mm/nommu.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
kobjsize
follow_pfn
vfree
__vmalloc
__vmalloc_node_flags
vmalloc_user
vmalloc_to_page
vmalloc_to_pfn
vread
vwrite
vmalloc
vzalloc
vmalloc_node
vzalloc_node
vmalloc_exec
vmalloc_32
vmalloc_32_user
vmap
vunmap
vm_map_ram
vm_unmap_ram
vm_unmap_aliases
vmalloc_sync_mappings
vmalloc_sync_unmappings
alloc_vm_area
free_vm_area
vm_insert_page
vm_map_pages
vm_map_pages_zero
SYSCALL_DEFINE1
mmap_init
validate_nommu_regions
validate_nommu_regions
add_nommu_region
delete_nommu_region
free_page_series
__put_nommu_region
put_nommu_region
add_vma_to_mm
delete_vma_from_mm
delete_vma
find_vma
find_extend_vma
expand_stack
find_vma_exact
validate_mmap_request
determine_vm_flags
do_mmap_shared_file
do_mmap_private
do_mmap
ksys_mmap_pgoff
SYSCALL_DEFINE6
SYSCALL_DEFINE1
split_vma
shrink_vma
do_munmap
vm_munmap
SYSCALL_DEFINE2
exit_mmap
vm_brk
do_mremap
SYSCALL_DEFINE5
follow_page
remap_pfn_range
vm_iomap_memory
remap_vmalloc_range
arch_get_unmapped_area
filemap_fault
filemap_map_pages
__access_remote_vm
access_remote_vm
access_process_vm
nommu_shrink_inode_mappings
init_user_reserve
init_admin_reserve
   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/nommu-mmap.txt
   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, pgprot_t prot)
 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_flags(unsigned long size, int node, gfp_t flags)
 154 {
 155         return __vmalloc(size, flags, PAGE_KERNEL);
 156 }
 157 
 158 void *vmalloc_user(unsigned long size)
 159 {
 160         void *ret;
 161 
 162         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 163         if (ret) {
 164                 struct vm_area_struct *vma;
 165 
 166                 down_write(&current->mm->mmap_sem);
 167                 vma = find_vma(current->mm, (unsigned long)ret);
 168                 if (vma)
 169                         vma->vm_flags |= VM_USERMAP;
 170                 up_write(&current->mm->mmap_sem);
 171         }
 172 
 173         return ret;
 174 }
 175 EXPORT_SYMBOL(vmalloc_user);
 176 
 177 struct page *vmalloc_to_page(const void *addr)
 178 {
 179         return virt_to_page(addr);
 180 }
 181 EXPORT_SYMBOL(vmalloc_to_page);
 182 
 183 unsigned long vmalloc_to_pfn(const void *addr)
 184 {
 185         return page_to_pfn(virt_to_page(addr));
 186 }
 187 EXPORT_SYMBOL(vmalloc_to_pfn);
 188 
 189 long vread(char *buf, char *addr, unsigned long count)
 190 {
 191         /* Don't allow overflow */
 192         if ((unsigned long) buf + count < count)
 193                 count = -(unsigned long) buf;
 194 
 195         memcpy(buf, addr, count);
 196         return count;
 197 }
 198 
 199 long vwrite(char *buf, char *addr, unsigned long count)
 200 {
 201         /* Don't allow overflow */
 202         if ((unsigned long) addr + count < count)
 203                 count = -(unsigned long) addr;
 204 
 205         memcpy(addr, buf, count);
 206         return count;
 207 }
 208 
 209 /*
 210  *      vmalloc  -  allocate virtually contiguous memory
 211  *
 212  *      @size:          allocation size
 213  *
 214  *      Allocate enough pages to cover @size from the page level
 215  *      allocator and map them into contiguous kernel virtual space.
 216  *
 217  *      For tight control over page level allocator and protection flags
 218  *      use __vmalloc() instead.
 219  */
 220 void *vmalloc(unsigned long size)
 221 {
 222        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 223 }
 224 EXPORT_SYMBOL(vmalloc);
 225 
 226 /*
 227  *      vzalloc - allocate virtually contiguous memory with zero fill
 228  *
 229  *      @size:          allocation size
 230  *
 231  *      Allocate enough pages to cover @size from the page level
 232  *      allocator and map them into contiguous kernel virtual space.
 233  *      The memory allocated is set to zero.
 234  *
 235  *      For tight control over page level allocator and protection flags
 236  *      use __vmalloc() instead.
 237  */
 238 void *vzalloc(unsigned long size)
 239 {
 240         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
 241                         PAGE_KERNEL);
 242 }
 243 EXPORT_SYMBOL(vzalloc);
 244 
 245 /**
 246  * vmalloc_node - allocate memory on a specific node
 247  * @size:       allocation size
 248  * @node:       numa node
 249  *
 250  * Allocate enough pages to cover @size from the page level
 251  * allocator and map them into contiguous kernel virtual space.
 252  *
 253  * For tight control over page level allocator and protection flags
 254  * use __vmalloc() instead.
 255  */
 256 void *vmalloc_node(unsigned long size, int node)
 257 {
 258         return vmalloc(size);
 259 }
 260 EXPORT_SYMBOL(vmalloc_node);
 261 
 262 /**
 263  * vzalloc_node - allocate memory on a specific node with zero fill
 264  * @size:       allocation size
 265  * @node:       numa node
 266  *
 267  * Allocate enough pages to cover @size from the page level
 268  * allocator and map them into contiguous kernel virtual space.
 269  * The memory allocated is set to zero.
 270  *
 271  * For tight control over page level allocator and protection flags
 272  * use __vmalloc() instead.
 273  */
 274 void *vzalloc_node(unsigned long size, int node)
 275 {
 276         return vzalloc(size);
 277 }
 278 EXPORT_SYMBOL(vzalloc_node);
 279 
 280 /**
 281  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 282  *      @size:          allocation size
 283  *
 284  *      Kernel-internal function to allocate enough pages to cover @size
 285  *      the page level allocator and map them into contiguous and
 286  *      executable kernel virtual space.
 287  *
 288  *      For tight control over page level allocator and protection flags
 289  *      use __vmalloc() instead.
 290  */
 291 
 292 void *vmalloc_exec(unsigned long size)
 293 {
 294         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 295 }
 296 
 297 /**
 298  * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 299  *      @size:          allocation size
 300  *
 301  *      Allocate enough 32bit PA addressable pages to cover @size from the
 302  *      page level allocator and map them into contiguous kernel virtual space.
 303  */
 304 void *vmalloc_32(unsigned long size)
 305 {
 306         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 307 }
 308 EXPORT_SYMBOL(vmalloc_32);
 309 
 310 /**
 311  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 312  *      @size:          allocation size
 313  *
 314  * The resulting memory area is 32bit addressable and zeroed so it can be
 315  * mapped to userspace without leaking data.
 316  *
 317  * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 318  * remap_vmalloc_range() are permissible.
 319  */
 320 void *vmalloc_32_user(unsigned long size)
 321 {
 322         /*
 323          * We'll have to sort out the ZONE_DMA bits for 64-bit,
 324          * but for now this can simply use vmalloc_user() directly.
 325          */
 326         return vmalloc_user(size);
 327 }
 328 EXPORT_SYMBOL(vmalloc_32_user);
 329 
 330 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 331 {
 332         BUG();
 333         return NULL;
 334 }
 335 EXPORT_SYMBOL(vmap);
 336 
 337 void vunmap(const void *addr)
 338 {
 339         BUG();
 340 }
 341 EXPORT_SYMBOL(vunmap);
 342 
 343 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
 344 {
 345         BUG();
 346         return NULL;
 347 }
 348 EXPORT_SYMBOL(vm_map_ram);
 349 
 350 void vm_unmap_ram(const void *mem, unsigned int count)
 351 {
 352         BUG();
 353 }
 354 EXPORT_SYMBOL(vm_unmap_ram);
 355 
 356 void vm_unmap_aliases(void)
 357 {
 358 }
 359 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 360 
 361 /*
 362  * Implement a stub for vmalloc_sync_[un]mapping() if the architecture
 363  * chose not to have one.
 364  */
 365 void __weak vmalloc_sync_mappings(void)
 366 {
 367 }
 368 
 369 void __weak vmalloc_sync_unmappings(void)
 370 {
 371 }
 372 
 373 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
 374 {
 375         BUG();
 376         return NULL;
 377 }
 378 EXPORT_SYMBOL_GPL(alloc_vm_area);
 379 
 380 void free_vm_area(struct vm_struct *area)
 381 {
 382         BUG();
 383 }
 384 EXPORT_SYMBOL_GPL(free_vm_area);
 385 
 386 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 387                    struct page *page)
 388 {
 389         return -EINVAL;
 390 }
 391 EXPORT_SYMBOL(vm_insert_page);
 392 
 393 int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
 394                         unsigned long num)
 395 {
 396         return -EINVAL;
 397 }
 398 EXPORT_SYMBOL(vm_map_pages);
 399 
 400 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
 401                                 unsigned long num)
 402 {
 403         return -EINVAL;
 404 }
 405 EXPORT_SYMBOL(vm_map_pages_zero);
 406 
 407 /*
 408  *  sys_brk() for the most part doesn't need the global kernel
 409  *  lock, except when an application is doing something nasty
 410  *  like trying to un-brk an area that has already been mapped
 411  *  to a regular file.  in this case, the unmapping will need
 412  *  to invoke file system routines that need the global lock.
 413  */
 414 SYSCALL_DEFINE1(brk, unsigned long, brk)
 415 {
 416         struct mm_struct *mm = current->mm;
 417 
 418         if (brk < mm->start_brk || brk > mm->context.end_brk)
 419                 return mm->brk;
 420 
 421         if (mm->brk == brk)
 422                 return mm->brk;
 423 
 424         /*
 425          * Always allow shrinking brk
 426          */
 427         if (brk <= mm->brk) {
 428                 mm->brk = brk;
 429                 return brk;
 430         }
 431 
 432         /*
 433          * Ok, looks good - let it rip.
 434          */
 435         flush_icache_range(mm->brk, brk);
 436         return mm->brk = brk;
 437 }
 438 
 439 /*
 440  * initialise the percpu counter for VM and region record slabs
 441  */
 442 void __init mmap_init(void)
 443 {
 444         int ret;
 445 
 446         ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
 447         VM_BUG_ON(ret);
 448         vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
 449 }
 450 
 451 /*
 452  * validate the region tree
 453  * - the caller must hold the region lock
 454  */
 455 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
 456 static noinline void validate_nommu_regions(void)
 457 {
 458         struct vm_region *region, *last;
 459         struct rb_node *p, *lastp;
 460 
 461         lastp = rb_first(&nommu_region_tree);
 462         if (!lastp)
 463                 return;
 464 
 465         last = rb_entry(lastp, struct vm_region, vm_rb);
 466         BUG_ON(last->vm_end <= last->vm_start);
 467         BUG_ON(last->vm_top < last->vm_end);
 468 
 469         while ((p = rb_next(lastp))) {
 470                 region = rb_entry(p, struct vm_region, vm_rb);
 471                 last = rb_entry(lastp, struct vm_region, vm_rb);
 472 
 473                 BUG_ON(region->vm_end <= region->vm_start);
 474                 BUG_ON(region->vm_top < region->vm_end);
 475                 BUG_ON(region->vm_start < last->vm_top);
 476 
 477                 lastp = p;
 478         }
 479 }
 480 #else
 481 static void validate_nommu_regions(void)
 482 {
 483 }
 484 #endif
 485 
 486 /*
 487  * add a region into the global tree
 488  */
 489 static void add_nommu_region(struct vm_region *region)
 490 {
 491         struct vm_region *pregion;
 492         struct rb_node **p, *parent;
 493 
 494         validate_nommu_regions();
 495 
 496         parent = NULL;
 497         p = &nommu_region_tree.rb_node;
 498         while (*p) {
 499                 parent = *p;
 500                 pregion = rb_entry(parent, struct vm_region, vm_rb);
 501                 if (region->vm_start < pregion->vm_start)
 502                         p = &(*p)->rb_left;
 503                 else if (region->vm_start > pregion->vm_start)
 504                         p = &(*p)->rb_right;
 505                 else if (pregion == region)
 506                         return;
 507                 else
 508                         BUG();
 509         }
 510 
 511         rb_link_node(&region->vm_rb, parent, p);
 512         rb_insert_color(&region->vm_rb, &nommu_region_tree);
 513 
 514         validate_nommu_regions();
 515 }
 516 
 517 /*
 518  * delete a region from the global tree
 519  */
 520 static void delete_nommu_region(struct vm_region *region)
 521 {
 522         BUG_ON(!nommu_region_tree.rb_node);
 523 
 524         validate_nommu_regions();
 525         rb_erase(&region->vm_rb, &nommu_region_tree);
 526         validate_nommu_regions();
 527 }
 528 
 529 /*
 530  * free a contiguous series of pages
 531  */
 532 static void free_page_series(unsigned long from, unsigned long to)
 533 {
 534         for (; from < to; from += PAGE_SIZE) {
 535                 struct page *page = virt_to_page(from);
 536 
 537                 atomic_long_dec(&mmap_pages_allocated);
 538                 put_page(page);
 539         }
 540 }
 541 
 542 /*
 543  * release a reference to a region
 544  * - the caller must hold the region semaphore for writing, which this releases
 545  * - the region may not have been added to the tree yet, in which case vm_top
 546  *   will equal vm_start
 547  */
 548 static void __put_nommu_region(struct vm_region *region)
 549         __releases(nommu_region_sem)
 550 {
 551         BUG_ON(!nommu_region_tree.rb_node);
 552 
 553         if (--region->vm_usage == 0) {
 554                 if (region->vm_top > region->vm_start)
 555                         delete_nommu_region(region);
 556                 up_write(&nommu_region_sem);
 557 
 558                 if (region->vm_file)
 559                         fput(region->vm_file);
 560 
 561                 /* IO memory and memory shared directly out of the pagecache
 562                  * from ramfs/tmpfs mustn't be released here */
 563                 if (region->vm_flags & VM_MAPPED_COPY)
 564                         free_page_series(region->vm_start, region->vm_top);
 565                 kmem_cache_free(vm_region_jar, region);
 566         } else {
 567                 up_write(&nommu_region_sem);
 568         }
 569 }
 570 
 571 /*
 572  * release a reference to a region
 573  */
 574 static void put_nommu_region(struct vm_region *region)
 575 {
 576         down_write(&nommu_region_sem);
 577         __put_nommu_region(region);
 578 }
 579 
 580 /*
 581  * add a VMA into a process's mm_struct in the appropriate place in the list
 582  * and tree and add to the address space's page tree also if not an anonymous
 583  * page
 584  * - should be called with mm->mmap_sem held writelocked
 585  */
 586 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 587 {
 588         struct vm_area_struct *pvma, *prev;
 589         struct address_space *mapping;
 590         struct rb_node **p, *parent, *rb_prev;
 591 
 592         BUG_ON(!vma->vm_region);
 593 
 594         mm->map_count++;
 595         vma->vm_mm = mm;
 596 
 597         /* add the VMA to the mapping */
 598         if (vma->vm_file) {
 599                 mapping = vma->vm_file->f_mapping;
 600 
 601                 i_mmap_lock_write(mapping);
 602                 flush_dcache_mmap_lock(mapping);
 603                 vma_interval_tree_insert(vma, &mapping->i_mmap);
 604                 flush_dcache_mmap_unlock(mapping);
 605                 i_mmap_unlock_write(mapping);
 606         }
 607 
 608         /* add the VMA to the tree */
 609         parent = rb_prev = NULL;
 610         p = &mm->mm_rb.rb_node;
 611         while (*p) {
 612                 parent = *p;
 613                 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 614 
 615                 /* sort by: start addr, end addr, VMA struct addr in that order
 616                  * (the latter is necessary as we may get identical VMAs) */
 617                 if (vma->vm_start < pvma->vm_start)
 618                         p = &(*p)->rb_left;
 619                 else if (vma->vm_start > pvma->vm_start) {
 620                         rb_prev = parent;
 621                         p = &(*p)->rb_right;
 622                 } else if (vma->vm_end < pvma->vm_end)
 623                         p = &(*p)->rb_left;
 624                 else if (vma->vm_end > pvma->vm_end) {
 625                         rb_prev = parent;
 626                         p = &(*p)->rb_right;
 627                 } else if (vma < pvma)
 628                         p = &(*p)->rb_left;
 629                 else if (vma > pvma) {
 630                         rb_prev = parent;
 631                         p = &(*p)->rb_right;
 632                 } else
 633                         BUG();
 634         }
 635 
 636         rb_link_node(&vma->vm_rb, parent, p);
 637         rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 638 
 639         /* add VMA to the VMA list also */
 640         prev = NULL;
 641         if (rb_prev)
 642                 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 643 
 644         __vma_link_list(mm, vma, prev, parent);
 645 }
 646 
 647 /*
 648  * delete a VMA from its owning mm_struct and address space
 649  */
 650 static void delete_vma_from_mm(struct vm_area_struct *vma)
 651 {
 652         int i;
 653         struct address_space *mapping;
 654         struct mm_struct *mm = vma->vm_mm;
 655         struct task_struct *curr = current;
 656 
 657         mm->map_count--;
 658         for (i = 0; i < VMACACHE_SIZE; i++) {
 659                 /* if the vma is cached, invalidate the entire cache */
 660                 if (curr->vmacache.vmas[i] == vma) {
 661                         vmacache_invalidate(mm);
 662                         break;
 663                 }
 664         }
 665 
 666         /* remove the VMA from the mapping */
 667         if (vma->vm_file) {
 668                 mapping = vma->vm_file->f_mapping;
 669 
 670                 i_mmap_lock_write(mapping);
 671                 flush_dcache_mmap_lock(mapping);
 672                 vma_interval_tree_remove(vma, &mapping->i_mmap);
 673                 flush_dcache_mmap_unlock(mapping);
 674                 i_mmap_unlock_write(mapping);
 675         }
 676 
 677         /* remove from the MM's tree and list */
 678         rb_erase(&vma->vm_rb, &mm->mm_rb);
 679 
 680         if (vma->vm_prev)
 681                 vma->vm_prev->vm_next = vma->vm_next;
 682         else
 683                 mm->mmap = vma->vm_next;
 684 
 685         if (vma->vm_next)
 686                 vma->vm_next->vm_prev = vma->vm_prev;
 687 }
 688 
 689 /*
 690  * destroy a VMA record
 691  */
 692 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 693 {
 694         if (vma->vm_ops && vma->vm_ops->close)
 695                 vma->vm_ops->close(vma);
 696         if (vma->vm_file)
 697                 fput(vma->vm_file);
 698         put_nommu_region(vma->vm_region);
 699         vm_area_free(vma);
 700 }
 701 
 702 /*
 703  * look up the first VMA in which addr resides, NULL if none
 704  * - should be called with mm->mmap_sem at least held readlocked
 705  */
 706 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 707 {
 708         struct vm_area_struct *vma;
 709 
 710         /* check the cache first */
 711         vma = vmacache_find(mm, addr);
 712         if (likely(vma))
 713                 return vma;
 714 
 715         /* trawl the list (there may be multiple mappings in which addr
 716          * resides) */
 717         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 718                 if (vma->vm_start > addr)
 719                         return NULL;
 720                 if (vma->vm_end > addr) {
 721                         vmacache_update(addr, vma);
 722                         return vma;
 723                 }
 724         }
 725 
 726         return NULL;
 727 }
 728 EXPORT_SYMBOL(find_vma);
 729 
 730 /*
 731  * find a VMA
 732  * - we don't extend stack VMAs under NOMMU conditions
 733  */
 734 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 735 {
 736         return find_vma(mm, addr);
 737 }
 738 
 739 /*
 740  * expand a stack to a given address
 741  * - not supported under NOMMU conditions
 742  */
 743 int expand_stack(struct vm_area_struct *vma, unsigned long address)
 744 {
 745         return -ENOMEM;
 746 }
 747 
 748 /*
 749  * look up the first VMA exactly that exactly matches addr
 750  * - should be called with mm->mmap_sem at least held readlocked
 751  */
 752 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 753                                              unsigned long addr,
 754                                              unsigned long len)
 755 {
 756         struct vm_area_struct *vma;
 757         unsigned long end = addr + len;
 758 
 759         /* check the cache first */
 760         vma = vmacache_find_exact(mm, addr, end);
 761         if (vma)
 762                 return vma;
 763 
 764         /* trawl the list (there may be multiple mappings in which addr
 765          * resides) */
 766         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 767                 if (vma->vm_start < addr)
 768                         continue;
 769                 if (vma->vm_start > addr)
 770                         return NULL;
 771                 if (vma->vm_end == end) {
 772                         vmacache_update(addr, vma);
 773                         return vma;
 774                 }
 775         }
 776 
 777         return NULL;
 778 }
 779 
 780 /*
 781  * determine whether a mapping should be permitted and, if so, what sort of
 782  * mapping we're capable of supporting
 783  */
 784 static int validate_mmap_request(struct file *file,
 785                                  unsigned long addr,
 786                                  unsigned long len,
 787                                  unsigned long prot,
 788                                  unsigned long flags,
 789                                  unsigned long pgoff,
 790                                  unsigned long *_capabilities)
 791 {
 792         unsigned long capabilities, rlen;
 793         int ret;
 794 
 795         /* do the simple checks first */
 796         if (flags & MAP_FIXED)
 797                 return -EINVAL;
 798 
 799         if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 800             (flags & MAP_TYPE) != MAP_SHARED)
 801                 return -EINVAL;
 802 
 803         if (!len)
 804                 return -EINVAL;
 805 
 806         /* Careful about overflows.. */
 807         rlen = PAGE_ALIGN(len);
 808         if (!rlen || rlen > TASK_SIZE)
 809                 return -ENOMEM;
 810 
 811         /* offset overflow? */
 812         if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 813                 return -EOVERFLOW;
 814 
 815         if (file) {
 816                 /* files must support mmap */
 817                 if (!file->f_op->mmap)
 818                         return -ENODEV;
 819 
 820                 /* work out if what we've got could possibly be shared
 821                  * - we support chardevs that provide their own "memory"
 822                  * - we support files/blockdevs that are memory backed
 823                  */
 824                 if (file->f_op->mmap_capabilities) {
 825                         capabilities = file->f_op->mmap_capabilities(file);
 826                 } else {
 827                         /* no explicit capabilities set, so assume some
 828                          * defaults */
 829                         switch (file_inode(file)->i_mode & S_IFMT) {
 830                         case S_IFREG:
 831                         case S_IFBLK:
 832                                 capabilities = NOMMU_MAP_COPY;
 833                                 break;
 834 
 835                         case S_IFCHR:
 836                                 capabilities =
 837                                         NOMMU_MAP_DIRECT |
 838                                         NOMMU_MAP_READ |
 839                                         NOMMU_MAP_WRITE;
 840                                 break;
 841 
 842                         default:
 843                                 return -EINVAL;
 844                         }
 845                 }
 846 
 847                 /* eliminate any capabilities that we can't support on this
 848                  * device */
 849                 if (!file->f_op->get_unmapped_area)
 850                         capabilities &= ~NOMMU_MAP_DIRECT;
 851                 if (!(file->f_mode & FMODE_CAN_READ))
 852                         capabilities &= ~NOMMU_MAP_COPY;
 853 
 854                 /* The file shall have been opened with read permission. */
 855                 if (!(file->f_mode & FMODE_READ))
 856                         return -EACCES;
 857 
 858                 if (flags & MAP_SHARED) {
 859                         /* do checks for writing, appending and locking */
 860                         if ((prot & PROT_WRITE) &&
 861                             !(file->f_mode & FMODE_WRITE))
 862                                 return -EACCES;
 863 
 864                         if (IS_APPEND(file_inode(file)) &&
 865                             (file->f_mode & FMODE_WRITE))
 866                                 return -EACCES;
 867 
 868                         if (locks_verify_locked(file))
 869                                 return -EAGAIN;
 870 
 871                         if (!(capabilities & NOMMU_MAP_DIRECT))
 872                                 return -ENODEV;
 873 
 874                         /* we mustn't privatise shared mappings */
 875                         capabilities &= ~NOMMU_MAP_COPY;
 876                 } else {
 877                         /* we're going to read the file into private memory we
 878                          * allocate */
 879                         if (!(capabilities & NOMMU_MAP_COPY))
 880                                 return -ENODEV;
 881 
 882                         /* we don't permit a private writable mapping to be
 883                          * shared with the backing device */
 884                         if (prot & PROT_WRITE)
 885                                 capabilities &= ~NOMMU_MAP_DIRECT;
 886                 }
 887 
 888                 if (capabilities & NOMMU_MAP_DIRECT) {
 889                         if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
 890                             ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
 891                             ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
 892                             ) {
 893                                 capabilities &= ~NOMMU_MAP_DIRECT;
 894                                 if (flags & MAP_SHARED) {
 895                                         pr_warn("MAP_SHARED not completely supported on !MMU\n");
 896                                         return -EINVAL;
 897                                 }
 898                         }
 899                 }
 900 
 901                 /* handle executable mappings and implied executable
 902                  * mappings */
 903                 if (path_noexec(&file->f_path)) {
 904                         if (prot & PROT_EXEC)
 905                                 return -EPERM;
 906                 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 907                         /* handle implication of PROT_EXEC by PROT_READ */
 908                         if (current->personality & READ_IMPLIES_EXEC) {
 909                                 if (capabilities & NOMMU_MAP_EXEC)
 910                                         prot |= PROT_EXEC;
 911                         }
 912                 } else if ((prot & PROT_READ) &&
 913                          (prot & PROT_EXEC) &&
 914                          !(capabilities & NOMMU_MAP_EXEC)
 915                          ) {
 916                         /* backing file is not executable, try to copy */
 917                         capabilities &= ~NOMMU_MAP_DIRECT;
 918                 }
 919         } else {
 920                 /* anonymous mappings are always memory backed and can be
 921                  * privately mapped
 922                  */
 923                 capabilities = NOMMU_MAP_COPY;
 924 
 925                 /* handle PROT_EXEC implication by PROT_READ */
 926                 if ((prot & PROT_READ) &&
 927                     (current->personality & READ_IMPLIES_EXEC))
 928                         prot |= PROT_EXEC;
 929         }
 930 
 931         /* allow the security API to have its say */
 932         ret = security_mmap_addr(addr);
 933         if (ret < 0)
 934                 return ret;
 935 
 936         /* looks okay */
 937         *_capabilities = capabilities;
 938         return 0;
 939 }
 940 
 941 /*
 942  * we've determined that we can make the mapping, now translate what we
 943  * now know into VMA flags
 944  */
 945 static unsigned long determine_vm_flags(struct file *file,
 946                                         unsigned long prot,
 947                                         unsigned long flags,
 948                                         unsigned long capabilities)
 949 {
 950         unsigned long vm_flags;
 951 
 952         vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
 953         /* vm_flags |= mm->def_flags; */
 954 
 955         if (!(capabilities & NOMMU_MAP_DIRECT)) {
 956                 /* attempt to share read-only copies of mapped file chunks */
 957                 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 958                 if (file && !(prot & PROT_WRITE))
 959                         vm_flags |= VM_MAYSHARE;
 960         } else {
 961                 /* overlay a shareable mapping on the backing device or inode
 962                  * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 963                  * romfs/cramfs */
 964                 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
 965                 if (flags & MAP_SHARED)
 966                         vm_flags |= VM_SHARED;
 967         }
 968 
 969         /* refuse to let anyone share private mappings with this process if
 970          * it's being traced - otherwise breakpoints set in it may interfere
 971          * with another untraced process
 972          */
 973         if ((flags & MAP_PRIVATE) && current->ptrace)
 974                 vm_flags &= ~VM_MAYSHARE;
 975 
 976         return vm_flags;
 977 }
 978 
 979 /*
 980  * set up a shared mapping on a file (the driver or filesystem provides and
 981  * pins the storage)
 982  */
 983 static int do_mmap_shared_file(struct vm_area_struct *vma)
 984 {
 985         int ret;
 986 
 987         ret = call_mmap(vma->vm_file, vma);
 988         if (ret == 0) {
 989                 vma->vm_region->vm_top = vma->vm_region->vm_end;
 990                 return 0;
 991         }
 992         if (ret != -ENOSYS)
 993                 return ret;
 994 
 995         /* getting -ENOSYS indicates that direct mmap isn't possible (as
 996          * opposed to tried but failed) so we can only give a suitable error as
 997          * it's not possible to make a private copy if MAP_SHARED was given */
 998         return -ENODEV;
 999 }
1000 
1001 /*
1002  * set up a private mapping or an anonymous shared mapping
1003  */
1004 static int do_mmap_private(struct vm_area_struct *vma,
1005                            struct vm_region *region,
1006                            unsigned long len,
1007                            unsigned long capabilities)
1008 {
1009         unsigned long total, point;
1010         void *base;
1011         int ret, order;
1012 
1013         /* invoke the file's mapping function so that it can keep track of
1014          * shared mappings on devices or memory
1015          * - VM_MAYSHARE will be set if it may attempt to share
1016          */
1017         if (capabilities & NOMMU_MAP_DIRECT) {
1018                 ret = call_mmap(vma->vm_file, vma);
1019                 if (ret == 0) {
1020                         /* shouldn't return success if we're not sharing */
1021                         BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1022                         vma->vm_region->vm_top = vma->vm_region->vm_end;
1023                         return 0;
1024                 }
1025                 if (ret != -ENOSYS)
1026                         return ret;
1027 
1028                 /* getting an ENOSYS error indicates that direct mmap isn't
1029                  * possible (as opposed to tried but failed) so we'll try to
1030                  * make a private copy of the data and map that instead */
1031         }
1032 
1033 
1034         /* allocate some memory to hold the mapping
1035          * - note that this may not return a page-aligned address if the object
1036          *   we're allocating is smaller than a page
1037          */
1038         order = get_order(len);
1039         total = 1 << order;
1040         point = len >> PAGE_SHIFT;
1041 
1042         /* we don't want to allocate a power-of-2 sized page set */
1043         if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1044                 total = point;
1045 
1046         base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1047         if (!base)
1048                 goto enomem;
1049 
1050         atomic_long_add(total, &mmap_pages_allocated);
1051 
1052         region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1053         region->vm_start = (unsigned long) base;
1054         region->vm_end   = region->vm_start + len;
1055         region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1056 
1057         vma->vm_start = region->vm_start;
1058         vma->vm_end   = region->vm_start + len;
1059 
1060         if (vma->vm_file) {
1061                 /* read the contents of a file into the copy */
1062                 loff_t fpos;
1063 
1064                 fpos = vma->vm_pgoff;
1065                 fpos <<= PAGE_SHIFT;
1066 
1067                 ret = kernel_read(vma->vm_file, base, len, &fpos);
1068                 if (ret < 0)
1069                         goto error_free;
1070 
1071                 /* clear the last little bit */
1072                 if (ret < len)
1073                         memset(base + ret, 0, len - ret);
1074 
1075         } else {
1076                 vma_set_anonymous(vma);
1077         }
1078 
1079         return 0;
1080 
1081 error_free:
1082         free_page_series(region->vm_start, region->vm_top);
1083         region->vm_start = vma->vm_start = 0;
1084         region->vm_end   = vma->vm_end = 0;
1085         region->vm_top   = 0;
1086         return ret;
1087 
1088 enomem:
1089         pr_err("Allocation of length %lu from process %d (%s) failed\n",
1090                len, current->pid, current->comm);
1091         show_free_areas(0, NULL);
1092         return -ENOMEM;
1093 }
1094 
1095 /*
1096  * handle mapping creation for uClinux
1097  */
1098 unsigned long do_mmap(struct file *file,
1099                         unsigned long addr,
1100                         unsigned long len,
1101                         unsigned long prot,
1102                         unsigned long flags,
1103                         vm_flags_t vm_flags,
1104                         unsigned long pgoff,
1105                         unsigned long *populate,
1106                         struct list_head *uf)
1107 {
1108         struct vm_area_struct *vma;
1109         struct vm_region *region;
1110         struct rb_node *rb;
1111         unsigned long capabilities, result;
1112         int ret;
1113 
1114         *populate = 0;
1115 
1116         /* decide whether we should attempt the mapping, and if so what sort of
1117          * mapping */
1118         ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1119                                     &capabilities);
1120         if (ret < 0)
1121                 return ret;
1122 
1123         /* we ignore the address hint */
1124         addr = 0;
1125         len = PAGE_ALIGN(len);
1126 
1127         /* we've determined that we can make the mapping, now translate what we
1128          * now know into VMA flags */
1129         vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1130 
1131         /* we're going to need to record the mapping */
1132         region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1133         if (!region)
1134                 goto error_getting_region;
1135 
1136         vma = vm_area_alloc(current->mm);
1137         if (!vma)
1138                 goto error_getting_vma;
1139 
1140         region->vm_usage = 1;
1141         region->vm_flags = vm_flags;
1142         region->vm_pgoff = pgoff;
1143 
1144         vma->vm_flags = vm_flags;
1145         vma->vm_pgoff = pgoff;
1146 
1147         if (file) {
1148                 region->vm_file = get_file(file);
1149                 vma->vm_file = get_file(file);
1150         }
1151 
1152         down_write(&nommu_region_sem);
1153 
1154         /* if we want to share, we need to check for regions created by other
1155          * mmap() calls that overlap with our proposed mapping
1156          * - we can only share with a superset match on most regular files
1157          * - shared mappings on character devices and memory backed files are
1158          *   permitted to overlap inexactly as far as we are concerned for in
1159          *   these cases, sharing is handled in the driver or filesystem rather
1160          *   than here
1161          */
1162         if (vm_flags & VM_MAYSHARE) {
1163                 struct vm_region *pregion;
1164                 unsigned long pglen, rpglen, pgend, rpgend, start;
1165 
1166                 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1167                 pgend = pgoff + pglen;
1168 
1169                 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1170                         pregion = rb_entry(rb, struct vm_region, vm_rb);
1171 
1172                         if (!(pregion->vm_flags & VM_MAYSHARE))
1173                                 continue;
1174 
1175                         /* search for overlapping mappings on the same file */
1176                         if (file_inode(pregion->vm_file) !=
1177                             file_inode(file))
1178                                 continue;
1179 
1180                         if (pregion->vm_pgoff >= pgend)
1181                                 continue;
1182 
1183                         rpglen = pregion->vm_end - pregion->vm_start;
1184                         rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1185                         rpgend = pregion->vm_pgoff + rpglen;
1186                         if (pgoff >= rpgend)
1187                                 continue;
1188 
1189                         /* handle inexactly overlapping matches between
1190                          * mappings */
1191                         if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1192                             !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1193                                 /* new mapping is not a subset of the region */
1194                                 if (!(capabilities & NOMMU_MAP_DIRECT))
1195                                         goto sharing_violation;
1196                                 continue;
1197                         }
1198 
1199                         /* we've found a region we can share */
1200                         pregion->vm_usage++;
1201                         vma->vm_region = pregion;
1202                         start = pregion->vm_start;
1203                         start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1204                         vma->vm_start = start;
1205                         vma->vm_end = start + len;
1206 
1207                         if (pregion->vm_flags & VM_MAPPED_COPY)
1208                                 vma->vm_flags |= VM_MAPPED_COPY;
1209                         else {
1210                                 ret = do_mmap_shared_file(vma);
1211                                 if (ret < 0) {
1212                                         vma->vm_region = NULL;
1213                                         vma->vm_start = 0;
1214                                         vma->vm_end = 0;
1215                                         pregion->vm_usage--;
1216                                         pregion = NULL;
1217                                         goto error_just_free;
1218                                 }
1219                         }
1220                         fput(region->vm_file);
1221                         kmem_cache_free(vm_region_jar, region);
1222                         region = pregion;
1223                         result = start;
1224                         goto share;
1225                 }
1226 
1227                 /* obtain the address at which to make a shared mapping
1228                  * - this is the hook for quasi-memory character devices to
1229                  *   tell us the location of a shared mapping
1230                  */
1231                 if (capabilities & NOMMU_MAP_DIRECT) {
1232                         addr = file->f_op->get_unmapped_area(file, addr, len,
1233                                                              pgoff, flags);
1234                         if (IS_ERR_VALUE(addr)) {
1235                                 ret = addr;
1236                                 if (ret != -ENOSYS)
1237                                         goto error_just_free;
1238 
1239                                 /* the driver refused to tell us where to site
1240                                  * the mapping so we'll have to attempt to copy
1241                                  * it */
1242                                 ret = -ENODEV;
1243                                 if (!(capabilities & NOMMU_MAP_COPY))
1244                                         goto error_just_free;
1245 
1246                                 capabilities &= ~NOMMU_MAP_DIRECT;
1247                         } else {
1248                                 vma->vm_start = region->vm_start = addr;
1249                                 vma->vm_end = region->vm_end = addr + len;
1250                         }
1251                 }
1252         }
1253 
1254         vma->vm_region = region;
1255 
1256         /* set up the mapping
1257          * - the region is filled in if NOMMU_MAP_DIRECT is still set
1258          */
1259         if (file && vma->vm_flags & VM_SHARED)
1260                 ret = do_mmap_shared_file(vma);
1261         else
1262                 ret = do_mmap_private(vma, region, len, capabilities);
1263         if (ret < 0)
1264                 goto error_just_free;
1265         add_nommu_region(region);
1266 
1267         /* clear anonymous mappings that don't ask for uninitialized data */
1268         if (!vma->vm_file &&
1269             (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1270              !(flags & MAP_UNINITIALIZED)))
1271                 memset((void *)region->vm_start, 0,
1272                        region->vm_end - region->vm_start);
1273 
1274         /* okay... we have a mapping; now we have to register it */
1275         result = vma->vm_start;
1276 
1277         current->mm->total_vm += len >> PAGE_SHIFT;
1278 
1279 share:
1280         add_vma_to_mm(current->mm, vma);
1281 
1282         /* we flush the region from the icache only when the first executable
1283          * mapping of it is made  */
1284         if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1285                 flush_icache_range(region->vm_start, region->vm_end);
1286                 region->vm_icache_flushed = true;
1287         }
1288 
1289         up_write(&nommu_region_sem);
1290 
1291         return result;
1292 
1293 error_just_free:
1294         up_write(&nommu_region_sem);
1295 error:
1296         if (region->vm_file)
1297                 fput(region->vm_file);
1298         kmem_cache_free(vm_region_jar, region);
1299         if (vma->vm_file)
1300                 fput(vma->vm_file);
1301         vm_area_free(vma);
1302         return ret;
1303 
1304 sharing_violation:
1305         up_write(&nommu_region_sem);
1306         pr_warn("Attempt to share mismatched mappings\n");
1307         ret = -EINVAL;
1308         goto error;
1309 
1310 error_getting_vma:
1311         kmem_cache_free(vm_region_jar, region);
1312         pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1313                         len, current->pid);
1314         show_free_areas(0, NULL);
1315         return -ENOMEM;
1316 
1317 error_getting_region:
1318         pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1319                         len, current->pid);
1320         show_free_areas(0, NULL);
1321         return -ENOMEM;
1322 }
1323 
1324 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1325                               unsigned long prot, unsigned long flags,
1326                               unsigned long fd, unsigned long pgoff)
1327 {
1328         struct file *file = NULL;
1329         unsigned long retval = -EBADF;
1330 
1331         audit_mmap_fd(fd, flags);
1332         if (!(flags & MAP_ANONYMOUS)) {
1333                 file = fget(fd);
1334                 if (!file)
1335                         goto out;
1336         }
1337 
1338         flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1339 
1340         retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1341 
1342         if (file)
1343                 fput(file);
1344 out:
1345         return retval;
1346 }
1347 
1348 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1349                 unsigned long, prot, unsigned long, flags,
1350                 unsigned long, fd, unsigned long, pgoff)
1351 {
1352         return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1353 }
1354 
1355 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1356 struct mmap_arg_struct {
1357         unsigned long addr;
1358         unsigned long len;
1359         unsigned long prot;
1360         unsigned long flags;
1361         unsigned long fd;
1362         unsigned long offset;
1363 };
1364 
1365 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1366 {
1367         struct mmap_arg_struct a;
1368 
1369         if (copy_from_user(&a, arg, sizeof(a)))
1370                 return -EFAULT;
1371         if (offset_in_page(a.offset))
1372                 return -EINVAL;
1373 
1374         return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1375                                a.offset >> PAGE_SHIFT);
1376 }
1377 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1378 
1379 /*
1380  * split a vma into two pieces at address 'addr', a new vma is allocated either
1381  * for the first part or the tail.
1382  */
1383 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1384               unsigned long addr, int new_below)
1385 {
1386         struct vm_area_struct *new;
1387         struct vm_region *region;
1388         unsigned long npages;
1389 
1390         /* we're only permitted to split anonymous regions (these should have
1391          * only a single usage on the region) */
1392         if (vma->vm_file)
1393                 return -ENOMEM;
1394 
1395         if (mm->map_count >= sysctl_max_map_count)
1396                 return -ENOMEM;
1397 
1398         region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1399         if (!region)
1400                 return -ENOMEM;
1401 
1402         new = vm_area_dup(vma);
1403         if (!new) {
1404                 kmem_cache_free(vm_region_jar, region);
1405                 return -ENOMEM;
1406         }
1407 
1408         /* most fields are the same, copy all, and then fixup */
1409         *region = *vma->vm_region;
1410         new->vm_region = region;
1411 
1412         npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1413 
1414         if (new_below) {
1415                 region->vm_top = region->vm_end = new->vm_end = addr;
1416         } else {
1417                 region->vm_start = new->vm_start = addr;
1418                 region->vm_pgoff = new->vm_pgoff += npages;
1419         }
1420 
1421         if (new->vm_ops && new->vm_ops->open)
1422                 new->vm_ops->open(new);
1423 
1424         delete_vma_from_mm(vma);
1425         down_write(&nommu_region_sem);
1426         delete_nommu_region(vma->vm_region);
1427         if (new_below) {
1428                 vma->vm_region->vm_start = vma->vm_start = addr;
1429                 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1430         } else {
1431                 vma->vm_region->vm_end = vma->vm_end = addr;
1432                 vma->vm_region->vm_top = addr;
1433         }
1434         add_nommu_region(vma->vm_region);
1435         add_nommu_region(new->vm_region);
1436         up_write(&nommu_region_sem);
1437         add_vma_to_mm(mm, vma);
1438         add_vma_to_mm(mm, new);
1439         return 0;
1440 }
1441 
1442 /*
1443  * shrink a VMA by removing the specified chunk from either the beginning or
1444  * the end
1445  */
1446 static int shrink_vma(struct mm_struct *mm,
1447                       struct vm_area_struct *vma,
1448                       unsigned long from, unsigned long to)
1449 {
1450         struct vm_region *region;
1451 
1452         /* adjust the VMA's pointers, which may reposition it in the MM's tree
1453          * and list */
1454         delete_vma_from_mm(vma);
1455         if (from > vma->vm_start)
1456                 vma->vm_end = from;
1457         else
1458                 vma->vm_start = to;
1459         add_vma_to_mm(mm, vma);
1460 
1461         /* cut the backing region down to size */
1462         region = vma->vm_region;
1463         BUG_ON(region->vm_usage != 1);
1464 
1465         down_write(&nommu_region_sem);
1466         delete_nommu_region(region);
1467         if (from > region->vm_start) {
1468                 to = region->vm_top;
1469                 region->vm_top = region->vm_end = from;
1470         } else {
1471                 region->vm_start = to;
1472         }
1473         add_nommu_region(region);
1474         up_write(&nommu_region_sem);
1475 
1476         free_page_series(from, to);
1477         return 0;
1478 }
1479 
1480 /*
1481  * release a mapping
1482  * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1483  *   VMA, though it need not cover the whole VMA
1484  */
1485 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1486 {
1487         struct vm_area_struct *vma;
1488         unsigned long end;
1489         int ret;
1490 
1491         len = PAGE_ALIGN(len);
1492         if (len == 0)
1493                 return -EINVAL;
1494 
1495         end = start + len;
1496 
1497         /* find the first potentially overlapping VMA */
1498         vma = find_vma(mm, start);
1499         if (!vma) {
1500                 static int limit;
1501                 if (limit < 5) {
1502                         pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1503                                         current->pid, current->comm,
1504                                         start, start + len - 1);
1505                         limit++;
1506                 }
1507                 return -EINVAL;
1508         }
1509 
1510         /* we're allowed to split an anonymous VMA but not a file-backed one */
1511         if (vma->vm_file) {
1512                 do {
1513                         if (start > vma->vm_start)
1514                                 return -EINVAL;
1515                         if (end == vma->vm_end)
1516                                 goto erase_whole_vma;
1517                         vma = vma->vm_next;
1518                 } while (vma);
1519                 return -EINVAL;
1520         } else {
1521                 /* the chunk must be a subset of the VMA found */
1522                 if (start == vma->vm_start && end == vma->vm_end)
1523                         goto erase_whole_vma;
1524                 if (start < vma->vm_start || end > vma->vm_end)
1525                         return -EINVAL;
1526                 if (offset_in_page(start))
1527                         return -EINVAL;
1528                 if (end != vma->vm_end && offset_in_page(end))
1529                         return -EINVAL;
1530                 if (start != vma->vm_start && end != vma->vm_end) {
1531                         ret = split_vma(mm, vma, start, 1);
1532                         if (ret < 0)
1533                                 return ret;
1534                 }
1535                 return shrink_vma(mm, vma, start, end);
1536         }
1537 
1538 erase_whole_vma:
1539         delete_vma_from_mm(vma);
1540         delete_vma(mm, vma);
1541         return 0;
1542 }
1543 EXPORT_SYMBOL(do_munmap);
1544 
1545 int vm_munmap(unsigned long addr, size_t len)
1546 {
1547         struct mm_struct *mm = current->mm;
1548         int ret;
1549 
1550         down_write(&mm->mmap_sem);
1551         ret = do_munmap(mm, addr, len, NULL);
1552         up_write(&mm->mmap_sem);
1553         return ret;
1554 }
1555 EXPORT_SYMBOL(vm_munmap);
1556 
1557 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1558 {
1559         return vm_munmap(addr, len);
1560 }
1561 
1562 /*
1563  * release all the mappings made in a process's VM space
1564  */
1565 void exit_mmap(struct mm_struct *mm)
1566 {
1567         struct vm_area_struct *vma;
1568 
1569         if (!mm)
1570                 return;
1571 
1572         mm->total_vm = 0;
1573 
1574         while ((vma = mm->mmap)) {
1575                 mm->mmap = vma->vm_next;
1576                 delete_vma_from_mm(vma);
1577                 delete_vma(mm, vma);
1578                 cond_resched();
1579         }
1580 }
1581 
1582 int vm_brk(unsigned long addr, unsigned long len)
1583 {
1584         return -ENOMEM;
1585 }
1586 
1587 /*
1588  * expand (or shrink) an existing mapping, potentially moving it at the same
1589  * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1590  *
1591  * under NOMMU conditions, we only permit changing a mapping's size, and only
1592  * as long as it stays within the region allocated by do_mmap_private() and the
1593  * block is not shareable
1594  *
1595  * MREMAP_FIXED is not supported under NOMMU conditions
1596  */
1597 static unsigned long do_mremap(unsigned long addr,
1598                         unsigned long old_len, unsigned long new_len,
1599                         unsigned long flags, unsigned long new_addr)
1600 {
1601         struct vm_area_struct *vma;
1602 
1603         /* insanity checks first */
1604         old_len = PAGE_ALIGN(old_len);
1605         new_len = PAGE_ALIGN(new_len);
1606         if (old_len == 0 || new_len == 0)
1607                 return (unsigned long) -EINVAL;
1608 
1609         if (offset_in_page(addr))
1610                 return -EINVAL;
1611 
1612         if (flags & MREMAP_FIXED && new_addr != addr)
1613                 return (unsigned long) -EINVAL;
1614 
1615         vma = find_vma_exact(current->mm, addr, old_len);
1616         if (!vma)
1617                 return (unsigned long) -EINVAL;
1618 
1619         if (vma->vm_end != vma->vm_start + old_len)
1620                 return (unsigned long) -EFAULT;
1621 
1622         if (vma->vm_flags & VM_MAYSHARE)
1623                 return (unsigned long) -EPERM;
1624 
1625         if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1626                 return (unsigned long) -ENOMEM;
1627 
1628         /* all checks complete - do it */
1629         vma->vm_end = vma->vm_start + new_len;
1630         return vma->vm_start;
1631 }
1632 
1633 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1634                 unsigned long, new_len, unsigned long, flags,
1635                 unsigned long, new_addr)
1636 {
1637         unsigned long ret;
1638 
1639         down_write(&current->mm->mmap_sem);
1640         ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1641         up_write(&current->mm->mmap_sem);
1642         return ret;
1643 }
1644 
1645 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1646                          unsigned int foll_flags)
1647 {
1648         return NULL;
1649 }
1650 
1651 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1652                 unsigned long pfn, unsigned long size, pgprot_t prot)
1653 {
1654         if (addr != (pfn << PAGE_SHIFT))
1655                 return -EINVAL;
1656 
1657         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1658         return 0;
1659 }
1660 EXPORT_SYMBOL(remap_pfn_range);
1661 
1662 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1663 {
1664         unsigned long pfn = start >> PAGE_SHIFT;
1665         unsigned long vm_len = vma->vm_end - vma->vm_start;
1666 
1667         pfn += vma->vm_pgoff;
1668         return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1669 }
1670 EXPORT_SYMBOL(vm_iomap_memory);
1671 
1672 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1673                         unsigned long pgoff)
1674 {
1675         unsigned int size = vma->vm_end - vma->vm_start;
1676 
1677         if (!(vma->vm_flags & VM_USERMAP))
1678                 return -EINVAL;
1679 
1680         vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1681         vma->vm_end = vma->vm_start + size;
1682 
1683         return 0;
1684 }
1685 EXPORT_SYMBOL(remap_vmalloc_range);
1686 
1687 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1688         unsigned long len, unsigned long pgoff, unsigned long flags)
1689 {
1690         return -ENOMEM;
1691 }
1692 
1693 vm_fault_t filemap_fault(struct vm_fault *vmf)
1694 {
1695         BUG();
1696         return 0;
1697 }
1698 EXPORT_SYMBOL(filemap_fault);
1699 
1700 void filemap_map_pages(struct vm_fault *vmf,
1701                 pgoff_t start_pgoff, pgoff_t end_pgoff)
1702 {
1703         BUG();
1704 }
1705 EXPORT_SYMBOL(filemap_map_pages);
1706 
1707 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1708                 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1709 {
1710         struct vm_area_struct *vma;
1711         int write = gup_flags & FOLL_WRITE;
1712 
1713         if (down_read_killable(&mm->mmap_sem))
1714                 return 0;
1715 
1716         /* the access must start within one of the target process's mappings */
1717         vma = find_vma(mm, addr);
1718         if (vma) {
1719                 /* don't overrun this mapping */
1720                 if (addr + len >= vma->vm_end)
1721                         len = vma->vm_end - addr;
1722 
1723                 /* only read or write mappings where it is permitted */
1724                 if (write && vma->vm_flags & VM_MAYWRITE)
1725                         copy_to_user_page(vma, NULL, addr,
1726                                          (void *) addr, buf, len);
1727                 else if (!write && vma->vm_flags & VM_MAYREAD)
1728                         copy_from_user_page(vma, NULL, addr,
1729                                             buf, (void *) addr, len);
1730                 else
1731                         len = 0;
1732         } else {
1733                 len = 0;
1734         }
1735 
1736         up_read(&mm->mmap_sem);
1737 
1738         return len;
1739 }
1740 
1741 /**
1742  * access_remote_vm - access another process' address space
1743  * @mm:         the mm_struct of the target address space
1744  * @addr:       start address to access
1745  * @buf:        source or destination buffer
1746  * @len:        number of bytes to transfer
1747  * @gup_flags:  flags modifying lookup behaviour
1748  *
1749  * The caller must hold a reference on @mm.
1750  */
1751 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1752                 void *buf, int len, unsigned int gup_flags)
1753 {
1754         return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1755 }
1756 
1757 /*
1758  * Access another process' address space.
1759  * - source/target buffer must be kernel space
1760  */
1761 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1762                 unsigned int gup_flags)
1763 {
1764         struct mm_struct *mm;
1765 
1766         if (addr + len < addr)
1767                 return 0;
1768 
1769         mm = get_task_mm(tsk);
1770         if (!mm)
1771                 return 0;
1772 
1773         len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
1774 
1775         mmput(mm);
1776         return len;
1777 }
1778 EXPORT_SYMBOL_GPL(access_process_vm);
1779 
1780 /**
1781  * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1782  * @inode: The inode to check
1783  * @size: The current filesize of the inode
1784  * @newsize: The proposed filesize of the inode
1785  *
1786  * Check the shared mappings on an inode on behalf of a shrinking truncate to
1787  * make sure that that any outstanding VMAs aren't broken and then shrink the
1788  * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1789  * automatically grant mappings that are too large.
1790  */
1791 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1792                                 size_t newsize)
1793 {
1794         struct vm_area_struct *vma;
1795         struct vm_region *region;
1796         pgoff_t low, high;
1797         size_t r_size, r_top;
1798 
1799         low = newsize >> PAGE_SHIFT;
1800         high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1801 
1802         down_write(&nommu_region_sem);
1803         i_mmap_lock_read(inode->i_mapping);
1804 
1805         /* search for VMAs that fall within the dead zone */
1806         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1807                 /* found one - only interested if it's shared out of the page
1808                  * cache */
1809                 if (vma->vm_flags & VM_SHARED) {
1810                         i_mmap_unlock_read(inode->i_mapping);
1811                         up_write(&nommu_region_sem);
1812                         return -ETXTBSY; /* not quite true, but near enough */
1813                 }
1814         }
1815 
1816         /* reduce any regions that overlap the dead zone - if in existence,
1817          * these will be pointed to by VMAs that don't overlap the dead zone
1818          *
1819          * we don't check for any regions that start beyond the EOF as there
1820          * shouldn't be any
1821          */
1822         vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1823                 if (!(vma->vm_flags & VM_SHARED))
1824                         continue;
1825 
1826                 region = vma->vm_region;
1827                 r_size = region->vm_top - region->vm_start;
1828                 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1829 
1830                 if (r_top > newsize) {
1831                         region->vm_top -= r_top - newsize;
1832                         if (region->vm_end > region->vm_top)
1833                                 region->vm_end = region->vm_top;
1834                 }
1835         }
1836 
1837         i_mmap_unlock_read(inode->i_mapping);
1838         up_write(&nommu_region_sem);
1839         return 0;
1840 }
1841 
1842 /*
1843  * Initialise sysctl_user_reserve_kbytes.
1844  *
1845  * This is intended to prevent a user from starting a single memory hogging
1846  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1847  * mode.
1848  *
1849  * The default value is min(3% of free memory, 128MB)
1850  * 128MB is enough to recover with sshd/login, bash, and top/kill.
1851  */
1852 static int __meminit init_user_reserve(void)
1853 {
1854         unsigned long free_kbytes;
1855 
1856         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1857 
1858         sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1859         return 0;
1860 }
1861 subsys_initcall(init_user_reserve);
1862 
1863 /*
1864  * Initialise sysctl_admin_reserve_kbytes.
1865  *
1866  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1867  * to log in and kill a memory hogging process.
1868  *
1869  * Systems with more than 256MB will reserve 8MB, enough to recover
1870  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1871  * only reserve 3% of free pages by default.
1872  */
1873 static int __meminit init_admin_reserve(void)
1874 {
1875         unsigned long free_kbytes;
1876 
1877         free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1878 
1879         sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1880         return 0;
1881 }
1882 subsys_initcall(init_admin_reserve);
/* [<][>][^][v][top][bottom][index][help] */
root/mm/nommu.c

DEFINITIONS
