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/kernel.h>
  10#include <linux/slab.h>
  11#include <linux/backing-dev.h>
  12#include <linux/mm.h>
  13#include <linux/shm.h>
  14#include <linux/mman.h>
  15#include <linux/pagemap.h>
  16#include <linux/swap.h>
  17#include <linux/syscalls.h>
  18#include <linux/capability.h>
  19#include <linux/init.h>
  20#include <linux/file.h>
  21#include <linux/fs.h>
  22#include <linux/personality.h>
  23#include <linux/security.h>
  24#include <linux/hugetlb.h>
  25#include <linux/profile.h>
  26#include <linux/export.h>
  27#include <linux/mount.h>
  28#include <linux/mempolicy.h>
  29#include <linux/rmap.h>
  30#include <linux/mmu_notifier.h>
  31#include <linux/perf_event.h>
  32#include <linux/audit.h>
  33#include <linux/khugepaged.h>
  34#include <linux/uprobes.h>
  35#include <linux/rbtree_augmented.h>
  36#include <linux/sched/sysctl.h>
  37#include <linux/notifier.h>
  38#include <linux/memory.h>
  39
  40#include <asm/uaccess.h>
  41#include <asm/cacheflush.h>
  42#include <asm/tlb.h>
  43#include <asm/mmu_context.h>
  44
  45#include "internal.h"
  46
  47#ifndef arch_mmap_check
  48#define arch_mmap_check(addr, len, flags)       (0)
  49#endif
  50
  51#ifndef arch_rebalance_pgtables
  52#define arch_rebalance_pgtables(addr, len)              (addr)
  53#endif
  54
  55static void unmap_region(struct mm_struct *mm,
  56                struct vm_area_struct *vma, struct vm_area_struct *prev,
  57                unsigned long start, unsigned long end);
  58
  59/* description of effects of mapping type and prot in current implementation.
  60 * this is due to the limited x86 page protection hardware.  The expected
  61 * behavior is in parens:
  62 *
  63 * map_type     prot
  64 *              PROT_NONE       PROT_READ       PROT_WRITE      PROT_EXEC
  65 * MAP_SHARED   r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  66 *              w: (no) no      w: (no) no      w: (yes) yes    w: (no) no
  67 *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  68 *              
  69 * MAP_PRIVATE  r: (no) no      r: (yes) yes    r: (no) yes     r: (no) yes
  70 *              w: (no) no      w: (no) no      w: (copy) copy  w: (no) no
  71 *              x: (no) no      x: (no) yes     x: (no) yes     x: (yes) yes
  72 *
  73 */
  74pgprot_t protection_map[16] = {
  75        __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  76        __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  77};
  78
  79pgprot_t vm_get_page_prot(unsigned long vm_flags)
  80{
  81        return __pgprot(pgprot_val(protection_map[vm_flags &
  82                                (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
  83                        pgprot_val(arch_vm_get_page_prot(vm_flags)));
  84}
  85EXPORT_SYMBOL(vm_get_page_prot);
  86
  87int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;  /* heuristic overcommit */
  88int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
  89int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
  90unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
  91unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
  92/*
  93 * Make sure vm_committed_as in one cacheline and not cacheline shared with
  94 * other variables. It can be updated by several CPUs frequently.
  95 */
  96struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
  97
  98/*
  99 * The global memory commitment made in the system can be a metric
 100 * that can be used to drive ballooning decisions when Linux is hosted
 101 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
 102 * balancing memory across competing virtual machines that are hosted.
 103 * Several metrics drive this policy engine including the guest reported
 104 * memory commitment.
 105 */
 106unsigned long vm_memory_committed(void)
 107{
 108        return percpu_counter_read_positive(&vm_committed_as);
 109}
 110EXPORT_SYMBOL_GPL(vm_memory_committed);
 111
 112/*
 113 * Check that a process has enough memory to allocate a new virtual
 114 * mapping. 0 means there is enough memory for the allocation to
 115 * succeed and -ENOMEM implies there is not.
 116 *
 117 * We currently support three overcommit policies, which are set via the
 118 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 119 *
 120 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 121 * Additional code 2002 Jul 20 by Robert Love.
 122 *
 123 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 124 *
 125 * Note this is a helper function intended to be used by LSMs which
 126 * wish to use this logic.
 127 */
 128int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 129{
 130        unsigned long free, allowed, reserve;
 131
 132        vm_acct_memory(pages);
 133
 134        /*
 135         * Sometimes we want to use more memory than we have
 136         */
 137        if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 138                return 0;
 139
 140        if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 141                free = global_page_state(NR_FREE_PAGES);
 142                free += global_page_state(NR_FILE_PAGES);
 143
 144                /*
 145                 * shmem pages shouldn't be counted as free in this
 146                 * case, they can't be purged, only swapped out, and
 147                 * that won't affect the overall amount of available
 148                 * memory in the system.
 149                 */
 150                free -= global_page_state(NR_SHMEM);
 151
 152                free += get_nr_swap_pages();
 153
 154                /*
 155                 * Any slabs which are created with the
 156                 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 157                 * which are reclaimable, under pressure.  The dentry
 158                 * cache and most inode caches should fall into this
 159                 */
 160                free += global_page_state(NR_SLAB_RECLAIMABLE);
 161
 162                /*
 163                 * Leave reserved pages. The pages are not for anonymous pages.
 164                 */
 165                if (free <= totalreserve_pages)
 166                        goto error;
 167                else
 168                        free -= totalreserve_pages;
 169
 170                /*
 171                 * Reserve some for root
 172                 */
 173                if (!cap_sys_admin)
 174                        free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
 175
 176                if (free > pages)
 177                        return 0;
 178
 179                goto error;
 180        }
 181
 182        allowed = (totalram_pages - hugetlb_total_pages())
 183                * sysctl_overcommit_ratio / 100;
 184        /*
 185         * Reserve some for root
 186         */
 187        if (!cap_sys_admin)
 188                allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
 189        allowed += total_swap_pages;
 190
 191        /*
 192         * Don't let a single process grow so big a user can't recover
 193         */
 194        if (mm) {
 195                reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
 196                allowed -= min(mm->total_vm / 32, reserve);
 197        }
 198
 199        if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 200                return 0;
 201error:
 202        vm_unacct_memory(pages);
 203
 204        return -ENOMEM;
 205}
 206
 207/*
 208 * Requires inode->i_mapping->i_mmap_mutex
 209 */
 210static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 211                struct file *file, struct address_space *mapping)
 212{
 213        if (vma->vm_flags & VM_DENYWRITE)
 214                atomic_inc(&file_inode(file)->i_writecount);
 215        if (vma->vm_flags & VM_SHARED)
 216                mapping->i_mmap_writable--;
 217
 218        flush_dcache_mmap_lock(mapping);
 219        if (unlikely(vma->vm_flags & VM_NONLINEAR))
 220                list_del_init(&vma->shared.nonlinear);
 221        else
 222                vma_interval_tree_remove(vma, &mapping->i_mmap);
 223        flush_dcache_mmap_unlock(mapping);
 224}
 225
 226/*
 227 * Unlink a file-based vm structure from its interval tree, to hide
 228 * vma from rmap and vmtruncate before freeing its page tables.
 229 */
 230void unlink_file_vma(struct vm_area_struct *vma)
 231{
 232        struct file *file = vma->vm_file;
 233
 234        if (file) {
 235                struct address_space *mapping = file->f_mapping;
 236                mutex_lock(&mapping->i_mmap_mutex);
 237                __remove_shared_vm_struct(vma, file, mapping);
 238                mutex_unlock(&mapping->i_mmap_mutex);
 239        }
 240}
 241
 242/*
 243 * Close a vm structure and free it, returning the next.
 244 */
 245static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 246{
 247        struct vm_area_struct *next = vma->vm_next;
 248
 249        might_sleep();
 250        if (vma->vm_ops && vma->vm_ops->close)
 251                vma->vm_ops->close(vma);
 252        if (vma->vm_file)
 253                fput(vma->vm_file);
 254        mpol_put(vma_policy(vma));
 255        kmem_cache_free(vm_area_cachep, vma);
 256        return next;
 257}
 258
 259static unsigned long do_brk(unsigned long addr, unsigned long len);
 260
 261SYSCALL_DEFINE1(brk, unsigned long, brk)
 262{
 263        unsigned long rlim, retval;
 264        unsigned long newbrk, oldbrk;
 265        struct mm_struct *mm = current->mm;
 266        unsigned long min_brk;
 267        bool populate;
 268
 269        down_write(&mm->mmap_sem);
 270
 271#ifdef CONFIG_COMPAT_BRK
 272        /*
 273         * CONFIG_COMPAT_BRK can still be overridden by setting
 274         * randomize_va_space to 2, which will still cause mm->start_brk
 275         * to be arbitrarily shifted
 276         */
 277        if (current->brk_randomized)
 278                min_brk = mm->start_brk;
 279        else
 280                min_brk = mm->end_data;
 281#else
 282        min_brk = mm->start_brk;
 283#endif
 284        if (brk < min_brk)
 285                goto out;
 286
 287        /*
 288         * Check against rlimit here. If this check is done later after the test
 289         * of oldbrk with newbrk then it can escape the test and let the data
 290         * segment grow beyond its set limit the in case where the limit is
 291         * not page aligned -Ram Gupta
 292         */
 293        rlim = rlimit(RLIMIT_DATA);
 294        if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
 295                        (mm->end_data - mm->start_data) > rlim)
 296                goto out;
 297
 298        newbrk = PAGE_ALIGN(brk);
 299        oldbrk = PAGE_ALIGN(mm->brk);
 300        if (oldbrk == newbrk)
 301                goto set_brk;
 302
 303        /* Always allow shrinking brk. */
 304        if (brk <= mm->brk) {
 305                if (!do_munmap(mm, newbrk, oldbrk-newbrk))
 306                        goto set_brk;
 307                goto out;
 308        }
 309
 310        /* Check against existing mmap mappings. */
 311        if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 312                goto out;
 313
 314        /* Ok, looks good - let it rip. */
 315        if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
 316                goto out;
 317
 318set_brk:
 319        mm->brk = brk;
 320        populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
 321        up_write(&mm->mmap_sem);
 322        if (populate)
 323                mm_populate(oldbrk, newbrk - oldbrk);
 324        return brk;
 325
 326out:
 327        retval = mm->brk;
 328        up_write(&mm->mmap_sem);
 329        return retval;
 330}
 331
 332static long vma_compute_subtree_gap(struct vm_area_struct *vma)
 333{
 334        unsigned long max, subtree_gap;
 335        max = vma->vm_start;
 336        if (vma->vm_prev)
 337                max -= vma->vm_prev->vm_end;
 338        if (vma->vm_rb.rb_left) {
 339                subtree_gap = rb_entry(vma->vm_rb.rb_left,
 340                                struct vm_area_struct, vm_rb)->rb_subtree_gap;
 341                if (subtree_gap > max)
 342                        max = subtree_gap;
 343        }
 344        if (vma->vm_rb.rb_right) {
 345                subtree_gap = rb_entry(vma->vm_rb.rb_right,
 346                                struct vm_area_struct, vm_rb)->rb_subtree_gap;
 347                if (subtree_gap > max)
 348                        max = subtree_gap;
 349        }
 350        return max;
 351}
 352
 353#ifdef CONFIG_DEBUG_VM_RB
 354static int browse_rb(struct rb_root *root)
 355{
 356        int i = 0, j, bug = 0;
 357        struct rb_node *nd, *pn = NULL;
 358        unsigned long prev = 0, pend = 0;
 359
 360        for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 361                struct vm_area_struct *vma;
 362                vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 363                if (vma->vm_start < prev) {
 364                        printk("vm_start %lx prev %lx\n", vma->vm_start, prev);
 365                        bug = 1;
 366                }
 367                if (vma->vm_start < pend) {
 368                        printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
 369                        bug = 1;
 370                }
 371                if (vma->vm_start > vma->vm_end) {
 372                        printk("vm_end %lx < vm_start %lx\n",
 373                                vma->vm_end, vma->vm_start);
 374                        bug = 1;
 375                }
 376                if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
 377                        printk("free gap %lx, correct %lx\n",
 378                               vma->rb_subtree_gap,
 379                               vma_compute_subtree_gap(vma));
 380                        bug = 1;
 381                }
 382                i++;
 383                pn = nd;
 384                prev = vma->vm_start;
 385                pend = vma->vm_end;
 386        }
 387        j = 0;
 388        for (nd = pn; nd; nd = rb_prev(nd))
 389                j++;
 390        if (i != j) {
 391                printk("backwards %d, forwards %d\n", j, i);
 392                bug = 1;
 393        }
 394        return bug ? -1 : i;
 395}
 396
 397static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
 398{
 399        struct rb_node *nd;
 400
 401        for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 402                struct vm_area_struct *vma;
 403                vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 404                BUG_ON(vma != ignore &&
 405                       vma->rb_subtree_gap != vma_compute_subtree_gap(vma));
 406        }
 407}
 408
 409void validate_mm(struct mm_struct *mm)
 410{
 411        int bug = 0;
 412        int i = 0;
 413        unsigned long highest_address = 0;
 414        struct vm_area_struct *vma = mm->mmap;
 415        while (vma) {
 416                struct anon_vma_chain *avc;
 417                vma_lock_anon_vma(vma);
 418                list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 419                        anon_vma_interval_tree_verify(avc);
 420                vma_unlock_anon_vma(vma);
 421                highest_address = vma->vm_end;
 422                vma = vma->vm_next;
 423                i++;
 424        }
 425        if (i != mm->map_count) {
 426                printk("map_count %d vm_next %d\n", mm->map_count, i);
 427                bug = 1;
 428        }
 429        if (highest_address != mm->highest_vm_end) {
 430                printk("mm->highest_vm_end %lx, found %lx\n",
 431                       mm->highest_vm_end, highest_address);
 432                bug = 1;
 433        }
 434        i = browse_rb(&mm->mm_rb);
 435        if (i != mm->map_count) {
 436                printk("map_count %d rb %d\n", mm->map_count, i);
 437                bug = 1;
 438        }
 439        BUG_ON(bug);
 440}
 441#else
 442#define validate_mm_rb(root, ignore) do { } while (0)
 443#define validate_mm(mm) do { } while (0)
 444#endif
 445
 446RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
 447                     unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
 448
 449/*
 450 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
 451 * vma->vm_prev->vm_end values changed, without modifying the vma's position
 452 * in the rbtree.
 453 */
 454static void vma_gap_update(struct vm_area_struct *vma)
 455{
 456        /*
 457         * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
 458         * function that does exacltly what we want.
 459         */
 460        vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
 461}
 462
 463static inline void vma_rb_insert(struct vm_area_struct *vma,
 464                                 struct rb_root *root)
 465{
 466        /* All rb_subtree_gap values must be consistent prior to insertion */
 467        validate_mm_rb(root, NULL);
 468
 469        rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
 470}
 471
 472static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
 473{
 474        /*
 475         * All rb_subtree_gap values must be consistent prior to erase,
 476         * with the possible exception of the vma being erased.
 477         */
 478        validate_mm_rb(root, vma);
 479
 480        /*
 481         * Note rb_erase_augmented is a fairly large inline function,
 482         * so make sure we instantiate it only once with our desired
 483         * augmented rbtree callbacks.
 484         */
 485        rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
 486}
 487
 488/*
 489 * vma has some anon_vma assigned, and is already inserted on that
 490 * anon_vma's interval trees.
 491 *
 492 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
 493 * vma must be removed from the anon_vma's interval trees using
 494 * anon_vma_interval_tree_pre_update_vma().
 495 *
 496 * After the update, the vma will be reinserted using
 497 * anon_vma_interval_tree_post_update_vma().
 498 *
 499 * The entire update must be protected by exclusive mmap_sem and by
 500 * the root anon_vma's mutex.
 501 */
 502static inline void
 503anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
 504{
 505        struct anon_vma_chain *avc;
 506
 507        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 508                anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
 509}
 510
 511static inline void
 512anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
 513{
 514        struct anon_vma_chain *avc;
 515
 516        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 517                anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
 518}
 519
 520static int find_vma_links(struct mm_struct *mm, unsigned long addr,
 521                unsigned long end, struct vm_area_struct **pprev,
 522                struct rb_node ***rb_link, struct rb_node **rb_parent)
 523{
 524        struct rb_node **__rb_link, *__rb_parent, *rb_prev;
 525
 526        __rb_link = &mm->mm_rb.rb_node;
 527        rb_prev = __rb_parent = NULL;
 528
 529        while (*__rb_link) {
 530                struct vm_area_struct *vma_tmp;
 531
 532                __rb_parent = *__rb_link;
 533                vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
 534
 535                if (vma_tmp->vm_end > addr) {
 536                        /* Fail if an existing vma overlaps the area */
 537                        if (vma_tmp->vm_start < end)
 538                                return -ENOMEM;
 539                        __rb_link = &__rb_parent->rb_left;
 540                } else {
 541                        rb_prev = __rb_parent;
 542                        __rb_link = &__rb_parent->rb_right;
 543                }
 544        }
 545
 546        *pprev = NULL;
 547        if (rb_prev)
 548                *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 549        *rb_link = __rb_link;
 550        *rb_parent = __rb_parent;
 551        return 0;
 552}
 553
 554static unsigned long count_vma_pages_range(struct mm_struct *mm,
 555                unsigned long addr, unsigned long end)
 556{
 557        unsigned long nr_pages = 0;
 558        struct vm_area_struct *vma;
 559
 560        /* Find first overlaping mapping */
 561        vma = find_vma_intersection(mm, addr, end);
 562        if (!vma)
 563                return 0;
 564
 565        nr_pages = (min(end, vma->vm_end) -
 566                max(addr, vma->vm_start)) >> PAGE_SHIFT;
 567
 568        /* Iterate over the rest of the overlaps */
 569        for (vma = vma->vm_next; vma; vma = vma->vm_next) {
 570                unsigned long overlap_len;
 571
 572                if (vma->vm_start > end)
 573                        break;
 574
 575                overlap_len = min(end, vma->vm_end) - vma->vm_start;
 576                nr_pages += overlap_len >> PAGE_SHIFT;
 577        }
 578
 579        return nr_pages;
 580}
 581
 582void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
 583                struct rb_node **rb_link, struct rb_node *rb_parent)
 584{
 585        /* Update tracking information for the gap following the new vma. */
 586        if (vma->vm_next)
 587                vma_gap_update(vma->vm_next);
 588        else
 589                mm->highest_vm_end = vma->vm_end;
 590
 591        /*
 592         * vma->vm_prev wasn't known when we followed the rbtree to find the
 593         * correct insertion point for that vma. As a result, we could not
 594         * update the vma vm_rb parents rb_subtree_gap values on the way down.
 595         * So, we first insert the vma with a zero rb_subtree_gap value
 596         * (to be consistent with what we did on the way down), and then
 597         * immediately update the gap to the correct value. Finally we
 598         * rebalance the rbtree after all augmented values have been set.
 599         */
 600        rb_link_node(&vma->vm_rb, rb_parent, rb_link);
 601        vma->rb_subtree_gap = 0;
 602        vma_gap_update(vma);
 603        vma_rb_insert(vma, &mm->mm_rb);
 604}
 605
 606static void __vma_link_file(struct vm_area_struct *vma)
 607{
 608        struct file *file;
 609
 610        file = vma->vm_file;
 611        if (file) {
 612                struct address_space *mapping = file->f_mapping;
 613
 614                if (vma->vm_flags & VM_DENYWRITE)
 615                        atomic_dec(&file_inode(file)->i_writecount);
 616                if (vma->vm_flags & VM_SHARED)
 617                        mapping->i_mmap_writable++;
 618
 619                flush_dcache_mmap_lock(mapping);
 620                if (unlikely(vma->vm_flags & VM_NONLINEAR))
 621                        vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 622                else
 623                        vma_interval_tree_insert(vma, &mapping->i_mmap);
 624                flush_dcache_mmap_unlock(mapping);
 625        }
 626}
 627
 628static void
 629__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 630        struct vm_area_struct *prev, struct rb_node **rb_link,
 631        struct rb_node *rb_parent)
 632{
 633        __vma_link_list(mm, vma, prev, rb_parent);
 634        __vma_link_rb(mm, vma, rb_link, rb_parent);
 635}
 636
 637static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 638                        struct vm_area_struct *prev, struct rb_node **rb_link,
 639                        struct rb_node *rb_parent)
 640{
 641        struct address_space *mapping = NULL;
 642
 643        if (vma->vm_file)
 644                mapping = vma->vm_file->f_mapping;
 645
 646        if (mapping)
 647                mutex_lock(&mapping->i_mmap_mutex);
 648
 649        __vma_link(mm, vma, prev, rb_link, rb_parent);
 650        __vma_link_file(vma);
 651
 652        if (mapping)
 653                mutex_unlock(&mapping->i_mmap_mutex);
 654
 655        mm->map_count++;
 656        validate_mm(mm);
 657}
 658
 659/*
 660 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
 661 * mm's list and rbtree.  It has already been inserted into the interval tree.
 662 */
 663static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
 664{
 665        struct vm_area_struct *prev;
 666        struct rb_node **rb_link, *rb_parent;
 667
 668        if (find_vma_links(mm, vma->vm_start, vma->vm_end,
 669                           &prev, &rb_link, &rb_parent))
 670                BUG();
 671        __vma_link(mm, vma, prev, rb_link, rb_parent);
 672        mm->map_count++;
 673}
 674
 675static inline void
 676__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
 677                struct vm_area_struct *prev)
 678{
 679        struct vm_area_struct *next;
 680
 681        vma_rb_erase(vma, &mm->mm_rb);
 682        prev->vm_next = next = vma->vm_next;
 683        if (next)
 684                next->vm_prev = prev;
 685        if (mm->mmap_cache == vma)
 686                mm->mmap_cache = prev;
 687}
 688
 689/*
 690 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
 691 * is already present in an i_mmap tree without adjusting the tree.
 692 * The following helper function should be used when such adjustments
 693 * are necessary.  The "insert" vma (if any) is to be inserted
 694 * before we drop the necessary locks.
 695 */
 696int vma_adjust(struct vm_area_struct *vma, unsigned long start,
 697        unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
 698{
 699        struct mm_struct *mm = vma->vm_mm;
 700        struct vm_area_struct *next = vma->vm_next;
 701        struct vm_area_struct *importer = NULL;
 702        struct address_space *mapping = NULL;
 703        struct rb_root *root = NULL;
 704        struct anon_vma *anon_vma = NULL;
 705        struct file *file = vma->vm_file;
 706        bool start_changed = false, end_changed = false;
 707        long adjust_next = 0;
 708        int remove_next = 0;
 709
 710        if (next && !insert) {
 711                struct vm_area_struct *exporter = NULL;
 712
 713                if (end >= next->vm_end) {
 714                        /*
 715                         * vma expands, overlapping all the next, and
 716                         * perhaps the one after too (mprotect case 6).
 717                         */
 718again:                  remove_next = 1 + (end > next->vm_end);
 719                        end = next->vm_end;
 720                        exporter = next;
 721                        importer = vma;
 722                } else if (end > next->vm_start) {
 723                        /*
 724                         * vma expands, overlapping part of the next:
 725                         * mprotect case 5 shifting the boundary up.
 726                         */
 727                        adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
 728                        exporter = next;
 729                        importer = vma;
 730                } else if (end < vma->vm_end) {
 731                        /*
 732                         * vma shrinks, and !insert tells it's not
 733                         * split_vma inserting another: so it must be
 734                         * mprotect case 4 shifting the boundary down.
 735                         */
 736                        adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
 737                        exporter = vma;
 738                        importer = next;
 739                }
 740
 741                /*
 742                 * Easily overlooked: when mprotect shifts the boundary,
 743                 * make sure the expanding vma has anon_vma set if the
 744                 * shrinking vma had, to cover any anon pages imported.
 745                 */
 746                if (exporter && exporter->anon_vma && !importer->anon_vma) {
 747                        if (anon_vma_clone(importer, exporter))
 748                                return -ENOMEM;
 749                        importer->anon_vma = exporter->anon_vma;
 750                }
 751        }
 752
 753        if (file) {
 754                mapping = file->f_mapping;
 755                if (!(vma->vm_flags & VM_NONLINEAR)) {
 756                        root = &mapping->i_mmap;
 757                        uprobe_munmap(vma, vma->vm_start, vma->vm_end);
 758
 759                        if (adjust_next)
 760                                uprobe_munmap(next, next->vm_start,
 761                                                        next->vm_end);
 762                }
 763
 764                mutex_lock(&mapping->i_mmap_mutex);
 765                if (insert) {
 766                        /*
 767                         * Put into interval tree now, so instantiated pages
 768                         * are visible to arm/parisc __flush_dcache_page
 769                         * throughout; but we cannot insert into address
 770                         * space until vma start or end is updated.
 771                         */
 772                        __vma_link_file(insert);
 773                }
 774        }
 775
 776        vma_adjust_trans_huge(vma, start, end, adjust_next);
 777
 778        anon_vma = vma->anon_vma;
 779        if (!anon_vma && adjust_next)
 780                anon_vma = next->anon_vma;
 781        if (anon_vma) {
 782                VM_BUG_ON(adjust_next && next->anon_vma &&
 783                          anon_vma != next->anon_vma);
 784                anon_vma_lock_write(anon_vma);
 785                anon_vma_interval_tree_pre_update_vma(vma);
 786                if (adjust_next)
 787                        anon_vma_interval_tree_pre_update_vma(next);
 788        }
 789
 790        if (root) {
 791                flush_dcache_mmap_lock(mapping);
 792                vma_interval_tree_remove(vma, root);
 793                if (adjust_next)
 794                        vma_interval_tree_remove(next, root);
 795        }
 796
 797        if (start != vma->vm_start) {
 798                vma->vm_start = start;
 799                start_changed = true;
 800        }
 801        if (end != vma->vm_end) {
 802                vma->vm_end = end;
 803                end_changed = true;
 804        }
 805        vma->vm_pgoff = pgoff;
 806        if (adjust_next) {
 807                next->vm_start += adjust_next << PAGE_SHIFT;
 808                next->vm_pgoff += adjust_next;
 809        }
 810
 811        if (root) {
 812                if (adjust_next)
 813                        vma_interval_tree_insert(next, root);
 814                vma_interval_tree_insert(vma, root);
 815                flush_dcache_mmap_unlock(mapping);
 816        }
 817
 818        if (remove_next) {
 819                /*
 820                 * vma_merge has merged next into vma, and needs
 821                 * us to remove next before dropping the locks.
 822                 */
 823                __vma_unlink(mm, next, vma);
 824                if (file)
 825                        __remove_shared_vm_struct(next, file, mapping);
 826        } else if (insert) {
 827                /*
 828                 * split_vma has split insert from vma, and needs
 829                 * us to insert it before dropping the locks
 830                 * (it may either follow vma or precede it).
 831                 */
 832                __insert_vm_struct(mm, insert);
 833        } else {
 834                if (start_changed)
 835                        vma_gap_update(vma);
 836                if (end_changed) {
 837                        if (!next)
 838                                mm->highest_vm_end = end;
 839                        else if (!adjust_next)
 840                                vma_gap_update(next);
 841                }
 842        }
 843
 844        if (anon_vma) {
 845                anon_vma_interval_tree_post_update_vma(vma);
 846                if (adjust_next)
 847                        anon_vma_interval_tree_post_update_vma(next);
 848                anon_vma_unlock_write(anon_vma);
 849        }
 850        if (mapping)
 851                mutex_unlock(&mapping->i_mmap_mutex);
 852
 853        if (root) {
 854                uprobe_mmap(vma);
 855
 856                if (adjust_next)
 857                        uprobe_mmap(next);
 858        }
 859
 860        if (remove_next) {
 861                if (file) {
 862                        uprobe_munmap(next, next->vm_start, next->vm_end);
 863                        fput(file);
 864                }
 865                if (next->anon_vma)
 866                        anon_vma_merge(vma, next);
 867                mm->map_count--;
 868                mpol_put(vma_policy(next));
 869                kmem_cache_free(vm_area_cachep, next);
 870                /*
 871                 * In mprotect's case 6 (see comments on vma_merge),
 872                 * we must remove another next too. It would clutter
 873                 * up the code too much to do both in one go.
 874                 */
 875                next = vma->vm_next;
 876                if (remove_next == 2)
 877                        goto again;
 878                else if (next)
 879                        vma_gap_update(next);
 880                else
 881                        mm->highest_vm_end = end;
 882        }
 883        if (insert && file)
 884                uprobe_mmap(insert);
 885
 886        validate_mm(mm);
 887
 888        return 0;
 889}
 890
 891/*
 892 * If the vma has a ->close operation then the driver probably needs to release
 893 * per-vma resources, so we don't attempt to merge those.
 894 */
 895static inline int is_mergeable_vma(struct vm_area_struct *vma,
 896                        struct file *file, unsigned long vm_flags)
 897{
 898        if (vma->vm_flags ^ vm_flags)
 899                return 0;
 900        if (vma->vm_file != file)
 901                return 0;
 902        if (vma->vm_ops && vma->vm_ops->close)
 903                return 0;
 904        return 1;
 905}
 906
 907static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
 908                                        struct anon_vma *anon_vma2,
 909                                        struct vm_area_struct *vma)
 910{
 911        /*
 912         * The list_is_singular() test is to avoid merging VMA cloned from
 913         * parents. This can improve scalability caused by anon_vma lock.
 914         */
 915        if ((!anon_vma1 || !anon_vma2) && (!vma ||
 916                list_is_singular(&vma->anon_vma_chain)))
 917                return 1;
 918        return anon_vma1 == anon_vma2;
 919}
 920
 921/*
 922 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 923 * in front of (at a lower virtual address and file offset than) the vma.
 924 *
 925 * We cannot merge two vmas if they have differently assigned (non-NULL)
 926 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 927 *
 928 * We don't check here for the merged mmap wrapping around the end of pagecache
 929 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
 930 * wrap, nor mmaps which cover the final page at index -1UL.
 931 */
 932static int
 933can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
 934        struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 935{
 936        if (is_mergeable_vma(vma, file, vm_flags) &&
 937            is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 938                if (vma->vm_pgoff == vm_pgoff)
 939                        return 1;
 940        }
 941        return 0;
 942}
 943
 944/*
 945 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 946 * beyond (at a higher virtual address and file offset than) the vma.
 947 *
 948 * We cannot merge two vmas if they have differently assigned (non-NULL)
 949 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 950 */
 951static int
 952can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 953        struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 954{
 955        if (is_mergeable_vma(vma, file, vm_flags) &&
 956            is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 957                pgoff_t vm_pglen;
 958                vm_pglen = vma_pages(vma);
 959                if (vma->vm_pgoff + vm_pglen == vm_pgoff)
 960                        return 1;
 961        }
 962        return 0;
 963}
 964
 965/*
 966 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
 967 * whether that can be merged with its predecessor or its successor.
 968 * Or both (it neatly fills a hole).
 969 *
 970 * In most cases - when called for mmap, brk or mremap - [addr,end) is
 971 * certain not to be mapped by the time vma_merge is called; but when
 972 * called for mprotect, it is certain to be already mapped (either at
 973 * an offset within prev, or at the start of next), and the flags of
 974 * this area are about to be changed to vm_flags - and the no-change
 975 * case has already been eliminated.
 976 *
 977 * The following mprotect cases have to be considered, where AAAA is
 978 * the area passed down from mprotect_fixup, never extending beyond one
 979 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
 980 *
 981 *     AAAA             AAAA                AAAA          AAAA
 982 *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
 983 *    cannot merge    might become    might become    might become
 984 *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
 985 *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
 986 *    mremap move:                                    PPPPNNNNNNNN 8
 987 *        AAAA
 988 *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
 989 *    might become    case 1 below    case 2 below    case 3 below
 990 *
 991 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
 992 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
 993 */
 994struct vm_area_struct *vma_merge(struct mm_struct *mm,
 995                        struct vm_area_struct *prev, unsigned long addr,
 996                        unsigned long end, unsigned long vm_flags,
 997                        struct anon_vma *anon_vma, struct file *file,
 998                        pgoff_t pgoff, struct mempolicy *policy)
 999{
1000        pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1001        struct vm_area_struct *area, *next;
1002        int err;
1003
1004        /*
1005         * We later require that vma->vm_flags == vm_flags,
1006         * so this tests vma->vm_flags & VM_SPECIAL, too.
1007         */
1008        if (vm_flags & VM_SPECIAL)
1009                return NULL;
1010
1011        if (prev)
1012                next = prev->vm_next;
1013        else
1014                next = mm->mmap;
1015        area = next;
1016        if (next && next->vm_end == end)                /* cases 6, 7, 8 */
1017                next = next->vm_next;
1018
1019        /*
1020         * Can it merge with the predecessor?
1021         */
1022        if (prev && prev->vm_end == addr &&
1023                        mpol_equal(vma_policy(prev), policy) &&
1024                        can_vma_merge_after(prev, vm_flags,
1025                                                anon_vma, file, pgoff)) {
1026                /*
1027                 * OK, it can.  Can we now merge in the successor as well?
1028                 */
1029                if (next && end == next->vm_start &&
1030                                mpol_equal(policy, vma_policy(next)) &&
1031                                can_vma_merge_before(next, vm_flags,
1032                                        anon_vma, file, pgoff+pglen) &&
1033                                is_mergeable_anon_vma(prev->anon_vma,
1034                                                      next->anon_vma, NULL)) {
1035                                                        /* cases 1, 6 */
1036                        err = vma_adjust(prev, prev->vm_start,
1037                                next->vm_end, prev->vm_pgoff, NULL);
1038                } else                                  /* cases 2, 5, 7 */
1039                        err = vma_adjust(prev, prev->vm_start,
1040                                end, prev->vm_pgoff, NULL);
1041                if (err)
1042                        return NULL;
1043                khugepaged_enter_vma_merge(prev);
1044                return prev;
1045        }
1046
1047        /*
1048         * Can this new request be merged in front of next?
1049         */
1050        if (next && end == next->vm_start &&
1051                        mpol_equal(policy, vma_policy(next)) &&
1052                        can_vma_merge_before(next, vm_flags,
1053                                        anon_vma, file, pgoff+pglen)) {
1054                if (prev && addr < prev->vm_end)        /* case 4 */
1055                        err = vma_adjust(prev, prev->vm_start,
1056                                addr, prev->vm_pgoff, NULL);
1057                else                                    /* cases 3, 8 */
1058                        err = vma_adjust(area, addr, next->vm_end,
1059                                next->vm_pgoff - pglen, NULL);
1060                if (err)
1061                        return NULL;
1062                khugepaged_enter_vma_merge(area);
1063                return area;
1064        }
1065
1066        return NULL;
1067}
1068
1069/*
1070 * Rough compatbility check to quickly see if it's even worth looking
1071 * at sharing an anon_vma.
1072 *
1073 * They need to have the same vm_file, and the flags can only differ
1074 * in things that mprotect may change.
1075 *
1076 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1077 * we can merge the two vma's. For example, we refuse to merge a vma if
1078 * there is a vm_ops->close() function, because that indicates that the
1079 * driver is doing some kind of reference counting. But that doesn't
1080 * really matter for the anon_vma sharing case.
1081 */
1082static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1083{
1084        return a->vm_end == b->vm_start &&
1085                mpol_equal(vma_policy(a), vma_policy(b)) &&
1086                a->vm_file == b->vm_file &&
1087                !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
1088                b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1089}
1090
1091/*
1092 * Do some basic sanity checking to see if we can re-use the anon_vma
1093 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1094 * the same as 'old', the other will be the new one that is trying
1095 * to share the anon_vma.
1096 *
1097 * NOTE! This runs with mm_sem held for reading, so it is possible that
1098 * the anon_vma of 'old' is concurrently in the process of being set up
1099 * by another page fault trying to merge _that_. But that's ok: if it
1100 * is being set up, that automatically means that it will be a singleton
1101 * acceptable for merging, so we can do all of this optimistically. But
1102 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1103 *
1104 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1105 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1106 * is to return an anon_vma that is "complex" due to having gone through
1107 * a fork).
1108 *
1109 * We also make sure that the two vma's are compatible (adjacent,
1110 * and with the same memory policies). That's all stable, even with just
1111 * a read lock on the mm_sem.
1112 */
1113static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1114{
1115        if (anon_vma_compatible(a, b)) {
1116                struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
1117
1118                if (anon_vma && list_is_singular(&old->anon_vma_chain))
1119                        return anon_vma;
1120        }
1121        return NULL;
1122}
1123
1124/*
1125 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1126 * neighbouring vmas for a suitable anon_vma, before it goes off
1127 * to allocate a new anon_vma.  It checks because a repetitive
1128 * sequence of mprotects and faults may otherwise lead to distinct
1129 * anon_vmas being allocated, preventing vma merge in subsequent
1130 * mprotect.
1131 */
1132struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1133{
1134        struct anon_vma *anon_vma;
1135        struct vm_area_struct *near;
1136
1137        near = vma->vm_next;
1138        if (!near)
1139                goto try_prev;
1140
1141        anon_vma = reusable_anon_vma(near, vma, near);
1142        if (anon_vma)
1143                return anon_vma;
1144try_prev:
1145        near = vma->vm_prev;
1146        if (!near)
1147                goto none;
1148
1149        anon_vma = reusable_anon_vma(near, near, vma);
1150        if (anon_vma)
1151                return anon_vma;
1152none:
1153        /*
1154         * There's no absolute need to look only at touching neighbours:
1155         * we could search further afield for "compatible" anon_vmas.
1156         * But it would probably just be a waste of time searching,
1157         * or lead to too many vmas hanging off the same anon_vma.
1158         * We're trying to allow mprotect remerging later on,
1159         * not trying to minimize memory used for anon_vmas.
1160         */
1161        return NULL;
1162}
1163
1164#ifdef CONFIG_PROC_FS
1165void vm_stat_account(struct mm_struct *mm, unsigned long flags,
1166                                                struct file *file, long pages)
1167{
1168        const unsigned long stack_flags
1169                = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
1170
1171        mm->total_vm += pages;
1172
1173        if (file) {
1174                mm->shared_vm += pages;
1175                if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
1176                        mm->exec_vm += pages;
1177        } else if (flags & stack_flags)
1178                mm->stack_vm += pages;
1179}
1180#endif /* CONFIG_PROC_FS */
1181
1182/*
1183 * If a hint addr is less than mmap_min_addr change hint to be as
1184 * low as possible but still greater than mmap_min_addr
1185 */
1186static inline unsigned long round_hint_to_min(unsigned long hint)
1187{
1188        hint &= PAGE_MASK;
1189        if (((void *)hint != NULL) &&
1190            (hint < mmap_min_addr))
1191                return PAGE_ALIGN(mmap_min_addr);
1192        return hint;
1193}
1194
1195/*
1196 * The caller must hold down_write(&current->mm->mmap_sem).
1197 */
1198
1199unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
1200                        unsigned long len, unsigned long prot,
1201                        unsigned long flags, unsigned long pgoff,
1202                        unsigned long *populate)
1203{
1204        struct mm_struct * mm = current->mm;
1205        vm_flags_t vm_flags;
1206
1207        *populate = 0;
1208
1209        /*
1210         * Does the application expect PROT_READ to imply PROT_EXEC?
1211         *
1212         * (the exception is when the underlying filesystem is noexec
1213         *  mounted, in which case we dont add PROT_EXEC.)
1214         */
1215        if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1216                if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
1217                        prot |= PROT_EXEC;
1218
1219        if (!len)
1220                return -EINVAL;
1221
1222        if (!(flags & MAP_FIXED))
1223                addr = round_hint_to_min(addr);
1224
1225        /* Careful about overflows.. */
1226        len = PAGE_ALIGN(len);
1227        if (!len)
1228                return -ENOMEM;
1229
1230        /* offset overflow? */
1231        if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1232               return -EOVERFLOW;
1233
1234        /* Too many mappings? */
1235        if (mm->map_count > sysctl_max_map_count)
1236                return -ENOMEM;
1237
1238        /* Obtain the address to map to. we verify (or select) it and ensure
1239         * that it represents a valid section of the address space.
1240         */
1241        addr = get_unmapped_area(file, addr, len, pgoff, flags);
1242        if (addr & ~PAGE_MASK)
1243                return addr;
1244
1245        /* Do simple checking here so the lower-level routines won't have
1246         * to. we assume access permissions have been handled by the open
1247         * of the memory object, so we don't do any here.
1248         */
1249        vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1250                        mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1251
1252        if (flags & MAP_LOCKED)
1253                if (!can_do_mlock())
1254                        return -EPERM;
1255
1256        /* mlock MCL_FUTURE? */
1257        if (vm_flags & VM_LOCKED) {
1258                unsigned long locked, lock_limit;
1259                locked = len >> PAGE_SHIFT;
1260                locked += mm->locked_vm;
1261                lock_limit = rlimit(RLIMIT_MEMLOCK);
1262                lock_limit >>= PAGE_SHIFT;
1263                if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1264                        return -EAGAIN;
1265        }
1266
1267        if (file) {
1268                struct inode *inode = file_inode(file);
1269
1270                switch (flags & MAP_TYPE) {
1271                case MAP_SHARED:
1272                        if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1273                                return -EACCES;
1274
1275                        /*
1276                         * Make sure we don't allow writing to an append-only
1277                         * file..
1278                         */
1279                        if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1280                                return -EACCES;
1281
1282                        /*
1283                         * Make sure there are no mandatory locks on the file.
1284                         */
1285                        if (locks_verify_locked(inode))
1286                                return -EAGAIN;
1287
1288                        vm_flags |= VM_SHARED | VM_MAYSHARE;
1289                        if (!(file->f_mode & FMODE_WRITE))
1290                                vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1291
1292                        /* fall through */
1293                case MAP_PRIVATE:
1294                        if (!(file->f_mode & FMODE_READ))
1295                                return -EACCES;
1296                        if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1297                                if (vm_flags & VM_EXEC)
1298                                        return -EPERM;
1299                                vm_flags &= ~VM_MAYEXEC;
1300                        }
1301
1302                        if (!file->f_op || !file->f_op->mmap)
1303                                return -ENODEV;
1304                        if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1305                                return -EINVAL;
1306                        break;
1307
1308                default:
1309                        return -EINVAL;
1310                }
1311        } else {
1312                switch (flags & MAP_TYPE) {
1313                case MAP_SHARED:
1314                        if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1315                                return -EINVAL;
1316                        /*
1317                         * Ignore pgoff.
1318                         */
1319                        pgoff = 0;
1320                        vm_flags |= VM_SHARED | VM_MAYSHARE;
1321                        break;
1322                case MAP_PRIVATE:
1323                        /*
1324                         * Set pgoff according to addr for anon_vma.
1325                         */
1326                        pgoff = addr >> PAGE_SHIFT;
1327                        break;
1328                default:
1329                        return -EINVAL;
1330                }
1331        }
1332
1333        /*
1334         * Set 'VM_NORESERVE' if we should not account for the
1335         * memory use of this mapping.
1336         */
1337        if (flags & MAP_NORESERVE) {
1338                /* We honor MAP_NORESERVE if allowed to overcommit */
1339                if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1340                        vm_flags |= VM_NORESERVE;
1341
1342                /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1343                if (file && is_file_hugepages(file))
1344                        vm_flags |= VM_NORESERVE;
1345        }
1346
1347        addr = mmap_region(file, addr, len, vm_flags, pgoff);
1348        if (!IS_ERR_VALUE(addr) &&
1349            ((vm_flags & VM_LOCKED) ||
1350             (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1351                *populate = len;
1352        return addr;
1353}
1354
1355SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1356                unsigned long, prot, unsigned long, flags,
1357                unsigned long, fd, unsigned long, pgoff)
1358{
1359        struct file *file = NULL;
1360        unsigned long retval = -EBADF;
1361
1362        if (!(flags & MAP_ANONYMOUS)) {
1363                audit_mmap_fd(fd, flags);
1364                file = fget(fd);
1365                if (!file)
1366                        goto out;
1367                if (is_file_hugepages(file))
1368                        len = ALIGN(len, huge_page_size(hstate_file(file)));
1369                retval = -EINVAL;
1370                if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1371                        goto out_fput;
1372        } else if (flags & MAP_HUGETLB) {
1373                struct user_struct *user = NULL;
1374                struct hstate *hs;
1375
1376                hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
1377                if (!hs)
1378                        return -EINVAL;
1379
1380                len = ALIGN(len, huge_page_size(hs));
1381                /*
1382                 * VM_NORESERVE is used because the reservations will be
1383                 * taken when vm_ops->mmap() is called
1384                 * A dummy user value is used because we are not locking
1385                 * memory so no accounting is necessary
1386                 */
1387                file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1388                                VM_NORESERVE,
1389                                &user, HUGETLB_ANONHUGE_INODE,
1390                                (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1391                if (IS_ERR(file))
1392                        return PTR_ERR(file);
1393        }
1394
1395        flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1396
1397        retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1398out_fput:
1399        if (file)
1400                fput(file);
1401out:
1402        return retval;
1403}
1404
1405#ifdef __ARCH_WANT_SYS_OLD_MMAP
1406struct mmap_arg_struct {
1407        unsigned long addr;
1408        unsigned long len;
1409        unsigned long prot;
1410        unsigned long flags;
1411        unsigned long fd;
1412        unsigned long offset;
1413};
1414
1415SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1416{
1417        struct mmap_arg_struct a;
1418
1419        if (copy_from_user(&a, arg, sizeof(a)))
1420                return -EFAULT;
1421        if (a.offset & ~PAGE_MASK)
1422                return -EINVAL;
1423
1424        return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1425                              a.offset >> PAGE_SHIFT);
1426}
1427#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1428
1429/*
1430 * Some shared mappigns will want the pages marked read-only
1431 * to track write events. If so, we'll downgrade vm_page_prot
1432 * to the private version (using protection_map[] without the
1433 * VM_SHARED bit).
1434 */
1435int vma_wants_writenotify(struct vm_area_struct *vma)
1436{
1437        vm_flags_t vm_flags = vma->vm_flags;
1438
1439        /* If it was private or non-writable, the write bit is already clear */
1440        if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1441                return 0;
1442
1443        /* The backer wishes to know when pages are first written to? */
1444        if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1445                return 1;
1446
1447        /* The open routine did something to the protections already? */
1448        if (pgprot_val(vma->vm_page_prot) !=
1449            pgprot_val(vm_get_page_prot(vm_flags)))
1450                return 0;
1451
1452        /* Specialty mapping? */
1453        if (vm_flags & VM_PFNMAP)
1454                return 0;
1455
1456        /* Can the mapping track the dirty pages? */
1457        return vma->vm_file && vma->vm_file->f_mapping &&
1458                mapping_cap_account_dirty(vma->vm_file->f_mapping);
1459}
1460
1461/*
1462 * We account for memory if it's a private writeable mapping,
1463 * not hugepages and VM_NORESERVE wasn't set.
1464 */
1465static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1466{
1467        /*
1468         * hugetlb has its own accounting separate from the core VM
1469         * VM_HUGETLB may not be set yet so we cannot check for that flag.
1470         */
1471        if (file && is_file_hugepages(file))
1472                return 0;
1473
1474        return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1475}
1476
1477unsigned long mmap_region(struct file *file, unsigned long addr,
1478                unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
1479{
1480        struct mm_struct *mm = current->mm;
1481        struct vm_area_struct *vma, *prev;
1482        int error;
1483        struct rb_node **rb_link, *rb_parent;
1484        unsigned long charged = 0;
1485
1486        /* Check against address space limit. */
1487        if (!may_expand_vm(mm, len >> PAGE_SHIFT)) {
1488                unsigned long nr_pages;
1489
1490                /*
1491                 * MAP_FIXED may remove pages of mappings that intersects with
1492                 * requested mapping. Account for the pages it would unmap.
1493                 */
1494                if (!(vm_flags & MAP_FIXED))
1495                        return -ENOMEM;
1496
1497                nr_pages = count_vma_pages_range(mm, addr, addr + len);
1498
1499                if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages))
1500                        return -ENOMEM;
1501        }
1502
1503        /* Clear old maps */
1504        error = -ENOMEM;
1505munmap_back:
1506        if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
1507                if (do_munmap(mm, addr, len))
1508                        return -ENOMEM;
1509                goto munmap_back;
1510        }
1511
1512        /*
1513         * Private writable mapping: check memory availability
1514         */
1515        if (accountable_mapping(file, vm_flags)) {
1516                charged = len >> PAGE_SHIFT;
1517                if (security_vm_enough_memory_mm(mm, charged))
1518                        return -ENOMEM;
1519                vm_flags |= VM_ACCOUNT;
1520        }
1521
1522        /*
1523         * Can we just expand an old mapping?
1524         */
1525        vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1526        if (vma)
1527                goto out;
1528
1529        /*
1530         * Determine the object being mapped and call the appropriate
1531         * specific mapper. the address has already been validated, but
1532         * not unmapped, but the maps are removed from the list.
1533         */
1534        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1535        if (!vma) {
1536                error = -ENOMEM;
1537                goto unacct_error;
1538        }
1539
1540        vma->vm_mm = mm;
1541        vma->vm_start = addr;
1542        vma->vm_end = addr + len;
1543        vma->vm_flags = vm_flags;
1544        vma->vm_page_prot = vm_get_page_prot(vm_flags);
1545        vma->vm_pgoff = pgoff;
1546        INIT_LIST_HEAD(&vma->anon_vma_chain);
1547
1548        if (file) {
1549                if (vm_flags & VM_DENYWRITE) {
1550                        error = deny_write_access(file);
1551                        if (error)
1552                                goto free_vma;
1553                }
1554                vma->vm_file = get_file(file);
1555                error = file->f_op->mmap(file, vma);
1556                if (error)
1557                        goto unmap_and_free_vma;
1558
1559                /* Can addr have changed??
1560                 *
1561                 * Answer: Yes, several device drivers can do it in their
1562                 *         f_op->mmap method. -DaveM
1563                 * Bug: If addr is changed, prev, rb_link, rb_parent should
1564                 *      be updated for vma_link()
1565                 */
1566                WARN_ON_ONCE(addr != vma->vm_start);
1567
1568                addr = vma->vm_start;
1569                vm_flags = vma->vm_flags;
1570        } else if (vm_flags & VM_SHARED) {
1571                error = shmem_zero_setup(vma);
1572                if (error)
1573                        goto free_vma;
1574        }
1575
1576        if (vma_wants_writenotify(vma)) {
1577                pgprot_t pprot = vma->vm_page_prot;
1578
1579                /* Can vma->vm_page_prot have changed??
1580                 *
1581                 * Answer: Yes, drivers may have changed it in their
1582                 *         f_op->mmap method.
1583                 *
1584                 * Ensures that vmas marked as uncached stay that way.
1585                 */
1586                vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1587                if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1588                        vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1589        }
1590
1591        vma_link(mm, vma, prev, rb_link, rb_parent);
1592        /* Once vma denies write, undo our temporary denial count */
1593        if (vm_flags & VM_DENYWRITE)
1594                allow_write_access(file);
1595        file = vma->vm_file;
1596out:
1597        perf_event_mmap(vma);
1598
1599        vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1600        if (vm_flags & VM_LOCKED) {
1601                if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1602                                        vma == get_gate_vma(current->mm)))
1603                        mm->locked_vm += (len >> PAGE_SHIFT);
1604                else
1605                        vma->vm_flags &= ~VM_LOCKED;
1606        }
1607
1608        if (file)
1609                uprobe_mmap(vma);
1610
1611        /*
1612         * New (or expanded) vma always get soft dirty status.
1613         * Otherwise user-space soft-dirty page tracker won't
1614         * be able to distinguish situation when vma area unmapped,
1615         * then new mapped in-place (which must be aimed as
1616         * a completely new data area).
1617         */
1618        vma->vm_flags |= VM_SOFTDIRTY;
1619
1620        return addr;
1621
1622unmap_and_free_vma:
1623        if (vm_flags & VM_DENYWRITE)
1624                allow_write_access(file);
1625        vma->vm_file = NULL;
1626        fput(file);
1627
1628        /* Undo any partial mapping done by a device driver. */
1629        unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1630        charged = 0;
1631free_vma:
1632        kmem_cache_free(vm_area_cachep, vma);
1633unacct_error:
1634        if (charged)
1635                vm_unacct_memory(charged);
1636        return error;
1637}
1638
1639unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1640{
1641        /*
1642         * We implement the search by looking for an rbtree node that
1643         * immediately follows a suitable gap. That is,
1644         * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1645         * - gap_end   = vma->vm_start        >= info->low_limit  + length;
1646         * - gap_end - gap_start >= length
1647         */
1648
1649        struct mm_struct *mm = current->mm;
1650        struct vm_area_struct *vma;
1651        unsigned long length, low_limit, high_limit, gap_start, gap_end;
1652
1653        /* Adjust search length to account for worst case alignment overhead */
1654        length = info->length + info->align_mask;
1655        if (length < info->length)
1656                return -ENOMEM;
1657
1658        /* Adjust search limits by the desired length */
1659        if (info->high_limit < length)
1660                return -ENOMEM;
1661        high_limit = info->high_limit - length;
1662
1663        if (info->low_limit > high_limit)
1664                return -ENOMEM;
1665        low_limit = info->low_limit + length;
1666
1667        /* Check if rbtree root looks promising */
1668        if (RB_EMPTY_ROOT(&mm->mm_rb))
1669                goto check_highest;
1670        vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1671        if (vma->rb_subtree_gap < length)
1672                goto check_highest;
1673
1674        while (true) {
1675                /* Visit left subtree if it looks promising */
1676                gap_end = vma->vm_start;
1677                if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1678                        struct vm_area_struct *left =
1679                                rb_entry(vma->vm_rb.rb_left,
1680                                         struct vm_area_struct, vm_rb);
1681                        if (left->rb_subtree_gap >= length) {
1682                                vma = left;
1683                                continue;
1684                        }
1685                }
1686
1687                gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1688check_current:
1689                /* Check if current node has a suitable gap */
1690                if (gap_start > high_limit)
1691                        return -ENOMEM;
1692                if (gap_end >= low_limit && gap_end - gap_start >= length)
1693                        goto found;
1694
1695                /* Visit right subtree if it looks promising */
1696                if (vma->vm_rb.rb_right) {
1697                        struct vm_area_struct *right =
1698                                rb_entry(vma->vm_rb.rb_right,
1699                                         struct vm_area_struct, vm_rb);
1700                        if (right->rb_subtree_gap >= length) {
1701                                vma = right;
1702                                continue;
1703                        }
1704                }
1705
1706                /* Go back up the rbtree to find next candidate node */
1707                while (true) {
1708                        struct rb_node *prev = &vma->vm_rb;
1709                        if (!rb_parent(prev))
1710                                goto check_highest;
1711                        vma = rb_entry(rb_parent(prev),
1712                                       struct vm_area_struct, vm_rb);
1713                        if (prev == vma->vm_rb.rb_left) {
1714                                gap_start = vma->vm_prev->vm_end;
1715                                gap_end = vma->vm_start;
1716                                goto check_current;
1717                        }
1718                }
1719        }
1720
1721check_highest:
1722        /* Check highest gap, which does not precede any rbtree node */
1723        gap_start = mm->highest_vm_end;
1724        gap_end = ULONG_MAX;  /* Only for VM_BUG_ON below */
1725        if (gap_start > high_limit)
1726                return -ENOMEM;
1727
1728found:
1729        /* We found a suitable gap. Clip it with the original low_limit. */
1730        if (gap_start < info->low_limit)
1731                gap_start = info->low_limit;
1732
1733        /* Adjust gap address to the desired alignment */
1734        gap_start += (info->align_offset - gap_start) & info->align_mask;
1735
1736        VM_BUG_ON(gap_start + info->length > info->high_limit);
1737        VM_BUG_ON(gap_start + info->length > gap_end);
1738        return gap_start;
1739}
1740
1741unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1742{
1743        struct mm_struct *mm = current->mm;
1744        struct vm_area_struct *vma;
1745        unsigned long length, low_limit, high_limit, gap_start, gap_end;
1746
1747        /* Adjust search length to account for worst case alignment overhead */
1748        length = info->length + info->align_mask;
1749        if (length < info->length)
1750                return -ENOMEM;
1751
1752        /*
1753         * Adjust search limits by the desired length.
1754         * See implementation comment at top of unmapped_area().
1755         */
1756        gap_end = info->high_limit;
1757        if (gap_end < length)
1758                return -ENOMEM;
1759        high_limit = gap_end - length;
1760
1761        if (info->low_limit > high_limit)
1762                return -ENOMEM;
1763        low_limit = info->low_limit + length;
1764
1765        /* Check highest gap, which does not precede any rbtree node */
1766        gap_start = mm->highest_vm_end;
1767        if (gap_start <= high_limit)
1768                goto found_highest;
1769
1770        /* Check if rbtree root looks promising */
1771        if (RB_EMPTY_ROOT(&mm->mm_rb))
1772                return -ENOMEM;
1773        vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1774        if (vma->rb_subtree_gap < length)
1775                return -ENOMEM;
1776
1777        while (true) {
1778                /* Visit right subtree if it looks promising */
1779                gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1780                if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1781                        struct vm_area_struct *right =
1782                                rb_entry(vma->vm_rb.rb_right,
1783                                         struct vm_area_struct, vm_rb);
1784                        if (right->rb_subtree_gap >= length) {
1785                                vma = right;
1786                                continue;
1787                        }
1788                }
1789
1790check_current:
1791                /* Check if current node has a suitable gap */
1792                gap_end = vma->vm_start;
1793                if (gap_end < low_limit)
1794                        return -ENOMEM;
1795                if (gap_start <= high_limit && gap_end - gap_start >= length)
1796                        goto found;
1797
1798                /* Visit left subtree if it looks promising */
1799                if (vma->vm_rb.rb_left) {
1800                        struct vm_area_struct *left =
1801                                rb_entry(vma->vm_rb.rb_left,
1802                                         struct vm_area_struct, vm_rb);
1803                        if (left->rb_subtree_gap >= length) {
1804                                vma = left;
1805                                continue;
1806                        }
1807                }
1808
1809                /* Go back up the rbtree to find next candidate node */
1810                while (true) {
1811                        struct rb_node *prev = &vma->vm_rb;
1812                        if (!rb_parent(prev))
1813                                return -ENOMEM;
1814                        vma = rb_entry(rb_parent(prev),
1815                                       struct vm_area_struct, vm_rb);
1816                        if (prev == vma->vm_rb.rb_right) {
1817                                gap_start = vma->vm_prev ?
1818                                        vma->vm_prev->vm_end : 0;
1819                                goto check_current;
1820                        }
1821                }
1822        }
1823
1824found:
1825        /* We found a suitable gap. Clip it with the original high_limit. */
1826        if (gap_end > info->high_limit)
1827                gap_end = info->high_limit;
1828
1829found_highest:
1830        /* Compute highest gap address at the desired alignment */
1831        gap_end -= info->length;
1832        gap_end -= (gap_end - info->align_offset) & info->align_mask;
1833
1834        VM_BUG_ON(gap_end < info->low_limit);
1835        VM_BUG_ON(gap_end < gap_start);
1836        return gap_end;
1837}
1838
1839/* Get an address range which is currently unmapped.
1840 * For shmat() with addr=0.
1841 *
1842 * Ugly calling convention alert:
1843 * Return value with the low bits set means error value,
1844 * ie
1845 *      if (ret & ~PAGE_MASK)
1846 *              error = ret;
1847 *
1848 * This function "knows" that -ENOMEM has the bits set.
1849 */
1850#ifndef HAVE_ARCH_UNMAPPED_AREA
1851unsigned long
1852arch_get_unmapped_area(struct file *filp, unsigned long addr,
1853                unsigned long len, unsigned long pgoff, unsigned long flags)
1854{
1855        struct mm_struct *mm = current->mm;
1856        struct vm_area_struct *vma;
1857        struct vm_unmapped_area_info info;
1858
1859        if (len > TASK_SIZE)
1860                return -ENOMEM;
1861
1862        if (flags & MAP_FIXED)
1863                return addr;
1864
1865        if (addr) {
1866                addr = PAGE_ALIGN(addr);
1867                vma = find_vma(mm, addr);
1868                if (TASK_SIZE - len >= addr &&
1869                    (!vma || addr + len <= vma->vm_start))
1870                        return addr;
1871        }
1872
1873        info.flags = 0;
1874        info.length = len;
1875        info.low_limit = TASK_UNMAPPED_BASE;
1876        info.high_limit = TASK_SIZE;
1877        info.align_mask = 0;
1878        return vm_unmapped_area(&info);
1879}
1880#endif  
1881
1882/*
1883 * This mmap-allocator allocates new areas top-down from below the
1884 * stack's low limit (the base):
1885 */
1886#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1887unsigned long
1888arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1889                          const unsigned long len, const unsigned long pgoff,
1890                          const unsigned long flags)
1891{
1892        struct vm_area_struct *vma;
1893        struct mm_struct *mm = current->mm;
1894        unsigned long addr = addr0;
1895        struct vm_unmapped_area_info info;
1896
1897        /* requested length too big for entire address space */
1898        if (len > TASK_SIZE)
1899                return -ENOMEM;
1900
1901        if (flags & MAP_FIXED)
1902                return addr;
1903
1904        /* requesting a specific address */
1905        if (addr) {
1906                addr = PAGE_ALIGN(addr);
1907                vma = find_vma(mm, addr);
1908                if (TASK_SIZE - len >= addr &&
1909                                (!vma || addr + len <= vma->vm_start))
1910                        return addr;
1911        }
1912
1913        info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1914        info.length = len;
1915        info.low_limit = PAGE_SIZE;
1916        info.high_limit = mm->mmap_base;
1917        info.align_mask = 0;
1918        addr = vm_unmapped_area(&info);
1919
1920        /*
1921         * A failed mmap() very likely causes application failure,
1922         * so fall back to the bottom-up function here. This scenario
1923         * can happen with large stack limits and large mmap()
1924         * allocations.
1925         */
1926        if (addr & ~PAGE_MASK) {
1927                VM_BUG_ON(addr != -ENOMEM);
1928                info.flags = 0;
1929                info.low_limit = TASK_UNMAPPED_BASE;
1930                info.high_limit = TASK_SIZE;
1931                addr = vm_unmapped_area(&info);
1932        }
1933
1934        return addr;
1935}
1936#endif
1937
1938unsigned long
1939get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1940                unsigned long pgoff, unsigned long flags)
1941{
1942        unsigned long (*get_area)(struct file *, unsigned long,
1943                                  unsigned long, unsigned long, unsigned long);
1944
1945        unsigned long error = arch_mmap_check(addr, len, flags);
1946        if (error)
1947                return error;
1948
1949        /* Careful about overflows.. */
1950        if (len > TASK_SIZE)
1951                return -ENOMEM;
1952
1953        get_area = current->mm->get_unmapped_area;
1954        if (file && file->f_op && file->f_op->get_unmapped_area)
1955                get_area = file->f_op->get_unmapped_area;
1956        addr = get_area(file, addr, len, pgoff, flags);
1957        if (IS_ERR_VALUE(addr))
1958                return addr;
1959
1960        if (addr > TASK_SIZE - len)
1961                return -ENOMEM;
1962        if (addr & ~PAGE_MASK)
1963                return -EINVAL;
1964
1965        addr = arch_rebalance_pgtables(addr, len);
1966        error = security_mmap_addr(addr);
1967        return error ? error : addr;
1968}
1969
1970EXPORT_SYMBOL(get_unmapped_area);
1971
1972/* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
1973struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1974{
1975        struct vm_area_struct *vma = NULL;
1976
1977        /* Check the cache first. */
1978        /* (Cache hit rate is typically around 35%.) */
1979        vma = ACCESS_ONCE(mm->mmap_cache);
1980        if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1981                struct rb_node *rb_node;
1982
1983                rb_node = mm->mm_rb.rb_node;
1984                vma = NULL;
1985
1986                while (rb_node) {
1987                        struct vm_area_struct *vma_tmp;
1988
1989                        vma_tmp = rb_entry(rb_node,
1990                                           struct vm_area_struct, vm_rb);
1991
1992                        if (vma_tmp->vm_end > addr) {
1993                                vma = vma_tmp;
1994                                if (vma_tmp->vm_start <= addr)
1995                                        break;
1996                                rb_node = rb_node->rb_left;
1997                        } else
1998                                rb_node = rb_node->rb_right;
1999                }
2000                if (vma)
2001                        mm->mmap_cache = vma;
2002        }
2003        return vma;
2004}
2005
2006EXPORT_SYMBOL(find_vma);
2007
2008/*
2009 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2010 */
2011struct vm_area_struct *
2012find_vma_prev(struct mm_struct *mm, unsigned long addr,
2013                        struct vm_area_struct **pprev)
2014{
2015        struct vm_area_struct *vma;
2016
2017        vma = find_vma(mm, addr);
2018        if (vma) {
2019                *pprev = vma->vm_prev;
2020        } else {
2021                struct rb_node *rb_node = mm->mm_rb.rb_node;
2022                *pprev = NULL;
2023                while (rb_node) {
2024                        *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2025                        rb_node = rb_node->rb_right;
2026                }
2027        }
2028        return vma;
2029}
2030
2031/*
2032 * Verify that the stack growth is acceptable and
2033 * update accounting. This is shared with both the
2034 * grow-up and grow-down cases.
2035 */
2036static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
2037{
2038        struct mm_struct *mm = vma->vm_mm;
2039        struct rlimit *rlim = current->signal->rlim;
2040        unsigned long new_start;
2041
2042        /* address space limit tests */
2043        if (!may_expand_vm(mm, grow))
2044                return -ENOMEM;
2045
2046        /* Stack limit test */
2047        if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
2048                return -ENOMEM;
2049
2050        /* mlock limit tests */
2051        if (vma->vm_flags & VM_LOCKED) {
2052                unsigned long locked;
2053                unsigned long limit;
2054                locked = mm->locked_vm + grow;
2055                limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2056                limit >>= PAGE_SHIFT;
2057                if (locked > limit && !capable(CAP_IPC_LOCK))
2058                        return -ENOMEM;
2059        }
2060
2061        /* Check to ensure the stack will not grow into a hugetlb-only region */
2062        new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2063                        vma->vm_end - size;
2064        if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2065                return -EFAULT;
2066
2067        /*
2068         * Overcommit..  This must be the final test, as it will
2069         * update security statistics.
2070         */
2071        if (security_vm_enough_memory_mm(mm, grow))
2072                return -ENOMEM;
2073
2074        /* Ok, everything looks good - let it rip */
2075        if (vma->vm_flags & VM_LOCKED)
2076                mm->locked_vm += grow;
2077        vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
2078        return 0;
2079}
2080
2081#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2082/*
2083 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2084 * vma is the last one with address > vma->vm_end.  Have to extend vma.
2085 */
2086int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2087{
2088        int error;
2089
2090        if (!(vma->vm_flags & VM_GROWSUP))
2091                return -EFAULT;
2092
2093        /*
2094         * We must make sure the anon_vma is allocated
2095         * so that the anon_vma locking is not a noop.
2096         */
2097        if (unlikely(anon_vma_prepare(vma)))
2098                return -ENOMEM;
2099        vma_lock_anon_vma(vma);
2100
2101        /*
2102         * vma->vm_start/vm_end cannot change under us because the caller
2103         * is required to hold the mmap_sem in read mode.  We need the
2104         * anon_vma lock to serialize against concurrent expand_stacks.
2105         * Also guard against wrapping around to address 0.
2106         */
2107        if (address < PAGE_ALIGN(address+4))
2108                address = PAGE_ALIGN(address+4);
2109        else {
2110                vma_unlock_anon_vma(vma);
2111                return -ENOMEM;
2112        }
2113        error = 0;
2114
2115        /* Somebody else might have raced and expanded it already */
2116        if (address > vma->vm_end) {
2117                unsigned long size, grow;
2118
2119                size = address - vma->vm_start;
2120                grow = (address - vma->vm_end) >> PAGE_SHIFT;
2121
2122                error = -ENOMEM;
2123                if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2124                        error = acct_stack_growth(vma, size, grow);
2125                        if (!error) {
2126                                /*
2127                                 * vma_gap_update() doesn't support concurrent
2128                                 * updates, but we only hold a shared mmap_sem
2129                                 * lock here, so we need to protect against
2130                                 * concurrent vma expansions.
2131                                 * vma_lock_anon_vma() doesn't help here, as
2132                                 * we don't guarantee that all growable vmas
2133                                 * in a mm share the same root anon vma.
2134                                 * So, we reuse mm->page_table_lock to guard
2135                                 * against concurrent vma expansions.
2136                                 */
2137                                spin_lock(&vma->vm_mm->page_table_lock);
2138                                anon_vma_interval_tree_pre_update_vma(vma);
2139                                vma->vm_end = address;
2140                                anon_vma_interval_tree_post_update_vma(vma);
2141                                if (vma->vm_next)
2142                                        vma_gap_update(vma->vm_next);
2143                                else
2144                                        vma->vm_mm->highest_vm_end = address;
2145                                spin_unlock(&vma->vm_mm->page_table_lock);
2146
2147                                perf_event_mmap(vma);
2148                        }
2149                }
2150        }
2151        vma_unlock_anon_vma(vma);
2152        khugepaged_enter_vma_merge(vma);
2153        validate_mm(vma->vm_mm);
2154        return error;
2155}
2156#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2157
2158/*
2159 * vma is the first one with address < vma->vm_start.  Have to extend vma.
2160 */
2161int expand_downwards(struct vm_area_struct *vma,
2162                                   unsigned long address)
2163{
2164        int error;
2165
2166        /*
2167         * We must make sure the anon_vma is allocated
2168         * so that the anon_vma locking is not a noop.
2169         */
2170        if (unlikely(anon_vma_prepare(vma)))
2171                return -ENOMEM;
2172
2173        address &= PAGE_MASK;
2174        error = security_mmap_addr(address);
2175        if (error)
2176                return error;
2177
2178        vma_lock_anon_vma(vma);
2179
2180        /*
2181         * vma->vm_start/vm_end cannot change under us because the caller
2182         * is required to hold the mmap_sem in read mode.  We need the
2183         * anon_vma lock to serialize against concurrent expand_stacks.
2184         */
2185
2186        /* Somebody else might have raced and expanded it already */
2187        if (address < vma->vm_start) {
2188                unsigned long size, grow;
2189
2190                size = vma->vm_end - address;
2191                grow = (vma->vm_start - address) >> PAGE_SHIFT;
2192
2193                error = -ENOMEM;
2194                if (grow <= vma->vm_pgoff) {
2195                        error = acct_stack_growth(vma, size, grow);
2196                        if (!error) {
2197                                /*
2198                                 * vma_gap_update() doesn't support concurrent
2199                                 * updates, but we only hold a shared mmap_sem
2200                                 * lock here, so we need to protect against
2201                                 * concurrent vma expansions.
2202                                 * vma_lock_anon_vma() doesn't help here, as
2203                                 * we don't guarantee that all growable vmas
2204                                 * in a mm share the same root anon vma.
2205                                 * So, we reuse mm->page_table_lock to guard
2206                                 * against concurrent vma expansions.
2207                                 */
2208                                spin_lock(&vma->vm_mm->page_table_lock);
2209                                anon_vma_interval_tree_pre_update_vma(vma);
2210                                vma->vm_start = address;
2211                                vma->vm_pgoff -= grow;
2212                                anon_vma_interval_tree_post_update_vma(vma);
2213                                vma_gap_update(vma);
2214                                spin_unlock(&vma->vm_mm->page_table_lock);
2215
2216                                perf_event_mmap(vma);
2217                        }
2218                }
2219        }
2220        vma_unlock_anon_vma(vma);
2221        khugepaged_enter_vma_merge(vma);
2222        validate_mm(vma->vm_mm);
2223        return error;
2224}
2225
2226/*
2227 * Note how expand_stack() refuses to expand the stack all the way to
2228 * abut the next virtual mapping, *unless* that mapping itself is also
2229 * a stack mapping. We want to leave room for a guard page, after all
2230 * (the guard page itself is not added here, that is done by the
2231 * actual page faulting logic)
2232 *
2233 * This matches the behavior of the guard page logic (see mm/memory.c:
2234 * check_stack_guard_page()), which only allows the guard page to be
2235 * removed under these circumstances.
2236 */
2237#ifdef CONFIG_STACK_GROWSUP
2238int expand_stack(struct vm_area_struct *vma, unsigned long address)
2239{
2240        struct vm_area_struct *next;
2241
2242        address &= PAGE_MASK;
2243        next = vma->vm_next;
2244        if (next && next->vm_start == address + PAGE_SIZE) {
2245                if (!(next->vm_flags & VM_GROWSUP))
2246                        return -ENOMEM;
2247        }
2248        return expand_upwards(vma, address);
2249}
2250
2251struct vm_area_struct *
2252find_extend_vma(struct mm_struct *mm, unsigned long addr)
2253{
2254        struct vm_area_struct *vma, *prev;
2255
2256        addr &= PAGE_MASK;
2257        vma = find_vma_prev(mm, addr, &prev);
2258        if (vma && (vma->vm_start <= addr))
2259                return vma;
2260        if (!prev || expand_stack(prev, addr))
2261                return NULL;
2262        if (prev->vm_flags & VM_LOCKED)
2263                __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL);
2264        return prev;
2265}
2266#else
2267int expand_stack(struct vm_area_struct *vma, unsigned long address)
2268{
2269        struct vm_area_struct *prev;
2270
2271        address &= PAGE_MASK;
2272        prev = vma->vm_prev;
2273        if (prev && prev->vm_end == address) {
2274                if (!(prev->vm_flags & VM_GROWSDOWN))
2275                        return -ENOMEM;
2276        }
2277        return expand_downwards(vma, address);
2278}
2279
2280struct vm_area_struct *
2281find_extend_vma(struct mm_struct * mm, unsigned long addr)
2282{
2283        struct vm_area_struct * vma;
2284        unsigned long start;
2285
2286        addr &= PAGE_MASK;
2287        vma = find_vma(mm,addr);
2288        if (!vma)
2289                return NULL;
2290        if (vma->vm_start <= addr)
2291                return vma;
2292        if (!(vma->vm_flags & VM_GROWSDOWN))
2293                return NULL;
2294        start = vma->vm_start;
2295        if (expand_stack(vma, addr))
2296                return NULL;
2297        if (vma->vm_flags & VM_LOCKED)
2298                __mlock_vma_pages_range(vma, addr, start, NULL);
2299        return vma;
2300}
2301#endif
2302
2303/*
2304 * Ok - we have the memory areas we should free on the vma list,
2305 * so release them, and do the vma updates.
2306 *
2307 * Called with the mm semaphore held.
2308 */
2309static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2310{
2311        unsigned long nr_accounted = 0;
2312
2313        /* Update high watermark before we lower total_vm */
2314        update_hiwater_vm(mm);
2315        do {
2316                long nrpages = vma_pages(vma);
2317
2318                if (vma->vm_flags & VM_ACCOUNT)
2319                        nr_accounted += nrpages;
2320                vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
2321                vma = remove_vma(vma);
2322        } while (vma);
2323        vm_unacct_memory(nr_accounted);
2324        validate_mm(mm);
2325}
2326
2327/*
2328 * Get rid of page table information in the indicated region.
2329 *
2330 * Called with the mm semaphore held.
2331 */
2332static void unmap_region(struct mm_struct *mm,
2333                struct vm_area_struct *vma, struct vm_area_struct *prev,
2334                unsigned long start, unsigned long end)
2335{
2336        struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
2337        struct mmu_gather tlb;
2338
2339        lru_add_drain();
2340        tlb_gather_mmu(&tlb, mm, start, end);
2341        update_hiwater_rss(mm);
2342        unmap_vmas(&tlb, vma, start, end);
2343        free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2344                                 next ? next->vm_start : USER_PGTABLES_CEILING);
2345        tlb_finish_mmu(&tlb, start, end);
2346}
2347
2348/*
2349 * Create a list of vma's touched by the unmap, removing them from the mm's
2350 * vma list as we go..
2351 */
2352static void
2353detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2354        struct vm_area_struct *prev, unsigned long end)
2355{
2356        struct vm_area_struct **insertion_point;
2357        struct vm_area_struct *tail_vma = NULL;
2358
2359        insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2360        vma->vm_prev = NULL;
2361        do {
2362                vma_rb_erase(vma, &mm->mm_rb);
2363                mm->map_count--;
2364                tail_vma = vma;
2365                vma = vma->vm_next;
2366        } while (vma && vma->vm_start < end);
2367        *insertion_point = vma;
2368        if (vma) {
2369                vma->vm_prev = prev;
2370                vma_gap_update(vma);
2371        } else
2372                mm->highest_vm_end = prev ? prev->vm_end : 0;
2373        tail_vma->vm_next = NULL;
2374        mm->mmap_cache = NULL;          /* Kill the cache. */
2375}
2376
2377/*
2378 * __split_vma() bypasses sysctl_max_map_count checking.  We use this on the
2379 * munmap path where it doesn't make sense to fail.
2380 */
2381static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
2382              unsigned long addr, int new_below)
2383{
2384        struct vm_area_struct *new;
2385        int err = -ENOMEM;
2386
2387        if (is_vm_hugetlb_page(vma) && (addr &
2388                                        ~(huge_page_mask(hstate_vma(vma)))))
2389                return -EINVAL;
2390
2391        new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2392        if (!new)
2393                goto out_err;
2394
2395        /* most fields are the same, copy all, and then fixup */
2396        *new = *vma;
2397
2398        INIT_LIST_HEAD(&new->anon_vma_chain);
2399
2400        if (new_below)
2401                new->vm_end = addr;
2402        else {
2403                new->vm_start = addr;
2404                new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2405        }
2406
2407        err = vma_dup_policy(vma, new);
2408        if (err)
2409                goto out_free_vma;
2410
2411        if (anon_vma_clone(new, vma))
2412                goto out_free_mpol;
2413
2414        if (new->vm_file)
2415                get_file(new->vm_file);
2416
2417        if (new->vm_ops && new->vm_ops->open)
2418                new->vm_ops->open(new);
2419
2420        if (new_below)
2421                err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2422                        ((addr - new->vm_start) >> PAGE_SHIFT), new);
2423        else
2424                err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2425
2426        /* Success. */
2427        if (!err)
2428                return 0;
2429
2430        /* Clean everything up if vma_adjust failed. */
2431        if (new->vm_ops && new->vm_ops->close)
2432                new->vm_ops->close(new);
2433        if (new->vm_file)
2434                fput(new->vm_file);
2435        unlink_anon_vmas(new);
2436 out_free_mpol:
2437        mpol_put(vma_policy(new));
2438 out_free_vma:
2439        kmem_cache_free(vm_area_cachep, new);
2440 out_err:
2441        return err;
2442}
2443
2444/*
2445 * Split a vma into two pieces at address 'addr', a new vma is allocated
2446 * either for the first part or the tail.
2447 */
2448int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2449              unsigned long addr, int new_below)
2450{
2451        if (mm->map_count >= sysctl_max_map_count)
2452                return -ENOMEM;
2453
2454        return __split_vma(mm, vma, addr, new_below);
2455}
2456
2457/* Munmap is split into 2 main parts -- this part which finds
2458 * what needs doing, and the areas themselves, which do the
2459 * work.  This now handles partial unmappings.
2460 * Jeremy Fitzhardinge <jeremy@goop.org>
2461 */
2462int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2463{
2464        unsigned long end;
2465        struct vm_area_struct *vma, *prev, *last;
2466
2467        if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2468                return -EINVAL;
2469
2470        if ((len = PAGE_ALIGN(len)) == 0)
2471                return -EINVAL;
2472
2473        /* Find the first overlapping VMA */
2474        vma = find_vma(mm, start);
2475        if (!vma)
2476                return 0;
2477        prev = vma->vm_prev;
2478        /* we have  start < vma->vm_end  */
2479
2480        /* if it doesn't overlap, we have nothing.. */
2481        end = start + len;
2482        if (vma->vm_start >= end)
2483                return 0;
2484
2485        /*
2486         * If we need to split any vma, do it now to save pain later.
2487         *
2488         * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2489         * unmapped vm_area_struct will remain in use: so lower split_vma
2490         * places tmp vma above, and higher split_vma places tmp vma below.
2491         */
2492        if (start > vma->vm_start) {
2493                int error;
2494
2495                /*
2496                 * Make sure that map_count on return from munmap() will
2497                 * not exceed its limit; but let map_count go just above
2498                 * its limit temporarily, to help free resources as expected.
2499                 */
2500                if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2501                        return -ENOMEM;
2502
2503                error = __split_vma(mm, vma, start, 0);
2504                if (error)
2505                        return error;
2506                prev = vma;
2507        }
2508
2509        /* Does it split the last one? */
2510        last = find_vma(mm, end);
2511        if (last && end > last->vm_start) {
2512                int error = __split_vma(mm, last, end, 1);
2513                if (error)
2514                        return error;
2515        }
2516        vma = prev? prev->vm_next: mm->mmap;
2517
2518        /*
2519         * unlock any mlock()ed ranges before detaching vmas
2520         */
2521        if (mm->locked_vm) {
2522                struct vm_area_struct *tmp = vma;
2523                while (tmp && tmp->vm_start < end) {
2524                        if (tmp->vm_flags & VM_LOCKED) {
2525                                mm->locked_vm -= vma_pages(tmp);
2526                                munlock_vma_pages_all(tmp);
2527                        }
2528                        tmp = tmp->vm_next;
2529                }
2530        }
2531
2532        /*
2533         * Remove the vma's, and unmap the actual pages
2534         */
2535        detach_vmas_to_be_unmapped(mm, vma, prev, end);
2536        unmap_region(mm, vma, prev, start, end);
2537
2538        /* Fix up all other VM information */
2539        remove_vma_list(mm, vma);
2540
2541        return 0;
2542}
2543
2544int vm_munmap(unsigned long start, size_t len)
2545{
2546        int ret;
2547        struct mm_struct *mm = current->mm;
2548
2549        down_write(&mm->mmap_sem);
2550        ret = do_munmap(mm, start, len);
2551        up_write(&mm->mmap_sem);
2552        return ret;
2553}
2554EXPORT_SYMBOL(vm_munmap);
2555
2556SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2557{
2558        profile_munmap(addr);
2559        return vm_munmap(addr, len);
2560}
2561
2562static inline void verify_mm_writelocked(struct mm_struct *mm)
2563{
2564#ifdef CONFIG_DEBUG_VM
2565        if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2566                WARN_ON(1);
2567                up_read(&mm->mmap_sem);
2568        }
2569#endif
2570}
2571
2572/*
2573 *  this is really a simplified "do_mmap".  it only handles
2574 *  anonymous maps.  eventually we may be able to do some
2575 *  brk-specific accounting here.
2576 */
2577static unsigned long do_brk(unsigned long addr, unsigned long len)
2578{
2579        struct mm_struct * mm = current->mm;
2580        struct vm_area_struct * vma, * prev;
2581        unsigned long flags;
2582        struct rb_node ** rb_link, * rb_parent;
2583        pgoff_t pgoff = addr >> PAGE_SHIFT;
2584        int error;
2585
2586        len = PAGE_ALIGN(len);
2587        if (!len)
2588                return addr;
2589
2590        flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2591
2592        error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2593        if (error & ~PAGE_MASK)
2594                return error;
2595
2596        /*
2597         * mlock MCL_FUTURE?
2598         */
2599        if (mm->def_flags & VM_LOCKED) {
2600                unsigned long locked, lock_limit;
2601                locked = len >> PAGE_SHIFT;
2602                locked += mm->locked_vm;
2603                lock_limit = rlimit(RLIMIT_MEMLOCK);
2604                lock_limit >>= PAGE_SHIFT;
2605                if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2606                        return -EAGAIN;
2607        }
2608
2609        /*
2610         * mm->mmap_sem is required to protect against another thread
2611         * changing the mappings in case we sleep.
2612         */
2613        verify_mm_writelocked(mm);
2614
2615        /*
2616         * Clear old maps.  this also does some error checking for us
2617         */
2618 munmap_back:
2619        if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) {
2620                if (do_munmap(mm, addr, len))
2621                        return -ENOMEM;
2622                goto munmap_back;
2623        }
2624
2625        /* Check against address space limits *after* clearing old maps... */
2626        if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2627                return -ENOMEM;
2628
2629        if (mm->map_count > sysctl_max_map_count)
2630                return -ENOMEM;
2631
2632        if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2633                return -ENOMEM;
2634
2635        /* Can we just expand an old private anonymous mapping? */
2636        vma = vma_merge(mm, prev, addr, addr + len, flags,
2637                                        NULL, NULL, pgoff, NULL);
2638        if (vma)
2639                goto out;
2640
2641        /*
2642         * create a vma struct for an anonymous mapping
2643         */
2644        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2645        if (!vma) {
2646                vm_unacct_memory(len >> PAGE_SHIFT);
2647                return -ENOMEM;
2648        }
2649
2650        INIT_LIST_HEAD(&vma->anon_vma_chain);
2651        vma->vm_mm = mm;
2652        vma->vm_start = addr;
2653        vma->vm_end = addr + len;
2654        vma->vm_pgoff = pgoff;
2655        vma->vm_flags = flags;
2656        vma->vm_page_prot = vm_get_page_prot(flags);
2657        vma_link(mm, vma, prev, rb_link, rb_parent);
2658out:
2659        perf_event_mmap(vma);
2660        mm->total_vm += len >> PAGE_SHIFT;
2661        if (flags & VM_LOCKED)
2662                mm->locked_vm += (len >> PAGE_SHIFT);
2663        vma->vm_flags |= VM_SOFTDIRTY;
2664        return addr;
2665}
2666
2667unsigned long vm_brk(unsigned long addr, unsigned long len)
2668{
2669        struct mm_struct *mm = current->mm;
2670        unsigned long ret;
2671        bool populate;
2672
2673        down_write(&mm->mmap_sem);
2674        ret = do_brk(addr, len);
2675        populate = ((mm->def_flags & VM_LOCKED) != 0);
2676        up_write(&mm->mmap_sem);
2677        if (populate)
2678                mm_populate(addr, len);
2679        return ret;
2680}
2681EXPORT_SYMBOL(vm_brk);
2682
2683/* Release all mmaps. */
2684void exit_mmap(struct mm_struct *mm)
2685{
2686        struct mmu_gather tlb;
2687        struct vm_area_struct *vma;
2688        unsigned long nr_accounted = 0;
2689
2690        /* mm's last user has gone, and its about to be pulled down */
2691        mmu_notifier_release(mm);
2692
2693        if (mm->locked_vm) {
2694                vma = mm->mmap;
2695                while (vma) {
2696                        if (vma->vm_flags & VM_LOCKED)
2697                                munlock_vma_pages_all(vma);
2698                        vma = vma->vm_next;
2699                }
2700        }
2701
2702        arch_exit_mmap(mm);
2703
2704        vma = mm->mmap;
2705        if (!vma)       /* Can happen if dup_mmap() received an OOM */
2706                return;
2707
2708        lru_add_drain();
2709        flush_cache_mm(mm);
2710        tlb_gather_mmu(&tlb, mm, 0, -1);
2711        /* update_hiwater_rss(mm) here? but nobody should be looking */
2712        /* Use -1 here to ensure all VMAs in the mm are unmapped */
2713        unmap_vmas(&tlb, vma, 0, -1);
2714
2715        free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
2716        tlb_finish_mmu(&tlb, 0, -1);
2717
2718        /*
2719         * Walk the list again, actually closing and freeing it,
2720         * with preemption enabled, without holding any MM locks.
2721         */
2722        while (vma) {
2723                if (vma->vm_flags & VM_ACCOUNT)
2724                        nr_accounted += vma_pages(vma);
2725                vma = remove_vma(vma);
2726        }
2727        vm_unacct_memory(nr_accounted);
2728
2729        WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2730}
2731
2732/* Insert vm structure into process list sorted by address
2733 * and into the inode's i_mmap tree.  If vm_file is non-NULL
2734 * then i_mmap_mutex is taken here.
2735 */
2736int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2737{
2738        struct vm_area_struct *prev;
2739        struct rb_node **rb_link, *rb_parent;
2740
2741        /*
2742         * The vm_pgoff of a purely anonymous vma should be irrelevant
2743         * until its first write fault, when page's anon_vma and index
2744         * are set.  But now set the vm_pgoff it will almost certainly
2745         * end up with (unless mremap moves it elsewhere before that
2746         * first wfault), so /proc/pid/maps tells a consistent story.
2747         *
2748         * By setting it to reflect the virtual start address of the
2749         * vma, merges and splits can happen in a seamless way, just
2750         * using the existing file pgoff checks and manipulations.
2751         * Similarly in do_mmap_pgoff and in do_brk.
2752         */
2753        if (!vma->vm_file) {
2754                BUG_ON(vma->anon_vma);
2755                vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2756        }
2757        if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2758                           &prev, &rb_link, &rb_parent))
2759                return -ENOMEM;
2760        if ((vma->vm_flags & VM_ACCOUNT) &&
2761             security_vm_enough_memory_mm(mm, vma_pages(vma)))
2762                return -ENOMEM;
2763
2764        vma_link(mm, vma, prev, rb_link, rb_parent);
2765        return 0;
2766}
2767
2768/*
2769 * Copy the vma structure to a new location in the same mm,
2770 * prior to moving page table entries, to effect an mremap move.
2771 */
2772struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2773        unsigned long addr, unsigned long len, pgoff_t pgoff,
2774        bool *need_rmap_locks)
2775{
2776        struct vm_area_struct *vma = *vmap;
2777        unsigned long vma_start = vma->vm_start;
2778        struct mm_struct *mm = vma->vm_mm;
2779        struct vm_area_struct *new_vma, *prev;
2780        struct rb_node **rb_link, *rb_parent;
2781        bool faulted_in_anon_vma = true;
2782
2783        /*
2784         * If anonymous vma has not yet been faulted, update new pgoff
2785         * to match new location, to increase its chance of merging.
2786         */
2787        if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2788                pgoff = addr >> PAGE_SHIFT;
2789                faulted_in_anon_vma = false;
2790        }
2791
2792        if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2793                return NULL;    /* should never get here */
2794        new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2795                        vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2796        if (new_vma) {
2797                /*
2798                 * Source vma may have been merged into new_vma
2799                 */
2800                if (unlikely(vma_start >= new_vma->vm_start &&
2801                             vma_start < new_vma->vm_end)) {
2802                        /*
2803                         * The only way we can get a vma_merge with
2804                         * self during an mremap is if the vma hasn't
2805                         * been faulted in yet and we were allowed to
2806                         * reset the dst vma->vm_pgoff to the
2807                         * destination address of the mremap to allow
2808                         * the merge to happen. mremap must change the
2809                         * vm_pgoff linearity between src and dst vmas
2810                         * (in turn preventing a vma_merge) to be
2811                         * safe. It is only safe to keep the vm_pgoff
2812                         * linear if there are no pages mapped yet.
2813                         */
2814                        VM_BUG_ON(faulted_in_anon_vma);
2815                        *vmap = vma = new_vma;
2816                }
2817                *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2818        } else {
2819                new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2820                if (new_vma) {
2821                        *new_vma = *vma;
2822                        new_vma->vm_start = addr;
2823                        new_vma->vm_end = addr + len;
2824                        new_vma->vm_pgoff = pgoff;
2825                        if (vma_dup_policy(vma, new_vma))
2826                                goto out_free_vma;
2827                        INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2828                        if (anon_vma_clone(new_vma, vma))
2829                                goto out_free_mempol;
2830                        if (new_vma->vm_file)
2831                                get_file(new_vma->vm_file);
2832                        if (new_vma->vm_ops && new_vma->vm_ops->open)
2833                                new_vma->vm_ops->open(new_vma);
2834                        vma_link(mm, new_vma, prev, rb_link, rb_parent);
2835                        *need_rmap_locks = false;
2836                }
2837        }
2838        return new_vma;
2839
2840 out_free_mempol:
2841        mpol_put(vma_policy(new_vma));
2842 out_free_vma:
2843        kmem_cache_free(vm_area_cachep, new_vma);
2844        return NULL;
2845}
2846
2847/*
2848 * Return true if the calling process may expand its vm space by the passed
2849 * number of pages
2850 */
2851int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2852{
2853        unsigned long cur = mm->total_vm;       /* pages */
2854        unsigned long lim;
2855
2856        lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2857
2858        if (cur + npages > lim)
2859                return 0;
2860        return 1;
2861}
2862
2863
2864static int special_mapping_fault(struct vm_area_struct *vma,
2865                                struct vm_fault *vmf)
2866{
2867        pgoff_t pgoff;
2868        struct page **pages;
2869
2870        /*
2871         * special mappings have no vm_file, and in that case, the mm
2872         * uses vm_pgoff internally. So we have to subtract it from here.
2873         * We are allowed to do this because we are the mm; do not copy
2874         * this code into drivers!
2875         */
2876        pgoff = vmf->pgoff - vma->vm_pgoff;
2877
2878        for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2879                pgoff--;
2880
2881        if (*pages) {
2882                struct page *page = *pages;
2883                get_page(page);
2884                vmf->page = page;
2885                return 0;
2886        }
2887
2888        return VM_FAULT_SIGBUS;
2889}
2890
2891/*
2892 * Having a close hook prevents vma merging regardless of flags.
2893 */
2894static void special_mapping_close(struct vm_area_struct *vma)
2895{
2896}
2897
2898static const struct vm_operations_struct special_mapping_vmops = {
2899        .close = special_mapping_close,
2900        .fault = special_mapping_fault,
2901};
2902
2903/*
2904 * Called with mm->mmap_sem held for writing.
2905 * Insert a new vma covering the given region, with the given flags.
2906 * Its pages are supplied by the given array of struct page *.
2907 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2908 * The region past the last page supplied will always produce SIGBUS.
2909 * The array pointer and the pages it points to are assumed to stay alive
2910 * for as long as this mapping might exist.
2911 */
2912int install_special_mapping(struct mm_struct *mm,
2913                            unsigned long addr, unsigned long len,
2914                            unsigned long vm_flags, struct page **pages)
2915{
2916        int ret;
2917        struct vm_area_struct *vma;
2918
2919        vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2920        if (unlikely(vma == NULL))
2921                return -ENOMEM;
2922
2923        INIT_LIST_HEAD(&vma->anon_vma_chain);
2924        vma->vm_mm = mm;
2925        vma->vm_start = addr;
2926        vma->vm_end = addr + len;
2927
2928        vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
2929        vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2930
2931        vma->vm_ops = &special_mapping_vmops;
2932        vma->vm_private_data = pages;
2933
2934        ret = insert_vm_struct(mm, vma);
2935        if (ret)
2936                goto out;
2937
2938        mm->total_vm += len >> PAGE_SHIFT;
2939
2940        perf_event_mmap(vma);
2941
2942        return 0;
2943
2944out:
2945        kmem_cache_free(vm_area_cachep, vma);
2946        return ret;
2947}
2948
2949static DEFINE_MUTEX(mm_all_locks_mutex);
2950
2951static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2952{
2953        if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
2954                /*
2955                 * The LSB of head.next can't change from under us
2956                 * because we hold the mm_all_locks_mutex.
2957                 */
2958                down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
2959                /*
2960                 * We can safely modify head.next after taking the
2961                 * anon_vma->root->rwsem. If some other vma in this mm shares
2962                 * the same anon_vma we won't take it again.
2963                 *
2964                 * No need of atomic instructions here, head.next
2965                 * can't change from under us thanks to the
2966                 * anon_vma->root->rwsem.
2967                 */
2968                if (__test_and_set_bit(0, (unsigned long *)
2969                                       &anon_vma->root->rb_root.rb_node))
2970                        BUG();
2971        }
2972}
2973
2974static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2975{
2976        if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2977                /*
2978                 * AS_MM_ALL_LOCKS can't change from under us because
2979                 * we hold the mm_all_locks_mutex.
2980                 *
2981                 * Operations on ->flags have to be atomic because
2982                 * even if AS_MM_ALL_LOCKS is stable thanks to the
2983                 * mm_all_locks_mutex, there may be other cpus
2984                 * changing other bitflags in parallel to us.
2985                 */
2986                if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2987                        BUG();
2988                mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2989        }
2990}
2991
2992/*
2993 * This operation locks against the VM for all pte/vma/mm related
2994 * operations that could ever happen on a certain mm. This includes
2995 * vmtruncate, try_to_unmap, and all page faults.
2996 *
2997 * The caller must take the mmap_sem in write mode before calling
2998 * mm_take_all_locks(). The caller isn't allowed to release the
2999 * mmap_sem until mm_drop_all_locks() returns.
3000 *
3001 * mmap_sem in write mode is required in order to block all operations
3002 * that could modify pagetables and free pages without need of
3003 * altering the vma layout (for example populate_range() with
3004 * nonlinear vmas). It's also needed in write mode to avoid new
3005 * anon_vmas to be associated with existing vmas.
3006 *
3007 * A single task can't take more than one mm_take_all_locks() in a row
3008 * or it would deadlock.
3009 *
3010 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3011 * mapping->flags avoid to take the same lock twice, if more than one
3012 * vma in this mm is backed by the same anon_vma or address_space.
3013 *
3014 * We can take all the locks in random order because the VM code
3015 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3016 * takes more than one of them in a row. Secondly we're protected
3017 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3018 *
3019 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3020 * that may have to take thousand of locks.
3021 *
3022 * mm_take_all_locks() can fail if it's interrupted by signals.
3023 */
3024int mm_take_all_locks(struct mm_struct *mm)
3025{
3026        struct vm_area_struct *vma;
3027        struct anon_vma_chain *avc;
3028
3029        BUG_ON(down_read_trylock(&mm->mmap_sem));
3030
3031        mutex_lock(&mm_all_locks_mutex);
3032
3033        for (vma = mm->mmap; vma; vma = vma->vm_next) {
3034                if (signal_pending(current))
3035                        goto out_unlock;
3036                if (vma->vm_file && vma->vm_file->f_mapping)
3037                        vm_lock_mapping(mm, vma->vm_file->f_mapping);
3038        }
3039
3040        for (vma = mm->mmap; vma; vma = vma->vm_next) {
3041                if (signal_pending(current))
3042                        goto out_unlock;
3043                if (vma->anon_vma)
3044                        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3045                                vm_lock_anon_vma(mm, avc->anon_vma);
3046        }
3047
3048        return 0;
3049
3050out_unlock:
3051        mm_drop_all_locks(mm);
3052        return -EINTR;
3053}
3054
3055static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3056{
3057        if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3058                /*
3059                 * The LSB of head.next can't change to 0 from under
3060                 * us because we hold the mm_all_locks_mutex.
3061                 *
3062                 * We must however clear the bitflag before unlocking
3063                 * the vma so the users using the anon_vma->rb_root will
3064                 * never see our bitflag.
3065                 *
3066                 * No need of atomic instructions here, head.next
3067                 * can't change from under us until we release the
3068                 * anon_vma->root->rwsem.
3069                 */
3070                if (!__test_and_clear_bit(0, (unsigned long *)
3071                                          &anon_vma->root->rb_root.rb_node))
3072                        BUG();
3073                anon_vma_unlock_write(anon_vma);
3074        }
3075}
3076
3077static void vm_unlock_mapping(struct address_space *mapping)
3078{
3079        if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3080                /*
3081                 * AS_MM_ALL_LOCKS can't change to 0 from under us
3082                 * because we hold the mm_all_locks_mutex.
3083                 */
3084                mutex_unlock(&mapping->i_mmap_mutex);
3085                if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3086                                        &mapping->flags))
3087                        BUG();
3088        }
3089}
3090
3091/*
3092 * The mmap_sem cannot be released by the caller until
3093 * mm_drop_all_locks() returns.
3094 */
3095void mm_drop_all_locks(struct mm_struct *mm)
3096{
3097        struct vm_area_struct *vma;
3098        struct anon_vma_chain *avc;
3099
3100        BUG_ON(down_read_trylock(&mm->mmap_sem));
3101        BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3102
3103        for (vma = mm->mmap; vma; vma = vma->vm_next) {
3104                if (vma->anon_vma)
3105                        list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3106                                vm_unlock_anon_vma(avc->anon_vma);
3107                if (vma->vm_file && vma->vm_file->f_mapping)
3108                        vm_unlock_mapping(vma->vm_file->f_mapping);
3109        }
3110
3111        mutex_unlock(&mm_all_locks_mutex);
3112}
3113
3114/*
3115 * initialise the VMA slab
3116 */
3117void __init mmap_init(void)
3118{
3119        int ret;
3120
3121        ret = percpu_counter_init(&vm_committed_as, 0);
3122        VM_BUG_ON(ret);
3123}
3124
3125/*
3126 * Initialise sysctl_user_reserve_kbytes.
3127 *
3128 * This is intended to prevent a user from starting a single memory hogging
3129 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3130 * mode.
3131 *
3132 * The default value is min(3% of free memory, 128MB)
3133 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3134 */
3135static int init_user_reserve(void)
3136{
3137        unsigned long free_kbytes;
3138
3139        free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3140
3141        sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3142        return 0;
3143}
3144module_init(init_user_reserve)
3145
3146/*
3147 * Initialise sysctl_admin_reserve_kbytes.
3148 *
3149 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3150 * to log in and kill a memory hogging process.
3151 *
3152 * Systems with more than 256MB will reserve 8MB, enough to recover
3153 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3154 * only reserve 3% of free pages by default.
3155 */
3156static int init_admin_reserve(void)
3157{
3158        unsigned long free_kbytes;
3159
3160        free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3161
3162        sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3163        return 0;
3164}
3165module_init(init_admin_reserve)
3166
3167/*
3168 * Reinititalise user and admin reserves if memory is added or removed.
3169 *
3170 * The default user reserve max is 128MB, and the default max for the
3171 * admin reserve is 8MB. These are usually, but not always, enough to
3172 * enable recovery from a memory hogging process using login/sshd, a shell,
3173 * and tools like top. It may make sense to increase or even disable the
3174 * reserve depending on the existence of swap or variations in the recovery
3175 * tools. So, the admin may have changed them.
3176 *
3177 * If memory is added and the reserves have been eliminated or increased above
3178 * the default max, then we'll trust the admin.
3179 *
3180 * If memory is removed and there isn't enough free memory, then we
3181 * need to reset the reserves.
3182 *
3183 * Otherwise keep the reserve set by the admin.
3184 */
3185static int reserve_mem_notifier(struct notifier_block *nb,
3186                             unsigned long action, void *data)
3187{
3188        unsigned long tmp, free_kbytes;
3189
3190        switch (action) {
3191        case MEM_ONLINE:
3192                /* Default max is 128MB. Leave alone if modified by operator. */
3193                tmp = sysctl_user_reserve_kbytes;
3194                if (0 < tmp && tmp < (1UL << 17))
3195                        init_user_reserve();
3196
3197                /* Default max is 8MB.  Leave alone if modified by operator. */
3198                tmp = sysctl_admin_reserve_kbytes;
3199                if (0 < tmp && tmp < (1UL << 13))
3200                        init_admin_reserve();
3201
3202                break;
3203        case MEM_OFFLINE:
3204                free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3205
3206                if (sysctl_user_reserve_kbytes > free_kbytes) {
3207                        init_user_reserve();
3208                        pr_info("vm.user_reserve_kbytes reset to %lu\n",
3209                                sysctl_user_reserve_kbytes);
3210                }
3211
3212                if (sysctl_admin_reserve_kbytes > free_kbytes) {
3213                        init_admin_reserve();
3214                        pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3215                                sysctl_admin_reserve_kbytes);
3216                }
3217                break;
3218        default:
3219                break;
3220        }
3221        return NOTIFY_OK;
3222}
3223
3224static struct notifier_block reserve_mem_nb = {
3225        .notifier_call = reserve_mem_notifier,
3226};
3227
3228static int __meminit init_reserve_notifier(void)
3229{
3230        if (register_hotmemory_notifier(&reserve_mem_nb))
3231                printk("Failed registering memory add/remove notifier for admin reserve");
3232
3233        return 0;
3234}
3235module_init(init_reserve_notifier)
3236
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.