mm/nommu.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/mm/nommu.c
   4  *
   5  *  Replacement code for mm functions to support CPU's that don't
   6  *  have any form of memory management unit (thus no virtual memory).
   7  *
   8  *  See Documentation/admin-guide/mm/nommu-mmap.rst
   9  *
  10  *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
  11  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  12  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  13  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  14  *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
  15  */
  16
  17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  18
  19 #include <linux/export.h>
  20 #include <linux/mm.h>
  21 #include <linux/sched/mm.h>
  22 #include <linux/vmacache.h>
  23 #include <linux/mman.h>
  24 #include <linux/swap.h>
  25 #include <linux/file.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/slab.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/blkdev.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/compiler.h>
  33 #include <linux/mount.h>
  34 #include <linux/personality.h>
  35 #include <linux/security.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/audit.h>
  38 #include <linux/printk.h>
  39
  40 #include <linux/uaccess.h>
  41 #include <asm/tlb.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/mmu_context.h>
  44 #include "internal.h"
  45
  46 void *high_memory;
  47 EXPORT_SYMBOL(high_memory);
  48 struct page *mem_map;
  49 unsigned long max_mapnr;
  50 EXPORT_SYMBOL(max_mapnr);
  51 unsigned long highest_memmap_pfn;
  52 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
  53 int heap_stack_gap = 0;
  54
  55 atomic_long_t mmap_pages_allocated;
  56
  57 EXPORT_SYMBOL(mem_map);
  58
  59 /* list of mapped, potentially shareable regions */
  60 static struct kmem_cache *vm_region_jar;
  61 struct rb_root nommu_region_tree = RB_ROOT;
  62 DECLARE_RWSEM(nommu_region_sem);
  63
  64 const struct vm_operations_struct generic_file_vm_ops = {
  65 };
  66
  67 /*
  68  * Return the total memory allocated for this pointer, not
  69  * just what the caller asked for.
  70  *
  71  * Doesn't have to be accurate, i.e. may have races.
  72  */
  73 unsigned int kobjsize(const void *objp)
  74 {
  75         struct page *page;
  76
  77         /*
  78          * If the object we have should not have ksize performed on it,
  79          * return size of 0
  80          */
  81         if (!objp || !virt_addr_valid(objp))
  82                 return 0;
  83
  84         page = virt_to_head_page(objp);
  85
  86         /*
  87          * If the allocator sets PageSlab, we know the pointer came from
  88          * kmalloc().
  89          */
  90         if (PageSlab(page))
  91                 return ksize(objp);
  92
  93         /*
  94          * If it's not a compound page, see if we have a matching VMA
  95          * region. This test is intentionally done in reverse order,
  96          * so if there's no VMA, we still fall through and hand back
  97          * PAGE_SIZE for 0-order pages.
  98          */
  99         if (!PageCompound(page)) {
 100                 struct vm_area_struct *vma;
 101
 102                 vma = find_vma(current->mm, (unsigned long)objp);
 103                 if (vma)
 104                         return vma->vm_end - vma->vm_start;
 105         }
 106
 107         /*
 108          * The ksize() function is only guaranteed to work for pointers
 109          * returned by kmalloc(). So handle arbitrary pointers here.
 110          */
 111         return page_size(page);
 112 }
 113
 114 /**
 115  * follow_pfn - look up PFN at a user virtual address
 116  * @vma: memory mapping
 117  * @address: user virtual address
 118  * @pfn: location to store found PFN
 119  *
 120  * Only IO mappings and raw PFN mappings are allowed.
 121  *
 122  * Returns zero and the pfn at @pfn on success, -ve otherwise.
 123  */
 124 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
 125         unsigned long *pfn)
 126 {
 127         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
 128                 return -EINVAL;
 129
 130         *pfn = address >> PAGE_SHIFT;
 131         return 0;
 132 }
 133 EXPORT_SYMBOL(follow_pfn);
 134
 135 LIST_HEAD(vmap_area_list);
 136
 137 void vfree(const void *addr)
 138 {
 139         kfree(addr);
 140 }
 141 EXPORT_SYMBOL(vfree);
 142
 143 void *__vmalloc(unsigned long size, gfp_t gfp_mask)
 144 {
 145         /*
 146          *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 147          * returns only a logical address.
 148          */
 149         return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 150 }
 151 EXPORT_SYMBOL(__vmalloc);
 152
 153 void *__vmalloc_node_range(unsigned long size, unsigned long align,
 154                 unsigned long start, unsigned long end, gfp_t gfp_mask,
 155                 pgprot_t prot, unsigned long vm_flags, int node,
 156                 const void *caller)
 157 {
 158         return __vmalloc(size, gfp_mask);
 159 }
 160
 161 void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
 162                 int node, const void *caller)
 163 {
 164         return __vmalloc(size, gfp_mask);
 165 }
 166
 167 static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
 168 {
 169         void *ret;
 170
 171         ret = __vmalloc(size, flags);
 172         if (ret) {
 173                 struct vm_area_struct *vma;
 174
 175                 mmap_write_lock(current->mm);
 176                 vma = find_vma(current->mm, (unsigned long)ret);
 177                 if (vma)
 178                         vma->vm_flags |= VM_USERMAP;
 179                 mmap_write_unlock(current->mm);
 180         }
 181
 182         return ret;
 183 }
 184
 185 void *vmalloc_user(unsigned long size)
 186 {
 187         return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
 188 }
 189 EXPORT_SYMBOL(vmalloc_user);
 190
 191 struct page *vmalloc_to_page(const void *addr)
 192 {
 193         return virt_to_page(addr);
 194 }
 195 EXPORT_SYMBOL(vmalloc_to_page);
 196
 197 unsigned long vmalloc_to_pfn(const void *addr)
 198 {
 199         return page_to_pfn(virt_to_page(addr));
 200 }
 201 EXPORT_SYMBOL(vmalloc_to_pfn);
 202
 203 long vread(char *buf, char *addr, unsigned long count)
 204 {
 205         /* Don't allow overflow */
 206         if ((unsigned long) buf + count < count)
 207                 count = -(unsigned long) buf;
 208
 209         memcpy(buf, addr, count);
 210         return count;
 211 }
 212
 213 /*
 214  *      vmalloc  -  allocate virtually contiguous memory
 215  *
 216  *      @size:          allocation size
 217  *
 218  *      Allocate enough pages to cover @size from the page level
 219  *      allocator and map them into contiguous kernel virtual space.
 220  *
 221  *      For tight control over page level allocator and protection flags
 222  *      use __vmalloc() instead.
 223  */
 224 void *vmalloc(unsigned long size)
 225 {
 226         return __vmalloc(size, GFP_KERNEL);
 227 }
 228 EXPORT_SYMBOL(vmalloc);
 229
 230 /*
 231  *      vzalloc - allocate virtually contiguous memory with zero fill
 232  *
 233  *      @size:          allocation size
 234  *
 235  *      Allocate enough pages to cover @size from the page level
 236  *      allocator and map them into contiguous kernel virtual space.
 237  *      The memory allocated is set to zero.
 238  *
 239  *      For tight control over page level allocator and protection flags
 240  *      use __vmalloc() instead.
 241  */
 242 void *vzalloc(unsigned long size)
 243 {
 244         return __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
 245 }
 246 EXPORT_SYMBOL(vzalloc);
 247
 248 /**
 249  * vmalloc_node - allocate memory on a specific node
 250  * @size:       allocation size
 251  * @node:       numa node
 252  *
 253  * Allocate enough pages to cover @size from the page level
 254  * allocator and map them into contiguous kernel virtual space.
 255  *
 256  * For tight control over page level allocator and protection flags
 257  * use __vmalloc() instead.
 258  */
 259 void *vmalloc_node(unsigned long size, int node)
 260 {
 261         return vmalloc(size);
 262 }
 263 EXPORT_SYMBOL(vmalloc_node);
 264
 265 /**
 266  * vzalloc_node - allocate memory on a specific node with zero fill
 267  * @size:       allocation size
 268  * @node:       numa node
 269  *
 270  * Allocate enough pages to cover @size from the page level
 271  * allocator and map them into contiguous kernel virtual space.
 272  * The memory allocated is set to zero.
 273  *
 274  * For tight control over page level allocator and protection flags
 275  * use __vmalloc() instead.
 276  */
 277 void *vzalloc_node(unsigned long size, int node)
 278 {
 279         return vzalloc(size);
 280 }
 281 EXPORT_SYMBOL(vzalloc_node);
 282
 283 /**
 284  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 285  *      @size:          allocation size
 286  *
 287  *      Allocate enough 32bit PA addressable pages to cover @size from the
 288  *      page level allocator and map them into contiguous kernel virtual space.
 289  */
 290 void *vmalloc_32(unsigned long size)
 291 {
 292         return __vmalloc(size, GFP_KERNEL);
 293 }
 294 EXPORT_SYMBOL(vmalloc_32);
 295
 296 /**
 297  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 298  *      @size:          allocation size
 299  *
 300  * The resulting memory area is 32bit addressable and zeroed so it can be
 301  * mapped to userspace without leaking data.
 302  *
 303  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 304  * remap_vmalloc_range() are permissible.
 305  */
 306 void *vmalloc_32_user(unsigned long size)
 307 {
 308         /*
 309          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 310          * but for now this can simply use vmalloc_user() directly.
 311          */
 312         return vmalloc_user(size);
 313 }
 314 EXPORT_SYMBOL(vmalloc_32_user);
 315
 316 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 317 {
 318         BUG();
 319         return NULL;
 320 }
 321 EXPORT_SYMBOL(vmap);
 322
 323 void vunmap(const void *addr)
 324 {
 325         BUG();
 326 }
 327 EXPORT_SYMBOL(vunmap);
 328
 329 void *vm_map_ram(struct page **pages, unsigned int count, int node)
 330 {
 331         BUG();
 332         return NULL;
 333 }
 334 EXPORT_SYMBOL(vm_map_ram);
 335
 336 void vm_unmap_ram(const void *mem, unsigned int count)
 337 {
 338         BUG();
 339 }
 340 EXPORT_SYMBOL(vm_unmap_ram);
 341
 342 void vm_unmap_aliases(void)
 343 {
 344 }
 345 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 346
 347 void free_vm_area(struct vm_struct *area)
 348 {
 349         BUG();
 350 }
 351 EXPORT_SYMBOL_GPL(free_vm_area);
 352
 353 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 354                    struct page *page)
 355 {
 356         return -EINVAL;
 357 }
 358 EXPORT_SYMBOL(vm_insert_page);
 359
 360 int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
 361                         unsigned long num)
 362 {
 363         return -EINVAL;
 364 }
 365 EXPORT_SYMBOL(vm_map_pages);
 366
 367 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
 368                                 unsigned long num)
 369 {
 370         return -EINVAL;
 371 }
 372 EXPORT_SYMBOL(vm_map_pages_zero);
 373
 374 /*
 375  *  sys_brk() for the most part doesn't need the global kernel
 376  *  lock, except when an application is doing something nasty
 377  *  like trying to un-brk an area that has already been mapped
 378  *  to a regular file.  in this case, the unmapping will need
 379  *  to invoke file system routines that need the global lock.
 380  */
 381 SYSCALL_DEFINE1(brk, unsigned long, brk)
 382 {
 383         struct mm_struct *mm = current->mm;
 384
 385         if (brk < mm->start_brk || brk > mm->context.end_brk)
 386                 return mm->brk;
 387
 388         if (mm->brk == brk)
 389                 return mm->brk;
 390
 391         /*
 392          * Always allow shrinking brk
 393          */
 394         if (brk <= mm->brk) {
 395                 mm->brk = brk;
 396                 return brk;
 397         }
 398
 399         /*
 400          * Ok, looks good - let it rip.
 401          */
 402         flush_icache_user_range(mm->brk, brk);
 403         return mm->brk = brk;
 404 }
 405
 406 /*
 407  * initialise the percpu counter for VM and region record slabs
 408  */
 409 void __init mmap_init(void)
 410 {
 411         int ret;
 412
 413         ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
 414         VM_BUG_ON(ret);
 415         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
 416 }
 417
 418 /*
 419  * validate the region tree
 420  * - the caller must hold the region lock
 421  */
 422 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 423 static noinline void validate_nommu_regions(void)
 424 {
 425         struct vm_region *region, *last;
 426         struct rb_node *p, *lastp;
 427
 428         lastp = rb_first(&nommu_region_tree);
 429         if (!lastp)
 430                 return;
 431
 432         last = rb_entry(lastp, struct vm_region, vm_rb);
 433         BUG_ON(last->vm_end <= last->vm_start);
 434         BUG_ON(last->vm_top < last->vm_end);
 435
 436         while ((p = rb_next(lastp))) {
 437                 region = rb_entry(p, struct vm_region, vm_rb);
 438                 last = rb_entry(lastp, struct vm_region, vm_rb);
 439
 440                 BUG_ON(region->vm_end <= region->vm_start);
 441                 BUG_ON(region->vm_top < region->vm_end);
 442                 BUG_ON(region->vm_start < last->vm_top);
 443
 444                 lastp = p;
 445         }
 446 }
 447 #else
 448 static void validate_nommu_regions(void)
 449 {
 450 }
 451 #endif
 452
 453 /*
 454  * add a region into the global tree
 455  */
 456 static void add_nommu_region(struct vm_region *region)
 457 {
 458         struct vm_region *pregion;
 459         struct rb_node **p, *parent;
 460
 461         validate_nommu_regions();
 462
 463         parent = NULL;
 464         p = &nommu_region_tree.rb_node;
 465         while (*p) {
 466                 parent = *p;
 467                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 468                 if (region->vm_start < pregion->vm_start)
 469                         p = &(*p)->rb_left;
 470                 else if (region->vm_start > pregion->vm_start)
 471                         p = &(*p)->rb_right;
 472                 else if (pregion == region)
 473                         return;
 474                 else
 475                         BUG();
 476         }
 477
 478         rb_link_node(&region->vm_rb, parent, p);
 479         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 480
 481         validate_nommu_regions();
 482 }
 483
 484 /*
 485  * delete a region from the global tree
 486  */
 487 static void delete_nommu_region(struct vm_region *region)
 488 {
 489         BUG_ON(!nommu_region_tree.rb_node);
 490
 491         validate_nommu_regions();
 492         rb_erase(&region->vm_rb, &nommu_region_tree);
 493         validate_nommu_regions();
 494 }
 495
 496 /*
 497  * free a contiguous series of pages
 498  */
 499 static void free_page_series(unsigned long from, unsigned long to)
 500 {
 501         for (; from < to; from += PAGE_SIZE) {
 502                 struct page *page = virt_to_page(from);
 503
 504                 atomic_long_dec(&mmap_pages_allocated);
 505                 put_page(page);
 506         }
 507 }
 508
 509 /*
 510  * release a reference to a region
 511  * - the caller must hold the region semaphore for writing, which this releases
 512  * - the region may not have been added to the tree yet, in which case vm_top
 513  *   will equal vm_start
 514  */
 515 static void __put_nommu_region(struct vm_region *region)
 516         __releases(nommu_region_sem)
 517 {
 518         BUG_ON(!nommu_region_tree.rb_node);
 519
 520         if (--region->vm_usage == 0) {
 521                 if (region->vm_top > region->vm_start)
 522                         delete_nommu_region(region);
 523                 up_write(&nommu_region_sem);
 524
 525                 if (region->vm_file)
 526                         fput(region->vm_file);
 527
 528                 /* IO memory and memory shared directly out of the pagecache
 529                  * from ramfs/tmpfs mustn't be released here */
 530                 if (region->vm_flags & VM_MAPPED_COPY)
 531                         free_page_series(region->vm_start, region->vm_top);
 532                 kmem_cache_free(vm_region_jar, region);
 533         } else {
 534                 up_write(&nommu_region_sem);
 535         }
 536 }
 537
 538 /*
 539  * release a reference to a region
 540  */
 541 static void put_nommu_region(struct vm_region *region)
 542 {
 543         down_write(&nommu_region_sem);
 544         __put_nommu_region(region);
 545 }
 546
 547 /*
 548  * add a VMA into a process's mm_struct in the appropriate place in the list
 549  * and tree and add to the address space's page tree also if not an anonymous
 550  * page
 551  * - should be called with mm->mmap_lock held writelocked
 552  */
 553 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 554 {
 555         struct vm_area_struct *pvma, *prev;
 556         struct address_space *mapping;
 557         struct rb_node **p, *parent, *rb_prev;
 558
 559         BUG_ON(!vma->vm_region);
 560
 561         mm->map_count++;
 562         vma->vm_mm = mm;
 563
 564         /* add the VMA to the mapping */
 565         if (vma->vm_file) {
 566                 mapping = vma->vm_file->f_mapping;
 567
 568                 i_mmap_lock_write(mapping);
 569                 flush_dcache_mmap_lock(mapping);
 570                 vma_interval_tree_insert(vma, &mapping->i_mmap);
 571                 flush_dcache_mmap_unlock(mapping);
 572                 i_mmap_unlock_write(mapping);
 573         }
 574
 575         /* add the VMA to the tree */
 576         parent = rb_prev = NULL;
 577         p = &mm->mm_rb.rb_node;
 578         while (*p) {
 579                 parent = *p;
 580                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 581
 582                 /* sort by: start addr, end addr, VMA struct addr in that order
 583                  * (the latter is necessary as we may get identical VMAs) */
 584                 if (vma->vm_start < pvma->vm_start)
 585                         p = &(*p)->rb_left;
 586                 else if (vma->vm_start > pvma->vm_start) {
 587                         rb_prev = parent;
 588                         p = &(*p)->rb_right;
 589                 } else if (vma->vm_end < pvma->vm_end)
 590                         p = &(*p)->rb_left;
 591                 else if (vma->vm_end > pvma->vm_end) {
 592                         rb_prev = parent;
 593                         p = &(*p)->rb_right;
 594                 } else if (vma < pvma)
 595                         p = &(*p)->rb_left;
 596                 else if (vma > pvma) {
 597                         rb_prev = parent;
 598                         p = &(*p)->rb_right;
 599                 } else
 600                         BUG();
 601         }
 602
 603         rb_link_node(&vma->vm_rb, parent, p);
 604         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 605
 606         /* add VMA to the VMA list also */
 607         prev = NULL;
 608         if (rb_prev)
 609                 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 610
 611         __vma_link_list(mm, vma, prev);
 612 }
 613
 614 /*
 615  * delete a VMA from its owning mm_struct and address space
 616  */
 617 static void delete_vma_from_mm(struct vm_area_struct *vma)
 618 {
 619         int i;
 620         struct address_space *mapping;
 621         struct mm_struct *mm = vma->vm_mm;
 622         struct task_struct *curr = current;
 623
 624         mm->map_count--;
 625         for (i = 0; i < VMACACHE_SIZE; i++) {
 626                 /* if the vma is cached, invalidate the entire cache */
 627                 if (curr->vmacache.vmas[i] == vma) {
 628                         vmacache_invalidate(mm);
 629                         break;
 630                 }
 631         }
 632
 633         /* remove the VMA from the mapping */
 634         if (vma->vm_file) {
 635                 mapping = vma->vm_file->f_mapping;
 636
 637                 i_mmap_lock_write(mapping);
 638                 flush_dcache_mmap_lock(mapping);
 639                 vma_interval_tree_remove(vma, &mapping->i_mmap);
 640                 flush_dcache_mmap_unlock(mapping);
 641                 i_mmap_unlock_write(mapping);
 642         }
 643
 644         /* remove from the MM's tree and list */
 645         rb_erase(&vma->vm_rb, &mm->mm_rb);
 646
 647         __vma_unlink_list(mm, vma);
 648 }
 649
 650 /*
 651  * destroy a VMA record
 652  */
 653 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 654 {
 655         if (vma->vm_ops && vma->vm_ops->close)
 656                 vma->vm_ops->close(vma);
 657         if (vma->vm_file)
 658                 fput(vma->vm_file);
 659         put_nommu_region(vma->vm_region);
 660         vm_area_free(vma);
 661 }
 662
 663 /*
 664  * look up the first VMA in which addr resides, NULL if none
 665  * - should be called with mm->mmap_lock at least held readlocked
 666  */
 667 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 668 {
 669         struct vm_area_struct *vma;
 670
 671         /* check the cache first */
 672         vma = vmacache_find(mm, addr);
 673         if (likely(vma))
 674                 return vma;
 675
 676         /* trawl the list (there may be multiple mappings in which addr
 677          * resides) */
 678         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 679                 if (vma->vm_start > addr)
 680                         return NULL;
 681                 if (vma->vm_end > addr) {
 682                         vmacache_update(addr, vma);
 683                         return vma;
 684                 }
 685         }
 686
 687         return NULL;
 688 }
 689 EXPORT_SYMBOL(find_vma);
 690
 691 /*
 692  * find a VMA
 693  * - we don't extend stack VMAs under NOMMU conditions
 694  */
 695 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 696 {
 697         return find_vma(mm, addr);
 698 }
 699
 700 /*
 701  * expand a stack to a given address
 702  * - not supported under NOMMU conditions
 703  */
 704 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 705 {
 706         return -ENOMEM;
 707 }
 708
 709 /*
 710  * look up the first VMA exactly that exactly matches addr
 711  * - should be called with mm->mmap_lock at least held readlocked
 712  */
 713 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 714                                              unsigned long addr,
 715                                              unsigned long len)
 716 {
 717         struct vm_area_struct *vma;
 718         unsigned long end = addr + len;
 719
 720         /* check the cache first */
 721         vma = vmacache_find_exact(mm, addr, end);
 722         if (vma)
 723                 return vma;
 724
 725         /* trawl the list (there may be multiple mappings in which addr
 726          * resides) */
 727         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 728                 if (vma->vm_start < addr)
 729                         continue;
 730                 if (vma->vm_start > addr)
 731                         return NULL;
 732                 if (vma->vm_end == end) {
 733                         vmacache_update(addr, vma);
 734                         return vma;
 735                 }
 736         }
 737
 738         return NULL;
 739 }
 740
 741 /*
 742  * determine whether a mapping should be permitted and, if so, what sort of
 743  * mapping we're capable of supporting
 744  */
 745 static int validate_mmap_request(struct file *file,
 746                                  unsigned long addr,
 747                                  unsigned long len,
 748                                  unsigned long prot,
 749                                  unsigned long flags,
 750                                  unsigned long pgoff,
 751                                  unsigned long *_capabilities)
 752 {
 753         unsigned long capabilities, rlen;
 754         int ret;
 755
 756         /* do the simple checks first */
 757         if (flags & MAP_FIXED)
 758                 return -EINVAL;
 759
 760         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 761             (flags & MAP_TYPE) != MAP_SHARED)
 762                 return -EINVAL;
 763
 764         if (!len)
 765                 return -EINVAL;
 766
 767         /* Careful about overflows.. */
 768         rlen = PAGE_ALIGN(len);
 769         if (!rlen || rlen > TASK_SIZE)
 770                 return -ENOMEM;
 771
 772         /* offset overflow? */
 773         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 774                 return -EOVERFLOW;
 775
 776         if (file) {
 777                 /* files must support mmap */
 778                 if (!file->f_op->mmap)
 779                         return -ENODEV;
 780
 781                 /* work out if what we've got could possibly be shared
 782                  * - we support chardevs that provide their own "memory"
 783                  * - we support files/blockdevs that are memory backed
 784                  */
 785                 if (file->f_op->mmap_capabilities) {
 786                         capabilities = file->f_op->mmap_capabilities(file);
 787                 } else {
 788                         /* no explicit capabilities set, so assume some
 789                          * defaults */
 790                         switch (file_inode(file)->i_mode & S_IFMT) {
 791                         case S_IFREG:
 792                         case S_IFBLK:
 793                                 capabilities = NOMMU_MAP_COPY;
 794                                 break;
 795
 796                         case S_IFCHR:
 797                                 capabilities =
 798                                         NOMMU_MAP_DIRECT |
 799                                         NOMMU_MAP_READ |
 800                                         NOMMU_MAP_WRITE;
 801                                 break;
 802
 803                         default:
 804                                 return -EINVAL;
 805                         }
 806                 }
 807
 808                 /* eliminate any capabilities that we can't support on this
 809                  * device */
 810                 if (!file->f_op->get_unmapped_area)
 811                         capabilities &= ~NOMMU_MAP_DIRECT;
 812                 if (!(file->f_mode & FMODE_CAN_READ))
 813                         capabilities &= ~NOMMU_MAP_COPY;
 814
 815                 /* The file shall have been opened with read permission. */
 816                 if (!(file->f_mode & FMODE_READ))
 817                         return -EACCES;
 818
 819                 if (flags & MAP_SHARED) {
 820                         /* do checks for writing, appending and locking */
 821                         if ((prot & PROT_WRITE) &&
 822                             !(file->f_mode & FMODE_WRITE))
 823                                 return -EACCES;
 824
 825                         if (IS_APPEND(file_inode(file)) &&
 826                             (file->f_mode & FMODE_WRITE))
 827                                 return -EACCES;
 828
 829                         if (!(capabilities & NOMMU_MAP_DIRECT))
 830                                 return -ENODEV;
 831
 832                         /* we mustn't privatise shared mappings */
 833                         capabilities &= ~NOMMU_MAP_COPY;
 834                 } else {
 835                         /* we're going to read the file into private memory we
 836                          * allocate */
 837                         if (!(capabilities & NOMMU_MAP_COPY))
 838                                 return -ENODEV;
 839
 840                         /* we don't permit a private writable mapping to be
 841                          * shared with the backing device */
 842                         if (prot & PROT_WRITE)
 843                                 capabilities &= ~NOMMU_MAP_DIRECT;
 844                 }
 845
 846                 if (capabilities & NOMMU_MAP_DIRECT) {
 847                         if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
 848                             ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
 849                             ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
 850                             ) {
 851                                 capabilities &= ~NOMMU_MAP_DIRECT;
 852                                 if (flags & MAP_SHARED) {
 853                                         pr_warn("MAP_SHARED not completely supported on !MMU\n");
 854                                         return -EINVAL;
 855                                 }
 856                         }
 857                 }
 858
 859                 /* handle executable mappings and implied executable
 860                  * mappings */
 861                 if (path_noexec(&file->f_path)) {
 862                         if (prot & PROT_EXEC)
 863                                 return -EPERM;
 864                 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 865                         /* handle implication of PROT_EXEC by PROT_READ */
 866                         if (current->personality & READ_IMPLIES_EXEC) {
 867                                 if (capabilities & NOMMU_MAP_EXEC)
 868                                         prot |= PROT_EXEC;
 869                         }
 870                 } else if ((prot & PROT_READ) &&
 871                          (prot & PROT_EXEC) &&
 872                          !(capabilities & NOMMU_MAP_EXEC)
 873                          ) {
 874                         /* backing file is not executable, try to copy */
 875                         capabilities &= ~NOMMU_MAP_DIRECT;
 876                 }
 877         } else {
 878                 /* anonymous mappings are always memory backed and can be
 879                  * privately mapped
 880                  */
 881                 capabilities = NOMMU_MAP_COPY;
 882
 883                 /* handle PROT_EXEC implication by PROT_READ */
 884                 if ((prot & PROT_READ) &&
 885                     (current->personality & READ_IMPLIES_EXEC))
 886                         prot |= PROT_EXEC;
 887         }
 888
 889         /* allow the security API to have its say */
 890         ret = security_mmap_addr(addr);
 891         if (ret < 0)
 892                 return ret;
 893
 894         /* looks okay */
 895         *_capabilities = capabilities;
 896         return 0;
 897 }
 898
 899 /*
 900  * we've determined that we can make the mapping, now translate what we
 901  * now know into VMA flags
 902  */
 903 static unsigned long determine_vm_flags(struct file *file,
 904                                         unsigned long prot,
 905                                         unsigned long flags,
 906                                         unsigned long capabilities)
 907 {
 908         unsigned long vm_flags;
 909
 910         vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
 911         /* vm_flags |= mm->def_flags; */
 912
 913         if (!(capabilities & NOMMU_MAP_DIRECT)) {
 914                 /* attempt to share read-only copies of mapped file chunks */
 915                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 916                 if (file && !(prot & PROT_WRITE))
 917                         vm_flags |= VM_MAYSHARE;
 918         } else {
 919                 /* overlay a shareable mapping on the backing device or inode
 920                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 921                  * romfs/cramfs */
 922                 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
 923                 if (flags & MAP_SHARED)
 924                         vm_flags |= VM_SHARED;
 925         }
 926
 927         /* refuse to let anyone share private mappings with this process if
 928          * it's being traced - otherwise breakpoints set in it may interfere
 929          * with another untraced process
 930          */
 931         if ((flags & MAP_PRIVATE) && current->ptrace)
 932                 vm_flags &= ~VM_MAYSHARE;
 933
 934         return vm_flags;
 935 }
 936
 937 /*
 938  * set up a shared mapping on a file (the driver or filesystem provides and
 939  * pins the storage)
 940  */
 941 static int do_mmap_shared_file(struct vm_area_struct *vma)
 942 {
 943         int ret;
 944
 945         ret = call_mmap(vma->vm_file, vma);
 946         if (ret == 0) {
 947                 vma->vm_region->vm_top = vma->vm_region->vm_end;
 948                 return 0;
 949         }
 950         if (ret != -ENOSYS)
 951                 return ret;
 952
 953         /* getting -ENOSYS indicates that direct mmap isn't possible (as
 954          * opposed to tried but failed) so we can only give a suitable error as
 955          * it's not possible to make a private copy if MAP_SHARED was given */
 956         return -ENODEV;
 957 }
 958
 959 /*
 960  * set up a private mapping or an anonymous shared mapping
 961  */
 962 static int do_mmap_private(struct vm_area_struct *vma,
 963                            struct vm_region *region,
 964                            unsigned long len,
 965                            unsigned long capabilities)
 966 {
 967         unsigned long total, point;
 968         void *base;
 969         int ret, order;
 970
 971         /* invoke the file's mapping function so that it can keep track of
 972          * shared mappings on devices or memory
 973          * - VM_MAYSHARE will be set if it may attempt to share
 974          */
 975         if (capabilities & NOMMU_MAP_DIRECT) {
 976                 ret = call_mmap(vma->vm_file, vma);
 977                 if (ret == 0) {
 978                         /* shouldn't return success if we're not sharing */
 979                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
 980                         vma->vm_region->vm_top = vma->vm_region->vm_end;
 981                         return 0;
 982                 }
 983                 if (ret != -ENOSYS)
 984                         return ret;
 985
 986                 /* getting an ENOSYS error indicates that direct mmap isn't
 987                  * possible (as opposed to tried but failed) so we'll try to
 988                  * make a private copy of the data and map that instead */
 989         }
 990
 991
 992         /* allocate some memory to hold the mapping
 993          * - note that this may not return a page-aligned address if the object
 994          *   we're allocating is smaller than a page
 995          */
 996         order = get_order(len);
 997         total = 1 << order;
 998         point = len >> PAGE_SHIFT;
 999
1000         /* we don't want to allocate a power-of-2 sized page set */
1001         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1002                 total = point;
1003
1004         base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1005         if (!base)
1006                 goto enomem;
1007
1008         atomic_long_add(total, &mmap_pages_allocated);
1009
1010         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1011         region->vm_start = (unsigned long) base;
1012         region->vm_end   = region->vm_start + len;
1013         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1014
1015         vma->vm_start = region->vm_start;
1016         vma->vm_end   = region->vm_start + len;
1017
1018         if (vma->vm_file) {
1019                 /* read the contents of a file into the copy */
1020                 loff_t fpos;
1021
1022                 fpos = vma->vm_pgoff;
1023                 fpos <<= PAGE_SHIFT;
1024
1025                 ret = kernel_read(vma->vm_file, base, len, &fpos);
1026                 if (ret < 0)
1027                         goto error_free;
1028
1029                 /* clear the last little bit */
1030                 if (ret < len)
1031                         memset(base + ret, 0, len - ret);
1032
1033         } else {
1034                 vma_set_anonymous(vma);
1035         }
1036
1037         return 0;
1038
1039 error_free:
1040         free_page_series(region->vm_start, region->vm_top);
1041         region->vm_start = vma->vm_start = 0;
1042         region->vm_end   = vma->vm_end = 0;
1043         region->vm_top   = 0;
1044         return ret;
1045
1046 enomem:
1047         pr_err("Allocation of length %lu from process %d (%s) failed\n",
1048                len, current->pid, current->comm);
1049         show_free_areas(0, NULL);
1050         return -ENOMEM;
1051 }
1052
1053 /*
1054  * handle mapping creation for uClinux
1055  */
1056 unsigned long do_mmap(struct file *file,
1057                         unsigned long addr,
1058                         unsigned long len,
1059                         unsigned long prot,
1060                         unsigned long flags,
1061                         unsigned long pgoff,
1062                         unsigned long *populate,
1063                         struct list_head *uf)
1064 {
1065         struct vm_area_struct *vma;
1066         struct vm_region *region;
1067         struct rb_node *rb;
1068         vm_flags_t vm_flags;
1069         unsigned long capabilities, result;
1070         int ret;
1071
1072         *populate = 0;
1073
1074         /* decide whether we should attempt the mapping, and if so what sort of
1075          * mapping */
1076         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1077                                     &capabilities);
1078         if (ret < 0)
1079                 return ret;
1080
1081         /* we ignore the address hint */
1082         addr = 0;
1083         len = PAGE_ALIGN(len);
1084
1085         /* we've determined that we can make the mapping, now translate what we
1086          * now know into VMA flags */
1087         vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1088
1089         /* we're going to need to record the mapping */
1090         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1091         if (!region)
1092                 goto error_getting_region;
1093
1094         vma = vm_area_alloc(current->mm);
1095         if (!vma)
1096                 goto error_getting_vma;
1097
1098         region->vm_usage = 1;
1099         region->vm_flags = vm_flags;
1100         region->vm_pgoff = pgoff;
1101
1102         vma->vm_flags = vm_flags;
1103         vma->vm_pgoff = pgoff;
1104
1105         if (file) {
1106                 region->vm_file = get_file(file);
1107                 vma->vm_file = get_file(file);
1108         }
1109
1110         down_write(&nommu_region_sem);
1111
1112         /* if we want to share, we need to check for regions created by other
1113          * mmap() calls that overlap with our proposed mapping
1114          * - we can only share with a superset match on most regular files
1115          * - shared mappings on character devices and memory backed files are
1116          *   permitted to overlap inexactly as far as we are concerned for in
1117          *   these cases, sharing is handled in the driver or filesystem rather
1118          *   than here
1119          */
1120         if (vm_flags & VM_MAYSHARE) {
1121                 struct vm_region *pregion;
1122                 unsigned long pglen, rpglen, pgend, rpgend, start;
1123
1124                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1125                 pgend = pgoff + pglen;
1126
1127                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1128                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1129
1130                         if (!(pregion->vm_flags & VM_MAYSHARE))
1131                                 continue;
1132
1133                         /* search for overlapping mappings on the same file */
1134                         if (file_inode(pregion->vm_file) !=
1135                             file_inode(file))
1136                                 continue;
1137
1138                         if (pregion->vm_pgoff >= pgend)
1139                                 continue;
1140
1141                         rpglen = pregion->vm_end - pregion->vm_start;
1142                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1143                         rpgend = pregion->vm_pgoff + rpglen;
1144                         if (pgoff >= rpgend)
1145                                 continue;
1146
1147                         /* handle inexactly overlapping matches between
1148                          * mappings */
1149                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1150                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1151                                 /* new mapping is not a subset of the region */
1152                                 if (!(capabilities & NOMMU_MAP_DIRECT))
1153                                         goto sharing_violation;
1154                                 continue;
1155                         }
1156
1157                         /* we've found a region we can share */
1158                         pregion->vm_usage++;
1159                         vma->vm_region = pregion;
1160                         start = pregion->vm_start;
1161                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1162                         vma->vm_start = start;
1163                         vma->vm_end = start + len;
1164
1165                         if (pregion->vm_flags & VM_MAPPED_COPY)
1166                                 vma->vm_flags |= VM_MAPPED_COPY;
1167                         else {
1168                                 ret = do_mmap_shared_file(vma);
1169                                 if (ret < 0) {
1170                                         vma->vm_region = NULL;
1171                                         vma->vm_start = 0;
1172                                         vma->vm_end = 0;
1173                                         pregion->vm_usage--;
1174                                         pregion = NULL;
1175                                         goto error_just_free;
1176                                 }
1177                         }
1178                         fput(region->vm_file);
1179                         kmem_cache_free(vm_region_jar, region);
1180                         region = pregion;
1181                         result = start;
1182                         goto share;
1183                 }
1184
1185                 /* obtain the address at which to make a shared mapping
1186                  * - this is the hook for quasi-memory character devices to
1187                  *   tell us the location of a shared mapping
1188                  */
1189                 if (capabilities & NOMMU_MAP_DIRECT) {
1190                         addr = file->f_op->get_unmapped_area(file, addr, len,
1191                                                              pgoff, flags);
1192                         if (IS_ERR_VALUE(addr)) {
1193                                 ret = addr;
1194                                 if (ret != -ENOSYS)
1195                                         goto error_just_free;
1196
1197                                 /* the driver refused to tell us where to site
1198                                  * the mapping so we'll have to attempt to copy
1199                                  * it */
1200                                 ret = -ENODEV;
1201                                 if (!(capabilities & NOMMU_MAP_COPY))
1202                                         goto error_just_free;
1203
1204                                 capabilities &= ~NOMMU_MAP_DIRECT;
1205                         } else {
1206                                 vma->vm_start = region->vm_start = addr;
1207                                 vma->vm_end = region->vm_end = addr + len;
1208                         }
1209                 }
1210         }
1211
1212         vma->vm_region = region;
1213
1214         /* set up the mapping
1215          * - the region is filled in if NOMMU_MAP_DIRECT is still set
1216          */
1217         if (file && vma->vm_flags & VM_SHARED)
1218                 ret = do_mmap_shared_file(vma);
1219         else
1220                 ret = do_mmap_private(vma, region, len, capabilities);
1221         if (ret < 0)
1222                 goto error_just_free;
1223         add_nommu_region(region);
1224
1225         /* clear anonymous mappings that don't ask for uninitialized data */
1226         if (!vma->vm_file &&
1227             (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1228              !(flags & MAP_UNINITIALIZED)))
1229                 memset((void *)region->vm_start, 0,
1230                        region->vm_end - region->vm_start);
1231
1232         /* okay... we have a mapping; now we have to register it */
1233         result = vma->vm_start;
1234
1235         current->mm->total_vm += len >> PAGE_SHIFT;
1236
1237 share:
1238         add_vma_to_mm(current->mm, vma);
1239
1240         /* we flush the region from the icache only when the first executable
1241          * mapping of it is made  */
1242         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1243                 flush_icache_user_range(region->vm_start, region->vm_end);
1244                 region->vm_icache_flushed = true;
1245         }
1246
1247         up_write(&nommu_region_sem);
1248
1249         return result;
1250
1251 error_just_free:
1252         up_write(&nommu_region_sem);
1253 error:
1254         if (region->vm_file)
1255                 fput(region->vm_file);
1256         kmem_cache_free(vm_region_jar, region);
1257         if (vma->vm_file)
1258                 fput(vma->vm_file);
1259         vm_area_free(vma);
1260         return ret;
1261
1262 sharing_violation:
1263         up_write(&nommu_region_sem);
1264         pr_warn("Attempt to share mismatched mappings\n");
1265         ret = -EINVAL;
1266         goto error;
1267
1268 error_getting_vma:
1269         kmem_cache_free(vm_region_jar, region);
1270         pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1271                         len, current->pid);
1272         show_free_areas(0, NULL);
1273         return -ENOMEM;
1274
1275 error_getting_region:
1276         pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1277                         len, current->pid);
1278         show_free_areas(0, NULL);
1279         return -ENOMEM;
1280 }
1281
1282 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1283                               unsigned long prot, unsigned long flags,
1284                               unsigned long fd, unsigned long pgoff)
1285 {
1286         struct file *file = NULL;
1287         unsigned long retval = -EBADF;
1288
1289         audit_mmap_fd(fd, flags);
1290         if (!(flags & MAP_ANONYMOUS)) {
1291                 file = fget(fd);
1292                 if (!file)
1293                         goto out;
1294         }
1295
1296         retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1297
1298         if (file)
1299                 fput(file);
1300 out:
1301         return retval;
1302 }
1303
1304 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1305                 unsigned long, prot, unsigned long, flags,
1306                 unsigned long, fd, unsigned long, pgoff)
1307 {
1308         return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1309 }
1310
1311 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1312 struct mmap_arg_struct {
1313         unsigned long addr;
1314         unsigned long len;
1315         unsigned long prot;
1316         unsigned long flags;
1317         unsigned long fd;
1318         unsigned long offset;
1319 };
1320
1321 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1322 {
1323         struct mmap_arg_struct a;
1324
1325         if (copy_from_user(&a, arg, sizeof(a)))
1326                 return -EFAULT;
1327         if (offset_in_page(a.offset))
1328                 return -EINVAL;
1329
1330         return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1331                                a.offset >> PAGE_SHIFT);
1332 }
1333 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1334
1335 /*
1336  * split a vma into two pieces at address 'addr', a new vma is allocated either
1337  * for the first part or the tail.
1338  */
1339 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1340               unsigned long addr, int new_below)
1341 {
1342         struct vm_area_struct *new;
1343         struct vm_region *region;
1344         unsigned long npages;
1345
1346         /* we're only permitted to split anonymous regions (these should have
1347          * only a single usage on the region) */
1348         if (vma->vm_file)
1349                 return -ENOMEM;
1350
1351         if (mm->map_count >= sysctl_max_map_count)
1352                 return -ENOMEM;
1353
1354         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1355         if (!region)
1356                 return -ENOMEM;
1357
1358         new = vm_area_dup(vma);
1359         if (!new) {
1360                 kmem_cache_free(vm_region_jar, region);
1361                 return -ENOMEM;
1362         }
1363
1364         /* most fields are the same, copy all, and then fixup */
1365         *region = *vma->vm_region;
1366         new->vm_region = region;
1367
1368         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1369
1370         if (new_below) {
1371                 region->vm_top = region->vm_end = new->vm_end = addr;
1372         } else {
1373                 region->vm_start = new->vm_start = addr;
1374                 region->vm_pgoff = new->vm_pgoff += npages;
1375         }
1376
1377         if (new->vm_ops && new->vm_ops->open)
1378                 new->vm_ops->open(new);
1379
1380         delete_vma_from_mm(vma);
1381         down_write(&nommu_region_sem);
1382         delete_nommu_region(vma->vm_region);
1383         if (new_below) {
1384                 vma->vm_region->vm_start = vma->vm_start = addr;
1385                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1386         } else {
1387                 vma->vm_region->vm_end = vma->vm_end = addr;
1388                 vma->vm_region->vm_top = addr;
1389         }
1390         add_nommu_region(vma->vm_region);
1391         add_nommu_region(new->vm_region);
1392         up_write(&nommu_region_sem);
1393         add_vma_to_mm(mm, vma);
1394         add_vma_to_mm(mm, new);
1395         return 0;
1396 }
1397
1398 /*
1399  * shrink a VMA by removing the specified chunk from either the beginning or
1400  * the end
1401  */
1402 static int shrink_vma(struct mm_struct *mm,
1403                       struct vm_area_struct *vma,
1404                       unsigned long from, unsigned long to)
1405 {
1406         struct vm_region *region;
1407
1408         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1409          * and list */
1410         delete_vma_from_mm(vma);
1411         if (from > vma->vm_start)
1412                 vma->vm_end = from;
1413         else
1414                 vma->vm_start = to;
1415         add_vma_to_mm(mm, vma);
1416
1417         /* cut the backing region down to size */
1418         region = vma->vm_region;
1419         BUG_ON(region->vm_usage != 1);
1420
1421         down_write(&nommu_region_sem);
1422         delete_nommu_region(region);
1423         if (from > region->vm_start) {
1424                 to = region->vm_top;
1425                 region->vm_top = region->vm_end = from;
1426         } else {
1427                 region->vm_start = to;
1428         }
1429         add_nommu_region(region);
1430         up_write(&nommu_region_sem);
1431
1432         free_page_series(from, to);
1433         return 0;
1434 }
1435
1436 /*
1437  * release a mapping
1438  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1439  *   VMA, though it need not cover the whole VMA
1440  */
1441 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1442 {
1443         struct vm_area_struct *vma;
1444         unsigned long end;
1445         int ret;
1446
1447         len = PAGE_ALIGN(len);
1448         if (len == 0)
1449                 return -EINVAL;
1450
1451         end = start + len;
1452
1453         /* find the first potentially overlapping VMA */
1454         vma = find_vma(mm, start);
1455         if (!vma) {
1456                 static int limit;
1457                 if (limit < 5) {
1458                         pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1459                                         current->pid, current->comm,
1460                                         start, start + len - 1);
1461                         limit++;
1462                 }
1463                 return -EINVAL;
1464         }
1465
1466         /* we're allowed to split an anonymous VMA but not a file-backed one */
1467         if (vma->vm_file) {
1468                 do {
1469                         if (start > vma->vm_start)
1470                                 return -EINVAL;
1471                         if (end == vma->vm_end)
1472                                 goto erase_whole_vma;
1473                         vma = vma->vm_next;
1474                 } while (vma);
1475                 return -EINVAL;
1476         } else {
1477                 /* the chunk must be a subset of the VMA found */
1478                 if (start == vma->vm_start && end == vma->vm_end)
1479                         goto erase_whole_vma;
1480                 if (start < vma->vm_start || end > vma->vm_end)
1481                         return -EINVAL;
1482                 if (offset_in_page(start))
1483                         return -EINVAL;
1484                 if (end != vma->vm_end && offset_in_page(end))
1485                         return -EINVAL;
1486                 if (start != vma->vm_start && end != vma->vm_end) {
1487                         ret = split_vma(mm, vma, start, 1);
1488                         if (ret < 0)
1489                                 return ret;
1490                 }
1491                 return shrink_vma(mm, vma, start, end);
1492         }
1493
1494 erase_whole_vma:
1495         delete_vma_from_mm(vma);
1496         delete_vma(mm, vma);
1497         return 0;
1498 }
1499
1500 int vm_munmap(unsigned long addr, size_t len)
1501 {
1502         struct mm_struct *mm = current->mm;
1503         int ret;
1504
1505         mmap_write_lock(mm);
1506         ret = do_munmap(mm, addr, len, NULL);
1507         mmap_write_unlock(mm);
1508         return ret;
1509 }
1510 EXPORT_SYMBOL(vm_munmap);
1511
1512 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1513 {
1514         return vm_munmap(addr, len);
1515 }
1516
1517 /*
1518  * release all the mappings made in a process's VM space
1519  */
1520 void exit_mmap(struct mm_struct *mm)
1521 {
1522         struct vm_area_struct *vma;
1523
1524         if (!mm)
1525                 return;
1526
1527         mm->total_vm = 0;
1528
1529         while ((vma = mm->mmap)) {
1530                 mm->mmap = vma->vm_next;
1531                 delete_vma_from_mm(vma);
1532                 delete_vma(mm, vma);
1533                 cond_resched();
1534         }
1535 }
1536
1537 int vm_brk(unsigned long addr, unsigned long len)
1538 {
1539         return -ENOMEM;
1540 }
1541
1542 /*
1543  * expand (or shrink) an existing mapping, potentially moving it at the same
1544  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1545  *
1546  * under NOMMU conditions, we only permit changing a mapping's size, and only
1547  * as long as it stays within the region allocated by do_mmap_private() and the
1548  * block is not shareable
1549  *
1550  * MREMAP_FIXED is not supported under NOMMU conditions
1551  */
1552 static unsigned long do_mremap(unsigned long addr,
1553                         unsigned long old_len, unsigned long new_len,
1554                         unsigned long flags, unsigned long new_addr)
1555 {
1556         struct vm_area_struct *vma;
1557
1558         /* insanity checks first */
1559         old_len = PAGE_ALIGN(old_len);
1560         new_len = PAGE_ALIGN(new_len);
1561         if (old_len == 0 || new_len == 0)
1562                 return (unsigned long) -EINVAL;
1563
1564         if (offset_in_page(addr))
1565                 return -EINVAL;
1566
1567         if (flags & MREMAP_FIXED && new_addr != addr)
1568                 return (unsigned long) -EINVAL;
1569
1570         vma = find_vma_exact(current->mm, addr, old_len);
1571         if (!vma)
1572                 return (unsigned long) -EINVAL;
1573
1574         if (vma->vm_end != vma->vm_start + old_len)
1575                 return (unsigned long) -EFAULT;
1576
1577         if (vma->vm_flags & VM_MAYSHARE)
1578                 return (unsigned long) -EPERM;
1579
1580         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1581                 return (unsigned long) -ENOMEM;
1582
1583         /* all checks complete - do it */
1584         vma->vm_end = vma->vm_start + new_len;
1585         return vma->vm_start;
1586 }
1587
1588 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1589                 unsigned long, new_len, unsigned long, flags,
1590                 unsigned long, new_addr)
1591 {
1592         unsigned long ret;
1593
1594         mmap_write_lock(current->mm);
1595         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1596         mmap_write_unlock(current->mm);
1597         return ret;
1598 }
1599
1600 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1601                          unsigned int foll_flags)
1602 {
1603         return NULL;
1604 }
1605
1606 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1607                 unsigned long pfn, unsigned long size, pgprot_t prot)
1608 {
1609         if (addr != (pfn << PAGE_SHIFT))
1610                 return -EINVAL;
1611
1612         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1613         return 0;
1614 }
1615 EXPORT_SYMBOL(remap_pfn_range);
1616
1617 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1618 {
1619         unsigned long pfn = start >> PAGE_SHIFT;
1620         unsigned long vm_len = vma->vm_end - vma->vm_start;
1621
1622         pfn += vma->vm_pgoff;
1623         return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1624 }
1625 EXPORT_SYMBOL(vm_iomap_memory);
1626
1627 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1628                         unsigned long pgoff)
1629 {
1630         unsigned int size = vma->vm_end - vma->vm_start;
1631
1632         if (!(vma->vm_flags & VM_USERMAP))
1633                 return -EINVAL;
1634
1635         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1636         vma->vm_end = vma->vm_start + size;
1637
1638         return 0;
1639 }
1640 EXPORT_SYMBOL(remap_vmalloc_range);
1641
1642 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1643         unsigned long len, unsigned long pgoff, unsigned long flags)
1644 {
1645         return -ENOMEM;
1646 }
1647
1648 vm_fault_t filemap_fault(struct vm_fault *vmf)
1649 {
1650         BUG();
1651         return 0;
1652 }
1653 EXPORT_SYMBOL(filemap_fault);
1654
1655 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
1656                 pgoff_t start_pgoff, pgoff_t end_pgoff)
1657 {
1658         BUG();
1659         return 0;
1660 }
1661 EXPORT_SYMBOL(filemap_map_pages);
1662
1663 int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
1664                        int len, unsigned int gup_flags)
1665 {
1666         struct vm_area_struct *vma;
1667         int write = gup_flags & FOLL_WRITE;
1668
1669         if (mmap_read_lock_killable(mm))
1670                 return 0;
1671
1672         /* the access must start within one of the target process's mappings */
1673         vma = find_vma(mm, addr);
1674         if (vma) {
1675                 /* don't overrun this mapping */
1676                 if (addr + len >= vma->vm_end)
1677                         len = vma->vm_end - addr;
1678
1679                 /* only read or write mappings where it is permitted */
1680                 if (write && vma->vm_flags & VM_MAYWRITE)
1681                         copy_to_user_page(vma, NULL, addr,
1682                                          (void *) addr, buf, len);
1683                 else if (!write && vma->vm_flags & VM_MAYREAD)
1684                         copy_from_user_page(vma, NULL, addr,
1685                                             buf, (void *) addr, len);
1686                 else
1687                         len = 0;
1688         } else {
1689                 len = 0;
1690         }
1691
1692         mmap_read_unlock(mm);
1693
1694         return len;
1695 }
1696
1697 /**
1698  * access_remote_vm - access another process' address space
1699  * @mm:         the mm_struct of the target address space
1700  * @addr:       start address to access
1701  * @buf:        source or destination buffer
1702  * @len:        number of bytes to transfer
1703  * @gup_flags:  flags modifying lookup behaviour
1704  *
1705  * The caller must hold a reference on @mm.
1706  */
1707 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1708                 void *buf, int len, unsigned int gup_flags)
1709 {
1710         return __access_remote_vm(mm, addr, buf, len, gup_flags);
1711 }
1712
1713 /*
1714  * Access another process' address space.
1715  * - source/target buffer must be kernel space
1716  */
1717 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1718                 unsigned int gup_flags)
1719 {
1720         struct mm_struct *mm;
1721
1722         if (addr + len < addr)
1723                 return 0;
1724
1725         mm = get_task_mm(tsk);
1726         if (!mm)
1727                 return 0;
1728
1729         len = __access_remote_vm(mm, addr, buf, len, gup_flags);
1730
1731         mmput(mm);
1732         return len;
1733 }
1734 EXPORT_SYMBOL_GPL(access_process_vm);
1735
1736 /**
1737  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1738  * @inode: The inode to check
1739  * @size: The current filesize of the inode
1740  * @newsize: The proposed filesize of the inode
1741  *
1742  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1743  * make sure that any outstanding VMAs aren't broken and then shrink the
1744  * vm_regions that extend beyond so that do_mmap() doesn't
1745  * automatically grant mappings that are too large.
1746  */
1747 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1748                                 size_t newsize)
1749 {
1750         struct vm_area_struct *vma;
1751         struct vm_region *region;
1752         pgoff_t low, high;
1753         size_t r_size, r_top;
1754
1755         low = newsize >> PAGE_SHIFT;
1756         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1757
1758         down_write(&nommu_region_sem);
1759         i_mmap_lock_read(inode->i_mapping);
1760
1761         /* search for VMAs that fall within the dead zone */
1762         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1763                 /* found one - only interested if it's shared out of the page
1764                  * cache */
1765                 if (vma->vm_flags & VM_SHARED) {
1766                         i_mmap_unlock_read(inode->i_mapping);
1767                         up_write(&nommu_region_sem);
1768                         return -ETXTBSY; /* not quite true, but near enough */
1769                 }
1770         }
1771
1772         /* reduce any regions that overlap the dead zone - if in existence,
1773          * these will be pointed to by VMAs that don't overlap the dead zone
1774          *
1775          * we don't check for any regions that start beyond the EOF as there
1776          * shouldn't be any
1777          */
1778         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1779                 if (!(vma->vm_flags & VM_SHARED))
1780                         continue;
1781
1782                 region = vma->vm_region;
1783                 r_size = region->vm_top - region->vm_start;
1784                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1785
1786                 if (r_top > newsize) {
1787                         region->vm_top -= r_top - newsize;
1788                         if (region->vm_end > region->vm_top)
1789                                 region->vm_end = region->vm_top;
1790                 }
1791         }
1792
1793         i_mmap_unlock_read(inode->i_mapping);
1794         up_write(&nommu_region_sem);
1795         return 0;
1796 }
1797
1798 /*
1799  * Initialise sysctl_user_reserve_kbytes.
1800  *
1801  * This is intended to prevent a user from starting a single memory hogging
1802  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1803  * mode.
1804  *
1805  * The default value is min(3% of free memory, 128MB)
1806  * 128MB is enough to recover with sshd/login, bash, and top/kill.
1807  */
1808 static int __meminit init_user_reserve(void)
1809 {
1810         unsigned long free_kbytes;
1811
1812         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1813
1814         sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1815         return 0;
1816 }
1817 subsys_initcall(init_user_reserve);
1818
1819 /*
1820  * Initialise sysctl_admin_reserve_kbytes.
1821  *
1822  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1823  * to log in and kill a memory hogging process.
1824  *
1825  * Systems with more than 256MB will reserve 8MB, enough to recover
1826  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1827  * only reserve 3% of free pages by default.
1828  */
1829 static int __meminit init_admin_reserve(void)
1830 {
1831         unsigned long free_kbytes;
1832
1833         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1834
1835         sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1836         return 0;
1837 }
1838 subsys_initcall(init_admin_reserve);