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