We found a suitable gap. Clip it with the original high_limit. */
1774        if (gap_end > info->high_limit)
1775                gap_end = info->high_limit;
1776
1777found_highest:
1778        /* Compute highest gap address at the desired alignment */
1779        gap_end -= info->length;
1780        gap_end -= (gap_end - info->align_offset) & info->align_mask;
1781
1782        VM_BUG_ON(gap_end < info->low_limit);
1783        VM_BUG_ON(gap_end < gap_start);
1784        return gap_end;
1785}
1786
1787/* Get an address range which is currently unmapped.
1788 * For shmat() with addr=0.
1789 *
1790 * Ugly calling convention alert:
1791 * Return value with the low bits set means error value,
1792 * ie
1793 *      if (ret & ~PAGE_MASK)
1794 *              error = ret;
1795 *
1796 * This function "knows" that -ENOMEM has the bits set.
1797 */
1798#ifndef HAVE_ARCH_UNMAPPED_AREA
1799unsigned long
1800arch_get_unmapped_area(struct file *filp, unsigned long addr,
1801                unsigned long len, unsigned long pgoff, unsigned long flags)
1802{
1803        struct mm_struct *mm = current->mm;
1804        struct vm_area_struct *vma;
1805        struct vm_unmapped_area_info info;
1806
1807        if (len > TASK_SIZE)
1808                return -ENOMEM;
1809
1810        if (flags & MAP_FIXED)
1811                return addr;
1812
1813        if (addr) {
1814                addr = PAGE_ALIGN(addr);
1815                vma = find_vma(mm, addr);
1816                if (TASK_SIZE - len >= addr &&
1817                    (!vma || addr + len <= vma->vm_start))
1818                        return addr;
1819        }
1820
1821        info.flags = 0;
1822        info.length = len;
1823        info.low_limit = TASK_UNMAPPED_BASE;
1824        info.high_limit = TASK_SIZE;
1825        info.align_mask = 0;
1826        return vm_unmapped_area(&info);
1827}
1828#endif  
1829
1830void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1831{
1832        /*
1833         * Is this a new hole at the lowest possible address?
1834         */
1835        if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1836                mm->free_area_cache = addr;
1837}
1838
1839/*
1840 * This mmap-allocator allocates new areas top-down from below the
1841 * stack's low limit (the base):
1842 */
1843#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1844unsigned long
1845arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1846                          const unsigned long len, const unsigned long pgoff,
1847                          const unsigned long flags)
1848{
1849        struct vm_area_struct *vma;
1850        struct mm_struct *mm = current->mm;
1851        unsigned long addr = addr0;
1852        struct vm_unmapped_area_info info;
1853
1854        /* requested length too big for entire address space */
1855        if (len > TASK_SIZE)
1856                return -ENOMEM;
1857
1858        if (flags & MAP_FIXED)
1859                return addr;
1860
1861        /* requesting a specific address */
1862        if (addr) {
1863                addr = PAGE_ALIGN(addr);
1864                vma = find_vma(mm, addr);
1865                if (TASK_SIZE - len >= addr &&
1866                                (!vma || addr + len <= vma->vm_start))
1867                        return addr;
1868        }
1869
1870        info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1871        info.length = len;
1872        info.low_limit = PAGE_SIZE;
1873        info.high_limit = mm->mmap_base;
1874        info.align_mask = 0;
1875        addr = vm_unmapped_area(&info);
1876
1877        /*
1878         * A failed mmap() very likely causes application failure,
1879         * so fall back to the bottom-up function here. This scenario
1880         * can happen with large stack limits and large mmap()
1881         * allocations.
1882         */
1883        if (addr & ~PAGE_MASK) {
1884                VM_BUG_ON(addr != -ENOMEM);
1885                info.flags = 0;
1886                info.low_limit = TASK_UNMAPPED_BASE;
1887                info.high_limit = TASK_SIZE;
1888                addr = vm_unmapped_area(&info);
1889        }
1890
1891        return addr;
1892}
1893#endif
1894
1895void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1896{
1897        /*
1898         * Is this a new hole at the highest possible address?
1899         */
1900        if (addr > mm->free_area_cache)
1901                mm->free_area_cache = addr;
1902
1903        /* dont allow allocations above current base */
1904        if (mm->free_area_cache > mm->mmap_base)
1905                mm->free_area_cache = mm->mmap_base;
1906}
1907
1908unsigned long
1909get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1910                unsigned long pgoff, unsigned long flags)
1911{
1912        unsigned long (*get_area)(struct file *, unsigned long,
1913                                  unsigned long, unsigned long, unsigned long);
1914
1915        unsigned long error = arch_mmap_check(addr, len, flags);
1916        if (error)
1917                return error;
1918
1919        /* Careful about overflows.. */
1920        if (len > TASK_SIZE)
1921                return -ENOMEM;
1922
1923        get_area = current->mm->get_unmapped_area;
1924        if (file && file->f_op && file->f_op->get_unmapped_area)
1925                get_area = file->f_op->get_unmapped_area;
1926        addr = get_area(file, addr, len, pgoff, flags);
1927        if (IS_ERR_VALUE(addr))
1928                return addr;
1929
1930        if (addr > TASK_SIZE - len)
1931                return -ENOMEM;
1932        if (addr & ~PAGE_MASK)
1933                return -EINVAL;
1934
1935        addr = arch_rebalance_pgtables(addr, len);
1936        error = security_mmap_addr(addr);
1937        return error ? error : addr;
1938}
1939
1940EXPORT_SYMBOL(get_unmapped_area);
1941
1942/* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
1943struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1944{
1945        struct vm_area_struct *vma = NULL;
1946
1947        if (WARN_ON_ONCE(!mm))          /* Remove this in linux-3.6 */
1948                return NULL;
1949
1950        /* Check the cache first. */
1951        /* (Cache hit rate is typically around 35%.) */
1952        vma = ACCESS_ONCE(mm->mmap_cache);
1953        if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1954                struct rb_node *rb_node;
1955
1956                rb_node = mm->mm_rb.rb_node;
1957                vma = NULL;
1958
1959                while (rb_node) {
1960                        struct vm_area_struct *vma_tmp;
1961
1962                        vma_tmp = rb_entry(rb_node,
1963                                           struct vm_area_struct, vm_rb);
1964
1965                        if (vma_tmp->vm_end > addr) {
1966                                vma = vma_tmp;
1967                                if (vma_tmp->vm_start <= addr)
1968                                        break;
1969                                rb_node = rb_node->rb_left;
1970                        } else
1971                                rb_node = rb_node->rb_right;
1972                }
1973                if (vma)
1974                        mm->mmap_cache = vma;
1975        }
1976        return vma;
1977}
1978
1979EXPORT_SYMBOL(find_vma);
1980
1981/*
1982 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1983 */
1984struct vm_area_struct *
1985find_vma_prev(struct mm_struct *mm, unsigned long addr,
1986                        struct vm_area_struct **pprev)
1987{
1988        struct vm_area_struct *vma;
1989
1990        vma = find_vma(mm, addr);
1991        if (vma) {
1992                *pprev = vma->vm_prev;
1993        } else {
1994                struct rb_node *rb_node = mm->mm_rb.rb_node;
1995                *pprev = NULL;
1996                while (rb_node) {
1997                        *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1998                        rb_node = rb_node->rb_right;
1999                }
2000        }
2001        return vma;
2002}
2003
2004/*
2005 * Verify that the stack growth is acceptable and
2006 * update accounting. This is shared with both the
2007 * grow-up and grow-down cases.
2008 */
2009static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
2010{
2011        struct mm_struct *mm = vma->vm_mm;
2012        struct rlimit *rlim = current->signal->rlim;
2013        unsigned long new_start;
2014
2015        /* address space limit tests */
2016        if (!may_expand_vm(mm, grow))
2017                return -ENOMEM;
2018
2019        /* Stack limit test */
2020        if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
2021                return -ENOMEM;
2022
2023        /* mlock limit tests */
2024        if (vma->vm_flags & VM_LOCKED) {
2025                unsigned long locked;
2026                unsigned long limit;
2027                locked = mm->locked_vm + grow;
2028                limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2029                limit >>= PAGE_SHIFT;
2030                if (locked > limit && !capable(CAP_IPC_LOCK))
2031                        return -ENOMEM;
2032        }
2033
2034        /* Check to ensure the stack will not grow into a hugetlb-only region */
2035        new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2036                        vma->vm_end - size;
2037        if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2038                return -EFAULT;
2039
2040        /*
2041         * Overcommit..  This must be the final test, as it will
2042         * update security statistics.
2043         */
2044        if (security_vm_enough_memory_mm(mm, grow))
2045                return -ENOMEM;
2046
2047        /* Ok, everything looks good - let it rip */
2048        if (vma->vm_flags & VM_LOCKED)
2049                mm->locked_vm += grow;
2050        vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
2051        return 0;
2052}
2053
2054#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2055/*
2056 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2057 * vma is the last one with address > vma->vm_end.  Have to extend vma.
2058 */
2059int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2060{
2061        int error;
2062
2063        if (!(vma->vm_flags & VM_GROWSUP))
2064                return -EFAULT;
2065
2066        /*
2067         * We must make sure the anon_vma is allocated
2068         * so that the anon_vma locking is not a noop.
2069         */
2070        if (unlikely(anon_vma_prepare(vma)))
2071                return -ENOMEM;
2072        vma_lock_anon_vma(vma);
2073
2074        /*
2075         * vma->vm_start/vm_end cannot change under us because the caller
2076         * is required to hold the mmap_sem in read mode.  We need the
2077         * anon_vma lock to serialize against concurrent expand_stacks.
2078         * Also guard against wrapping around to address 0.
2079         */
2080        if (address < PAGE_ALIGN(address+4))
2081                address = PAGE_ALIGN(address+4);
2082        else {
2083                vma_unlock_anon_vma(vma);
2084                return -ENOMEM;
2085        }
2086        error = 0;
2087
2088        /* Somebody else might have raced and expanded it already */
2089        if (address > vma->vm_end) {
2090                unsigned long size, grow;
2091
2092                size = address - vma->vm_start;
2093                grow = (address - vma->vm_end) >> PAGE_SHIFT;
2094
2095                error = -ENOMEM;
2096                if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2097                        error = acct_stack_growth(vma, size, grow);
2098                        if (!error) {
2099                                /*
2100                                 * vma_gap_update() doesn't support concurrent
2101                                 * updates, but we only hold a shared mmap_sem
2102                                 * lock here, so we need to protect against
2103                                 * concurrent vma expansions.
2104                                 * vma_lock_anon_vma() doesn't help here, as
2105                                 * we don't guarantee that all growable vmas
2106                                 * in a mm share the same root anon vma.
2107                                 * So, we reuse mm->page_table_lock to guard
2108                                 * against concurrent vma expansions.
2109                                 */
2110                                spin_lock(&vma->vm_mm->page_table_lock);
2111                                anon_vma_interval_tree_pre_update_vma(vma);
2112                                vma->vm_end = address;
2113                                anon_vma_interval_tree_post_update_vma(vma);
2114                                if (vma->vm_next)
2115                                        vma_gap_update(vma->vm_next);
2116                                else
2117                                        vma->vm_mm->highest_vm_end = address;
2118                                spin_unlock(&vma->vm_mm->page_table_lock);
2119
2120                                perf_event_mmap(vma);
2121                        }
2122                }
2123        }
2124        vma_unlock_anon_vma(vma);
2125        khugepaged_enter_vma_merge(vma);
2126        validate_mm(vma->vm_mm);
2127        return error;
2128}
2129#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2130
2131/*
2132 * vma is the first one with address < vma->vm_start.  Have to extend vma.
2133 */
2134int expand_downwards(struct vm_area_struct *vma,
2135                                   unsigned long address)
2136{
2137        int error;
2138
2139        /*
2140         * We must make sure the anon_vma is allocated
2141         * so that the anon_vma locking is not a noop.
2142         */
2143        if (unlikely(anon_vma_prepare(vma)))
2144                return -ENOMEM;
2145
2146        address &= PAGE_MASK;
2147        error = security_mmap_addr(address);
2148        if (error)
2149                return error;
2150
2151        vma_lock_anon_vma(vma);
2152
2153        /*
2154         * vma->vm_start/vm_end cannot change under us because the caller
2155         * is required to hold the mmap_sem in read mode.  We need the
2156         * anon_vma lock to serialize against concurrent expand_stacks.
2157         */
2158
2159        /* Somebody else might have raced and expanded it already */
2160        if (address < vma->vm_start) {
2161                unsigned long size, grow;
2162
2163                size = vma->vm_end - address;
2164                grow = (vma->vm_start - address) >> PAGE_SHIFT;
2165
2166                error = -ENOMEM;
2167                if (grow <= vma->vm_pgoff) {
2168                        error = acct_stack_growth(vma, size, grow);
2169                        if (!error) {
2170                                /*
2171                                 * vma_gap_update() doesn't support concurrent
2172                                 * updates, but we only hold a shared mmap_sem
2173                                 * lock here, so we need to protect against
2174                                 * concurrent vma expansions.
2175                                 * vma_lock_anon_vma() doesn't help here, as
2176                                 * we don't guarantee that all growable vmas
2177                                 * in a mm share the same root anon vma.
2178                                 * So, we reuse mm->page_table_lock to guard
2179                                 * against concurrent vma expansions.
2180                                 */
2181                                spin_lock(&vma->vm_mm->page_table_lock);
2182                                anon_vma_interval_tree_pre_update_vma(vma);
2183                                vma->vm_start = address;
2184                                vma->vm_pgoff -= grow;
2185                                anon_vma_interval_tree_post_update_vma(vma);
2186                                vma_gap_update(vma);
2187                                spin_unlock(&vma->vm_mm->page_table_lock);
2188
2189                                perf_event_mmap(vma);
2190                        }
2191                }
2192        }
2193        vma_unlock_anon_vma(vma);
2194        khugepaged_enter_vma_merge(vma);
2195        validate_mm(vma->vm_mm);
2196        return error;
2197}
2198
2199/*
2200 * Note how expand_stack() refuses to expand the stack all the way to
2201 * abut the next virtual mapping, *unless* that mapping itself is also
2202 * a stack mapping. We want to leave room for a guard page, after all
2203 * (the guard page itself is not added here, that is done by the
2204 * actual page faulting logic)
2205 *
2206 * This matches the behavior of the guard page logic (see mm/memory.c:
2207 * check_stack_guard_page()), which only allows the guard page to be
2208 * removed under these circumstances.
2209 */
2210#ifdef CONFIG_STACK_GROWSUP
2211int expand_stack(struct vm_area_struct *vma, unsigned long address)
2212{
2213        struct vm_area_struct *next;
2214
2215        address &= PAGE_MASK;
2216        next = vma->vm_next;
2217        if (next && next->vm_start == address + PAGE_SIZE) {
2218                if (!(next->vm_flags & VM_GROWSUP))
2219                        return -ENOMEM;
2220        }
2221        return expand_upwards(vma, address);
2222}
2223
2224struct vm_area_struct *
2225find_extend_vma(struct mm_struct *mm, unsigned long addr)
2226{
2227        struct vm_area_struct *vma, *prev;
2228
2229        addr &= PAGE_MASK;
2230        vma = find_vma_prev(mm, addr, &prev);
2231        if (vma && (vma->vm_start <= addr))
2232                return vma;
2233        if (!prev || expand_stack(prev, addr))
2234                return NULL;
2235        if (prev->vm_flags & VM_LOCKED)
2236                __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
2237        return prev;
2238}
2239#else
2240int expand_stack(struct vm_area_struct *vma, unsigned long address)
2241{
2242        struct vm_area_struct *prev;
2243
2244        address &= PAGE_MASK;
2245        prev = vma->vm_prev;
2246        if (prev && prev->vm_end == address) {
2247                if (!(prev->vm_flags & VM_GROWSDOWN))
2248                        return -ENOMEM;
2249        }
2250        return expand_downwards(vma, address);
2251}
2252
2253struct vm_area_struct *
2254find_extend_vma(struct mm_struct * mm, unsigned long addr)
2255{
2256        struct vm_area_struct * vma;
2257        unsigned long start;
2258
2259        addr &= PAGE_MASK;
2260        vma = find_vma(mm,addr);
2261        if (!vma)
2262                return NULL;
2263        if (vma->vm_start <= addr)
2264                return vma;
2265        if (!(vma->vm_flags & VM_GROWSDOWN))
2266                return NULL;
2267        start = vma->vm_start;
2268        if (expand_stack(vma, addr))
2269                return NULL;
2270        if (vma->vm_flags & VM_LOCKED)
2271                __mlock_vma_pages_range(vma, addr, start, NULL);
2272        return vma;
2273}
2274#endif
2275
2276/*
2277 * Ok - we have the memory areas we should free on the vma list,
2278 * so release them, and do the vma updates.
2279 *
2280 * Called with the mm semaphore held.
2281 */
2282static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2283{
2284        unsigned long nr_accounted = 0;
2285
2286        /* Update high watermark before we lower total_vm */
2287        update_hiwater_vm(mm);
2288        do {
2289                long nrpages = vma_pages(vma);
2290
2291                if (vma->vm_flags & VM_ACCOUNT)
2292                        nr_accounted += nrpages;
2293                vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
2294                vma = remove_vma(vma);
2295        } while (vma);
2296        vm_unacct_memory(nr_accounted);
2297        validate_mm(mm);
2298}
2299
2300/*
2301 * Get rid of page table information in the indicated region.
2302 *
2303 * Called with the mm semaphore held.
2304 */
2305static void unmap_region(struct mm_struct *mm,
2306                struct vm_area_struct *vma, struct vm_area_struct *prev,
2307                unsigned long start, unsigned long end)
2308{
2309        struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
2310        struct mmu_gather tlb;
2311
2312        lru_add_drain();
2313        tlb_gather_mmu(&tlb, mm, 0);
2314        update_hiwater_rss(mm);
2315        unmap_vmas(&tlb, vma, start, end);
2316        free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2317                                 next ? next->vm_start : USER_PGTABLES_CEILING);
2318        tlb_finish_mmu(&tlb, start, end);
2319}
2320
2321/*
2322 * Create a list of vma's touched by the unmap, removing them from the mm's
2323 * vma list as we go..
2324 */
2325static void
2326detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2327        struct vm_area_struct *prev, unsigned long end)
2328{
2329        struct vm_area_struct **insertion_point;
2330        struct vm_area_struct *tail_vma = NULL;
2331        unsigned long addr;
2332
2333        insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2334        vma->vm_prev = NULL;
2335        do {
2336                vma_rb_erase(vma, &mm->mm_rb);
2337                mm->map_count--;
2338                tail_vma = vma;
2339                vma = vma->vm_next;
2340        } while (vma && vma->vm_start < end);
2341        *insertion_point = vma;
2342        if (vma) {
2343                vma->vm_prev = prev;
2344                vma_gap_update(vma);
2345        } else
2346                mm->highest_vm_end = prev ? prev->vm_end : 0;
2347        tail_vma->vm_next = NULL;
2348        if (mm->unmap_area == arch_unmap_area)
2349                addr = prev ? prev->vm_end : mm->mmap_base;
2350        else
2351                addr = vma ?  vma->vm_start : mm->mmap_base;
2352        mm->unmap_area(mm, addr);
2353        mm->mmap_cache = NULL;          /* Kill the cache. */
2354}
2355
2356/*
2357 * __split_vma() bypasses sysctl_max_map_count checking.  We use this on the
2358 * munmap path where it doesn't make sense to fail.
2359 */
2360static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
2361              unsigned long addr, int new_below)
2362{
2363        struct mempolicy *pol;
2364        struct vm_area_struct *new;
2365        int err = -ENOMEM;
2366
2367        if (is_vm_hugetlb_page(vma) && (addr &
2368                                        ~(huge_page_mask(hstate_vma(vma)))))
2369                return -EINVAL;
2370
2371        new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2372        if (!new)
2373                goto out_err;
2374
2375        /* most fields are the same, copy all, and then fixup */
2376        *new = *vma;
2377
2378        INIT_LIST_HEAD(&new->anon_vma_chain);
2379
2380        if (new_below)
2381                new->vm_end = addr;
2382        else {
2383                new->vm_start = addr;
2384                new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2385        }
2386
2387        pol = mpol_dup(vma_policy(vma));
2388        if (IS_ERR(pol)) {
2389                err = PTR_ERR(pol);
2390                goto out_free_vma;
2391        }
2392        vma_set_policy(new, pol);
2393
2394        if (anon_vma_clone(new, vma))
2395                goto out_free_mpol;
2396
2397        if (new->vm_file)
2398                get_file(new->vm_file);
2399
2400        if (new->vm_ops && new->vm_ops->open)
2401                new->vm_ops->open(new);
2402
2403        if (new_below)
2404                err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2405                        ((addr - new->vm_start) >> PAGE_SHIFT), new);
2406        else
2407                err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2408
2409        /* Success. */
2410        if (!err)
2411                return 0;
2412
2413        /* Clean everything up if vma_adjust failed. */
2414        if (new->vm_ops && new->vm_ops->close)
2415                new->vm_ops->close(new);
2416        if (new->vm_file)
2417                fput(new->vm_file);
2418        unlink_anon_vmas(new);
2419 out_free_mpol:
2420        mpol_put(pol);
2421 out_free_vma:
2422        kmem_cache_free(vm_area_cachep, new);
2423 out_err:
2424        return err;
2425}
2426
2427/*
2428 * Split a vma into two pieces at address 'addr', a new vma is allocated
2429 * either for the first part or the tail.
2430 */
2431int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2432              unsigned long addr, int new_below)
2433{
2434        if (mm->map_count >= sysctl_max_map_count)
2435                return -ENOMEM;
2436
2437        return __split_vma(mm, vma, addr, new_below);
2438}
2439
2440/* Munmap is split into 2 main parts -- this part which finds
2441 * what needs doing, and the areas themselves, which do the
2442 * work.  This now handles partial unmappings.
2443 * Jeremy Fitzhardinge <jeremy@goop.org>
2444 */
2445int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2446{
2447        unsigned long end;
2448        struct vm_area_struct *vma, *prev, *last;
2449
2450        if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2451                return -EINVAL;
2452
2453        if ((len = PAGE_ALIGN(len)) == 0)
2454                return -EINVAL;
2455
2456        /* Find the first overlapping VMA */
2457        vma = find_vma(mm, start);
2458        if (!vma)
2459                return 0;
2460        prev = vma->vm_prev;
2461        /* we have  start < vma->vm_end  */
2462
2463        /* if it doesn't overlap, we have nothing.. */
2464        end = start + len;
2465        if (vma->vm_start >= end)
2466                return 0;
2467
2468        /*
2469         * If we need to split any vma, do it now to save pain later.
2470         *
2471         * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2472         * unmapped vm_area_struct will remain in use: so lower split_vma
2473         * places tmp vma above, and higher split_vma places tmp vma below.
2474         */
2475        if (start > vma->vm_start) {
2476                int error;
2477
2478                /*
2479                 * Make sure that map_count on return from munmap() will
2480                 * not exceed its limit; but let map_count go just above
2481                 * its limit temporarily, to help free resources as expected.
2482                 */
2483                if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2484                        return -ENOMEM;
2485
2486                error = __split_vma(mm, vma, start, 0);
2487                if (error)
2488                        return error;
2489                prev = vma;
2490        }
2491
2492        /* Does it split the last one? */
2493        last = find_vma(mm, end);
2494        if (last && end > last->vm_start) {
2495                int error = __split_vma(mm, last, end, 1);
2496                if (error)
2497                        return error;
2498        }
2499        vma = prev? prev->vm_next: mm->mmap;
2500
2501        /*
2502         * unlock any mlock()ed ranges before detaching vmas
2503         */
2504        if (mm->locked_vm) {
2505                struct vm_area_struct *tmp = vma;
2506                while (tmp && tmp->vm_start < end) {
2507                        if (tmp->vm_flags & VM_LOCKED) {
2508                                mm->locked_vm -= vma_pages(tmp);
2509                                munlock_vma_pages_all(tmp);
2510                        }
2511                        tmp = tmp->vm_next;
2512                }
2513        }
2514
2515        /*
2516         * Remove the vma's, and unmap the actual pages
2517         */
2518        detach_vmas_to_be_unmapped(mm, vma, prev, end);
2519        unmap_region(mm, vma, prev, start, end);
2520
2521        /* Fix up all other VM information */
2522        remove_vma_list(mm, vma);
2523
2524        return 0;
2525}
2526
2527int vm_munmap(unsigned long start, size_t len)
2528{
2529        int ret;
2530        struct mm_struct *mm = current->mm;
2531
2532        down_write(&mm->mmap_sem);
2533        ret = do_munmap(mm, start, len);
2534        up_write(&mm->mmap_sem);
2535        return ret;
2536}
2537EXPORT_SYMBOL(vm_munmap);
2538
2539SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2540{
2541        profile_munmap(addr);
2542        return vm_munmap(addr, len);
2543}
2544
2545static inline void verify_mm_writelocked(struct mm_struct *mm)
2546{
2547#ifdef CONFIG_DEBUG_VM
2548        if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2549                WARN_ON(1);
2550                up_read(&mm->mmap_sem);
2551        }
2552#endif
2553}
2554
2555/*
2556 *  this is really a simplified "do_mmap".  it only handles
2557 *  anonymous maps.  eventually we may be able to do some
2558 *  brk-specific accounting here.
2559 */
2560static unsigned long do_brk(unsigned long addr, unsigned long len)
2561{
2562        struct mm_struct * mm = current->mm;
2563        struct vm_area_struct * vma, * prev;
2564        unsigned long flags;
2565        struct rb_node ** rb_link, * rb_parent;
2566        pgoff_t pgoff = addr >> PAGE_SHIFT;
2567        int error;
2568
2569        len = PAGE_ALIGN(len);
2570        if (!len)
2571                return addr;
2572
2573        flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2574
2575        error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2576        if (error & ~PAGE_MASK)
2577                return error;
2578
2579        /*
2580         * mlock MCL_FUTURE?
2581         */
2582        if (mm->def_flags & VM_LOCKED) {
2583                unsigned long locked, lock_limit;
2584                locked = len >> PAGE_SHIFT;
2585                locked += mm->locked_vm;
2586                lock_limit = rlimit(RLIMIT_MEMLOCK);
2587                lock_limit >>= PAGE_SHIFT;
2588                if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2589                        return -EAGAIN;
2590        }
2591
2592        /*
2593         * mm->mmap_sem is required to protect against another thread
2594         * changing the mappings in case we sleep.
2595         */
2596        verify_mm_writelocked(mm);
2597
2598        /*
2599         * Clear old maps.  this also does some error checking for us
2600         */
2601 munmap_back:
2602        if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
2603                if (do_munmap(mm, addr, len))
2604                        return -ENOMEM;
2605                goto munmap_back;
2606        }
2607
2608        /* Check against address space limits *after* clearing old maps... */
2609        if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2610                return -ENOMEM;
2611
2612        if (mm->map_count > sysctl_max_map_count)
2613                return -ENOMEM;
2614
2615        if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2616                return -ENOMEM;
2617
2618        /* Can we just expand an old private anonymous mapping? */
2619        vma = vma_merge(mm, prev, addr, addr + len, flags,
2620                                        NULL, NULL, pgoff, NULL);
2621        if (vma)
2622                goto out;
2623
2624        /*
2625         * create a vma struct for an anonymous mapping
2626         */
2627        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2628        if (!vma) {
2629                vm_unacct_memory(len >> PAGE_SHIFT);
2630                return -ENOMEM;
2631        }
2632
2633        INIT_LIST_HEAD(&vma->anon_vma_chain);
2634        vma->vm_mm = mm;
2635        vma->vm_start = addr;
2636        vma->vm_end = addr + len;
2637        vma->vm_pgoff = pgoff;
2638        vma->vm_flags = flags;
2639        vma->vm_page_prot = vm_get_page_prot(flags);
2640        vma_link(mm, vma, prev, rb_link, rb_parent);
2641out:
2642        perf_event_mmap(vma);
2643        mm->total_vm += len >> PAGE_SHIFT;
2644        if (flags & VM_LOCKED)
2645                mm->locked_vm += (len >> PAGE_SHIFT);
2646        return addr;
2647}
2648
2649unsigned long vm_brk(unsigned long addr, unsigned long len)
2650{
2651        struct mm_struct *mm = current->mm;
2652        unsigned long ret;
2653        bool populate;
2654
2655        down_write(&mm->mmap_sem);
2656        ret = do_brk(addr, len);
2657        populate = ((mm->def_flags & VM_LOCKED) != 0);
2658        up_write(&mm->mmap_sem);
2659        if (populate)
2660                mm_populate(addr, len);
2661        return ret;
2662}
2663EXPORT_SYMBOL(vm_brk);
2664
2665/* Release all mmaps. */
2666void exit_mmap(struct mm_struct *mm)
2667{
2668        struct mmu_gather tlb;
2669        struct vm_area_struct *vma;
2670        unsigned long nr_accounted = 0;
2671
2672        /* mm's last user has gone, and its about to be pulled down */
2673        mmu_notifier_release(mm);
2674
2675        if (mm->locked_vm) {
2676                vma = mm->mmap;
2677                while (vma) {
2678                        if (vma->vm_flags & VM_LOCKED)
2679                                munlock_vma_pages_all(vma);
2680                        vma = vma->vm_next;
2681                }
2682        }
2683
2684        arch_exit_mmap(mm);
2685
2686        vma = mm->mmap;
2687        if (!vma)       /* Can happen if dup_mmap() received an OOM */
2688                return;
2689
2690        lru_add_drain();
2691        flush_cache_mm(mm);
2692        tlb_gather_mmu(&tlb, mm, 1);
2693        /* update_hiwater_rss(mm) here? but nobody should be looking */
2694        /* Use -1 here to ensure all VMAs in the mm are unmapped */
2695        unmap_vmas(&tlb, vma, 0, -1);
2696
2697        free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
2698        tlb_finish_mmu(&tlb, 0, -1);
2699
2700        /*
2701         * Walk the list again, actually closing and freeing it,
2702         * with preemption enabled, without holding any MM locks.
2703         */
2704        while (vma) {
2705                if (vma->vm_flags & VM_ACCOUNT)
2706                        nr_accounted += vma_pages(vma);
2707                vma = remove_vma(vma);
2708        }
2709        vm_unacct_memory(nr_accounted);
2710
2711        WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2712}
2713
2714/* Insert vm structure into process list sorted by address
2715 * and into the inode's i_mmap tree.  If vm_file is non-NULL
2716 * then i_mmap_mutex is taken here.
2717 */
2718int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2719{
2720        struct vm_area_struct *prev;
2721        struct rb_node **rb_link, *rb_parent;
2722
2723        /*
2724         * The vm_pgoff of a purely anonymous vma should be irrelevant
2725         * until its first write fault, when page's anon_vma and index
2726         * are set.  But now set the vm_pgoff it will almost certainly
2727         * end up with (unless mremap moves it elsewhere before that
2728         * first wfault), so /proc/pid/maps tells a consistent story.
2729         *
2730         * By setting it to reflect the virtual start address of the
2731         * vma, merges and splits can happen in a seamless way, just
2732         * using the existing file pgoff checks and manipulations.
2733         * Similarly in do_mmap_pgoff and in do_brk.
2734         */
2735        if (!vma->vm_file) {
2736                BUG_ON(vma->anon_vma);
2737                vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2738        }
2739        if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2740                           &prev, &rb_link, &rb_parent))
2741                return -ENOMEM;
2742        if ((vma->vm_flags & VM_ACCOUNT) &&
2743             security_vm_enough_memory_mm(mm, vma_pages(vma)))
2744                return -ENOMEM;
2745
2746        vma_link(mm, vma, prev, rb_link, rb_parent);
2747        return 0;
2748}
2749
2750/*
2751 * Copy the vma structure to a new location in the same mm,
2752 * prior to moving page table entries, to effect an mremap move.
2753 */
2754struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2755        unsigned long addr, unsigned long len, pgoff_t pgoff,
2756        bool *need_rmap_locks)
2757{
2758        struct vm_area_struct *vma = *vmap;
2759        unsigned long vma_start = vma->vm_start;
2760        struct mm_struct *mm = vma->vm_mm;
2761        struct vm_area_struct *new_vma, *prev;
2762        struct rb_node **rb_link, *rb_parent;
2763        struct mempolicy *pol;
2764        bool faulted_in_anon_vma = true;
2765
2766        /*
2767         * If anonymous vma has not yet been faulted, update new pgoff
2768         * to match new location, to increase its chance of merging.
2769         */
2770        if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2771                pgoff = addr >> PAGE_SHIFT;
2772                faulted_in_anon_vma = false;
2773        }
2774
2775        if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2776                return NULL;    /* should never get here */
2777        new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2778                        vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2779        if (new_vma) {
2780                /*
2781                 * Source vma may have been merged into new_vma
2782                 */
2783                if (unlikely(vma_start >= new_vma->vm_start &&
2784                             vma_start < new_vma->vm_end)) {
2785                        /*
2786                         * The only way we can get a vma_merge with
2787                         * self during an mremap is if the vma hasn't
2788                         * been faulted in yet and we were allowed to
2789                         * reset the dst vma->vm_pgoff to the
2790                         * destination address of the mremap to allow
2791                         * the merge to happen. mremap must change the
2792                         * vm_pgoff linearity between src and dst vmas
2793                         * (in turn preventing a vma_merge) to be
2794                         * safe. It is only safe to keep the vm_pgoff
2795                         * linear if there are no pages mapped yet.
2796                         */
2797                        VM_BUG_ON(faulted_in_anon_vma);
2798                        *vmap = vma = new_vma;
2799                }
2800                *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2801        } else {
2802                new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2803                if (new_vma) {
2804                        *new_vma = *vma;
2805                        new_vma->vm_start = addr;
2806                        new_vma->vm_end = addr + len;
2807                        new_vma->vm_pgoff = pgoff;
2808                        pol = mpol_dup(vma_policy(vma));
2809                        if (IS_ERR(pol))
2810                                goto out_free_vma;
2811                        vma_set_policy(new_vma, pol);
2812                        INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2813                        if (anon_vma_clone(new_vma, vma))
2814                                goto out_free_mempol;
2815                        if (new_vma->vm_file)
2816                                get_file(new_vma->vm_file);
2817                        if (new_vma->vm_ops && new_vma->vm_ops->open)
2818                                new_vma->vm_ops->open(new_vma);
2819                        vma_link(mm, new_vma, prev, rb_link, rb_parent);
2820                        *need_rmap_locks = false;
2821                }
2822        }
2823        return new_vma;
2824
2825 out_free_mempol:
2826        mpol_put(pol);
2827 out_free_vma:
2828        kmem_cache_free(vm_area_cachep, new_vma);
2829        return NULL;
2830}
2831
2832/*
2833 * Return true if the calling process may expand its vm space by the passed
2834 * number of pages
2835 */
2836int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2837{
2838        unsigned long cur = mm->total_vm;       /* pages */
2839        unsigned long lim;
2840
2841        lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2842
2843        if (cur + npages > lim)
2844                return 0;
2845        return 1;
2846}
2847
2848
2849static int special_mapping_fault(struct vm_area_struct *vma,
2850                                struct vm_fault *vmf)
2851{
2852        pgoff_t pgoff;
2853        struct page **pages;
2854
2855        /*
2856         * special mappings have no vm_file, and in that case, the mm
2857         * uses vm_pgoff internally. So we have to subtract it from here.
2858         * We are allowed to do this because we are the mm; do not copy
2859         * this code into drivers!
2860         */
2861        pgoff = vmf->pgoff - vma->vm_pgoff;
2862
2863        for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2864                pgoff--;
2865
2866        if (*pages) {
2867                struct page *page = *pages;
2868                get_page(page);
2869                vmf->page = page;
2870                return 0;
2871        }
2872
2873        return VM_FAULT_SIGBUS;
2874}
2875
2876/*
2877 * Having a close hook prevents vma merging regardless of flags.
2878 */
2879static void special_mapping_close(struct vm_area_struct *vma)
2880{
2881}
2882
2883static const struct vm_operations_struct special_mapping_vmops = {
2884        .close = special_mapping_close,
2885        .fault = special_mapping_fault,
2886};
2887
2888/*
2889 * Called with mm->mmap_sem held for writing.
2890 * Insert a new vma covering the given region, with the given flags.
2891 * Its pages are supplied by the given array of struct page *.
2892 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2893 * The region past the last page supplied will always produce SIGBUS.
2894 * The array pointer and the pages it points to are assumed to stay alive
2895 * for as long as this mapping might exist.
2896 */
2897int install_special_mapping(struct mm_struct *mm,
2898                            unsigned long addr, unsigned long len,
2899                            unsigned long vm_flags, struct page **pages)
2900{
2901        int ret;
2902        struct vm_area_struct *vma;
2903
2904        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2905        if (unlikely(vma == NULL))
2906                return -ENOMEM;
2907
2908        INIT_LIST_HEAD(&vma->anon_vma_chain);
2909        vma->vm_mm = mm;
2910        vma->vm_start = addr;
2911        vma->vm_end = addr + len;
2912
2913        vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2914        vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2915
2916        vma->vm_ops = &special_mapping_vmops;
2917        vma->vm_private_data = pages;
2918
2919        ret = insert_vm_struct(mm, vma);
2920        if (ret)
2921                goto out;
2922
2923        mm->total_vm += len >> PAGE_SHIFT;
2924
2925        perf_event_mmap(vma);
2926
2927        return 0;
2928
2929out:
2930        kmem_cache_free(vm_area_cachep, vma);
2931        return ret;
2932}
2933
2934static DEFINE_MUTEX(mm_all_locks_mutex);
2935
2936static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2937{
2938        if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
2939                /*
2940                 * The LSB of head.next can't change from under us
2941                 * because we hold the mm_all_locks_mutex.
2942                 */
2943                down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
2944                /*
2945                 * We can safely modify head.next after taking the
2946                 * anon_vma->root->rwsem. If some other vma in this mm shares
2947                 * the same anon_vma we won't take it again.
2948                 *
2949                 * No need of atomic instructions here, head.next
2950                 * can't change from under us thanks to the
2951                 * anon_vma->root->rwsem.
2952                 */
2953                if (__test_and_set_bit(0, (unsigned long *)
2954                                       &anon_vma->root->rb_root.rb_node))
2955                        BUG();
2956        }
2957}
2958
2959static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2960{
2961        if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2962                /*
2963                 * AS_MM_ALL_LOCKS can't change from under us because
2964                 * we hold the mm_all_locks_mutex.
2965                 *
2966                 * Operations on ->flags have to be atomic because
2967                 * even if AS_MM_ALL_LOCKS is stable thanks to the
2968                 * mm_all_locks_mutex, there may be other cpus
2969                 * changing other bitflags in parallel to us.
2970                 */
2971                if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2972                        BUG();
2973                mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2974        }
2975}
2976
2977/*
2978 * This operation locks against the VM for all pte/vma/mm related
2979 * operations that could ever happen on a certain mm. This includes
2980 * vmtruncate, try_to_unmap, and all page faults.
2981 *
2982 * The caller must take the mmap_sem in write mode before calling
2983 * mm_take_all_locks(). The caller isn't allowed to release the
2984 * mmap_sem until mm_drop_all_locks() returns.
2985 *
2986 * mmap_sem in write mode is required in order to block all operations
2987 * that could modify pagetables and free pages without need of
2988 * altering the vma layout (for example populate_range() with
2989 * nonlinear vmas). It's also needed in write mode to avoid new
2990 * anon_vmas to be associated with existing vmas.
2991 *
2992 * A single task can't take more than one mm_take_all_locks() in a row
2993 * or it would deadlock.
2994 *
2995 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
2996 * mapping->flags avoid to take the same lock twice, if more than one
2997 * vma in this mm is backed by the same anon_vma or address_space.
2998 *
2999 * We can take all the locks in random order because the VM code
3000 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3001 * takes more than one of them in a row. Secondly we're protected
3002 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3003 *
3004 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3005 * that may have to take thousand of locks.
3006 *
3007 * mm_take_all_locks() can fail if it's interrupted by signals.
3008 */
3009int mm_take_all_locks(struct mm_struct *mm)
3010{
3011        struct vm_area_struct *vma;
3012        struct anon_vma_chain *avc;
3013
3014        BUG_ON(down_read_trylock(&mm->mmap_sem));
3015
3016        mutex_lock(&mm_all_locks_mutex);
3017
3018        for (vma = mm->mmap; vma; vma = vma->vm_next) {
3019                if (signal_pending(current))
3020                        goto out_unlock;
3021                if (vma->vm_file && vma->vm_file->f_mapping)
3022                        vm_lock_mapping(mm, vma->vm_file->f_mapping);
3023        }
3024
3025        for (vma = mm->mmap; vma; vma = vma->vm_next) {
3026                if (signal_pending(current))
3027                        goto out_unlock;
3028                if (vma->anon_vma)
3029                        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3030                                vm_lock_anon_vma(mm, avc->anon_vma);
3031        }
3032
3033        return 0;
3034
3035out_unlock:
3036        mm_drop_all_locks(mm);
3037        return -EINTR;
3038}
3039
3040static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3041{
3042        if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3043                /*
3044                 * The LSB of head.next can't change to 0 from under
3045                 * us because we hold the mm_all_locks_mutex.
3046                 *
3047                 * We must however clear the bitflag before unlocking
3048                 * the vma so the users using the anon_vma->rb_root will
3049                 * never see our bitflag.
3050                 *
3051                 * No need of atomic instructions here, head.next
3052                 * can't change from under us until we release the
3053                 * anon_vma->root->rwsem.
3054                 */
3055                if (!__test_and_clear_bit(0, (unsigned long *)
3056                                          &anon_vma->root->rb_root.rb_node))
3057                        BUG();
3058                anon_vma_unlock_write(anon_vma);
3059        }
3060}
3061
3062static void vm_unlock_mapping(struct address_space *mapping)
3063{
3064        if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3065                /*
3066                 * AS_MM_ALL_LOCKS can't change to 0 from under us
3067                 * because we hold the mm_all_locks_mutex.
3068                 */
3069                mutex_unlock(&mapping->i_mmap_mutex);
3070                if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3071                                        &mapping->flags))
3072                        BUG();
3073        }
3074}
3075
3076/*
3077 * The mmap_sem cannot be released by the caller until
3078 * mm_drop_all_locks() returns.
3079 */
3080void mm_drop_all_locks(struct mm_struct *mm)
3081{
3082        struct vm_area_struct *vma;
3083        struct anon_vma_chain *avc;
3084
3085        BUG_ON(down_read_trylock(&mm->mmap_sem));
3086        BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3087
3088        for (vma = mm->mmap; vma; vma = vma->vm_next) {
3089                if (vma->anon_vma)
3090                        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3091                                vm_unlock_anon_vma(avc->anon_vma);
3092                if (vma->vm_file && vma->vm_file->f_mapping)
3093                        vm_unlock_mapping(vma->vm_file->f_mapping);
3094        }
3095
3096        mutex_unlock(&mm_all_locks_mutex);
3097}
3098
3099/*
3100 * initialise the VMA slab
3101 */
3102void __init mmap_init(void)
3103{
3104        int ret;
3105
3106        ret = percpu_counter_init(&vm_committed_as, 0);
3107        VM_BUG_ON(ret);
3108}
3109
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.