1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/mm/nommu.c
   4 *
   5 *  Replacement code for mm functions to support CPU's that don't
   6 *  have any form of memory management unit (thus no virtual memory).
   7 *
   8 *  See Documentation/admin-guide/mm/nommu-mmap.rst
   9 *
  10 *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
  11 *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  12 *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  13 *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  14 *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
  15 */
  16
  17#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  18
  19#include <linux/export.h>
  20#include <linux/mm.h>
  21#include <linux/sched/mm.h>
  22#include <linux/vmacache.h>
  23#include <linux/mman.h>
  24#include <linux/swap.h>
  25#include <linux/file.h>
  26#include <linux/highmem.h>
  27#include <linux/pagemap.h>
  28#include <linux/slab.h>
  29#include <linux/vmalloc.h>
  30#include <linux/blkdev.h>
  31#include <linux/backing-dev.h>
  32#include <linux/compiler.h>
  33#include <linux/mount.h>
  34#include <linux/personality.h>
  35#include <linux/security.h>
  36#include <linux/syscalls.h>
  37#include <linux/audit.h>
  38#include <linux/printk.h>
  39
  40#include <linux/uaccess.h>
  41#include <asm/tlb.h>
  42#include <asm/tlbflush.h>
  43#include <asm/mmu_context.h>
  44#include "internal.h"
  45
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  46void *high_memory;
  47EXPORT_SYMBOL(high_memory);
  48struct page *mem_map;
  49unsigned long max_mapnr;
  50EXPORT_SYMBOL(max_mapnr);
  51unsigned long highest_memmap_pfn;
 
 
 
 
  52int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
  53int heap_stack_gap = 0;
  54
  55atomic_long_t mmap_pages_allocated;
  56
  57EXPORT_SYMBOL(mem_map);
 
  58
  59/* list of mapped, potentially shareable regions */
  60static struct kmem_cache *vm_region_jar;
  61struct rb_root nommu_region_tree = RB_ROOT;
  62DECLARE_RWSEM(nommu_region_sem);
  63
  64const 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 */
  73unsigned 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 */
 124int 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}
 133EXPORT_SYMBOL(follow_pfn);
 134
 135LIST_HEAD(vmap_area_list);
 
 136
 137void vfree(const void *addr)
 138{
 139	kfree(addr);
 140}
 141EXPORT_SYMBOL(vfree);
 142
 143void *__vmalloc(unsigned long size, gfp_t gfp_mask)
 144{
 145	/*
 146	 *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 147	 * returns only a logical address.
 148	 */
 149	return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 150}
 151EXPORT_SYMBOL(__vmalloc);
 152
 153void *__vmalloc_node_range(unsigned long size, unsigned long align,
 154		unsigned long start, unsigned long end, gfp_t gfp_mask,
 155		pgprot_t prot, unsigned long vm_flags, int node,
 156		const void *caller)
 157{
 158	return __vmalloc(size, gfp_mask);
 159}
 160
 161void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
 162		int node, const void *caller)
 163{
 164	return __vmalloc(size, gfp_mask);
 165}
 166
 167static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
 168{
 169	void *ret;
 170
 171	ret = __vmalloc(size, flags);
 
 172	if (ret) {
 173		struct vm_area_struct *vma;
 174
 175		mmap_write_lock(current->mm);
 176		vma = find_vma(current->mm, (unsigned long)ret);
 177		if (vma)
 178			vma->vm_flags |= VM_USERMAP;
 179		mmap_write_unlock(current->mm);
 180	}
 181
 182	return ret;
 183}
 184
 185void *vmalloc_user(unsigned long size)
 186{
 187	return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
 188}
 189EXPORT_SYMBOL(vmalloc_user);
 190
 191struct page *vmalloc_to_page(const void *addr)
 192{
 193	return virt_to_page(addr);
 194}
 195EXPORT_SYMBOL(vmalloc_to_page);
 196
 197unsigned long vmalloc_to_pfn(const void *addr)
 198{
 199	return page_to_pfn(virt_to_page(addr));
 200}
 201EXPORT_SYMBOL(vmalloc_to_pfn);
 202
 203long vread(char *buf, char *addr, unsigned long count)
 204{
 205	/* Don't allow overflow */
 206	if ((unsigned long) buf + count < count)
 207		count = -(unsigned long) buf;
 208
 209	memcpy(buf, addr, count);
 210	return count;
 211}
 212
 
 
 
 
 
 
 
 
 
 
 213/*
 214 *	vmalloc  -  allocate virtually contiguous memory
 215 *
 216 *	@size:		allocation size
 217 *
 218 *	Allocate enough pages to cover @size from the page level
 219 *	allocator and map them into contiguous kernel virtual space.
 220 *
 221 *	For tight control over page level allocator and protection flags
 222 *	use __vmalloc() instead.
 223 */
 224void *vmalloc(unsigned long size)
 225{
 226	return __vmalloc(size, GFP_KERNEL);
 227}
 228EXPORT_SYMBOL(vmalloc);
 229
 230/*
 231 *	vzalloc - allocate virtually contiguous memory with zero fill
 232 *
 233 *	@size:		allocation size
 234 *
 235 *	Allocate enough pages to cover @size from the page level
 236 *	allocator and map them into contiguous kernel virtual space.
 237 *	The memory allocated is set to zero.
 238 *
 239 *	For tight control over page level allocator and protection flags
 240 *	use __vmalloc() instead.
 241 */
 242void *vzalloc(unsigned long size)
 243{
 244	return __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
 
 245}
 246EXPORT_SYMBOL(vzalloc);
 247
 248/**
 249 * vmalloc_node - allocate memory on a specific node
 250 * @size:	allocation size
 251 * @node:	numa node
 252 *
 253 * Allocate enough pages to cover @size from the page level
 254 * allocator and map them into contiguous kernel virtual space.
 255 *
 256 * For tight control over page level allocator and protection flags
 257 * use __vmalloc() instead.
 258 */
 259void *vmalloc_node(unsigned long size, int node)
 260{
 261	return vmalloc(size);
 262}
 263EXPORT_SYMBOL(vmalloc_node);
 264
 265/**
 266 * vzalloc_node - allocate memory on a specific node with zero fill
 267 * @size:	allocation size
 268 * @node:	numa node
 269 *
 270 * Allocate enough pages to cover @size from the page level
 271 * allocator and map them into contiguous kernel virtual space.
 272 * The memory allocated is set to zero.
 273 *
 274 * For tight control over page level allocator and protection flags
 275 * use __vmalloc() instead.
 276 */
 277void *vzalloc_node(unsigned long size, int node)
 278{
 279	return vzalloc(size);
 280}
 281EXPORT_SYMBOL(vzalloc_node);
 282
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 283/**
 284 * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 285 *	@size:		allocation size
 286 *
 287 *	Allocate enough 32bit PA addressable pages to cover @size from the
 288 *	page level allocator and map them into contiguous kernel virtual space.
 289 */
 290void *vmalloc_32(unsigned long size)
 291{
 292	return __vmalloc(size, GFP_KERNEL);
 293}
 294EXPORT_SYMBOL(vmalloc_32);
 295
 296/**
 297 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 298 *	@size:		allocation size
 299 *
 300 * The resulting memory area is 32bit addressable and zeroed so it can be
 301 * mapped to userspace without leaking data.
 302 *
 303 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 304 * remap_vmalloc_range() are permissible.
 305 */
 306void *vmalloc_32_user(unsigned long size)
 307{
 308	/*
 309	 * We'll have to sort out the ZONE_DMA bits for 64-bit,
 310	 * but for now this can simply use vmalloc_user() directly.
 311	 */
 312	return vmalloc_user(size);
 313}
 314EXPORT_SYMBOL(vmalloc_32_user);
 315
 316void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 317{
 318	BUG();
 319	return NULL;
 320}
 321EXPORT_SYMBOL(vmap);
 322
 323void vunmap(const void *addr)
 324{
 325	BUG();
 326}
 327EXPORT_SYMBOL(vunmap);
 328
 329void *vm_map_ram(struct page **pages, unsigned int count, int node)
 330{
 331	BUG();
 332	return NULL;
 333}
 334EXPORT_SYMBOL(vm_map_ram);
 335
 336void vm_unmap_ram(const void *mem, unsigned int count)
 337{
 338	BUG();
 339}
 340EXPORT_SYMBOL(vm_unmap_ram);
 341
 342void vm_unmap_aliases(void)
 343{
 344}
 345EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 346
 347void free_vm_area(struct vm_struct *area)
 
 
 
 
 348{
 349	BUG();
 350}
 351EXPORT_SYMBOL_GPL(free_vm_area);
 352
 353int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 354		   struct page *page)
 
 
 
 
 
 
 
 
 
 
 
 355{
 356	return -EINVAL;
 
 357}
 358EXPORT_SYMBOL(vm_insert_page);
 359
 360int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
 361			unsigned long num)
 362{
 363	return -EINVAL;
 364}
 365EXPORT_SYMBOL(vm_map_pages);
 366
 367int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
 368				unsigned long num)
 369{
 370	return -EINVAL;
 371}
 372EXPORT_SYMBOL(vm_map_pages_zero);
 373
 374/*
 375 *  sys_brk() for the most part doesn't need the global kernel
 376 *  lock, except when an application is doing something nasty
 377 *  like trying to un-brk an area that has already been mapped
 378 *  to a regular file.  in this case, the unmapping will need
 379 *  to invoke file system routines that need the global lock.
 380 */
 381SYSCALL_DEFINE1(brk, unsigned long, brk)
 382{
 383	struct mm_struct *mm = current->mm;
 384
 385	if (brk < mm->start_brk || brk > mm->context.end_brk)
 386		return mm->brk;
 387
 388	if (mm->brk == brk)
 389		return mm->brk;
 390
 391	/*
 392	 * Always allow shrinking brk
 393	 */
 394	if (brk <= mm->brk) {
 395		mm->brk = brk;
 396		return brk;
 397	}
 398
 399	/*
 400	 * Ok, looks good - let it rip.
 401	 */
 402	flush_icache_user_range(mm->brk, brk);
 403	return mm->brk = brk;
 404}
 405
 406/*
 407 * initialise the percpu counter for VM and region record slabs
 408 */
 409void __init mmap_init(void)
 410{
 411	int ret;
 412
 413	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
 414	VM_BUG_ON(ret);
 415	vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
 416}
 417
 418/*
 419 * validate the region tree
 420 * - the caller must hold the region lock
 421 */
 422#ifdef CONFIG_DEBUG_NOMMU_REGIONS
 423static noinline void validate_nommu_regions(void)
 424{
 425	struct vm_region *region, *last;
 426	struct rb_node *p, *lastp;
 427
 428	lastp = rb_first(&nommu_region_tree);
 429	if (!lastp)
 430		return;
 431
 432	last = rb_entry(lastp, struct vm_region, vm_rb);
 433	BUG_ON(last->vm_end <= last->vm_start);
 434	BUG_ON(last->vm_top < last->vm_end);
 435
 436	while ((p = rb_next(lastp))) {
 437		region = rb_entry(p, struct vm_region, vm_rb);
 438		last = rb_entry(lastp, struct vm_region, vm_rb);
 439
 440		BUG_ON(region->vm_end <= region->vm_start);
 441		BUG_ON(region->vm_top < region->vm_end);
 442		BUG_ON(region->vm_start < last->vm_top);
 443
 444		lastp = p;
 445	}
 446}
 447#else
 448static void validate_nommu_regions(void)
 449{
 450}
 451#endif
 452
 453/*
 454 * add a region into the global tree
 455 */
 456static void add_nommu_region(struct vm_region *region)
 457{
 458	struct vm_region *pregion;
 459	struct rb_node **p, *parent;
 460
 461	validate_nommu_regions();
 462
 463	parent = NULL;
 464	p = &nommu_region_tree.rb_node;
 465	while (*p) {
 466		parent = *p;
 467		pregion = rb_entry(parent, struct vm_region, vm_rb);
 468		if (region->vm_start < pregion->vm_start)
 469			p = &(*p)->rb_left;
 470		else if (region->vm_start > pregion->vm_start)
 471			p = &(*p)->rb_right;
 472		else if (pregion == region)
 473			return;
 474		else
 475			BUG();
 476	}
 477
 478	rb_link_node(&region->vm_rb, parent, p);
 479	rb_insert_color(&region->vm_rb, &nommu_region_tree);
 480
 481	validate_nommu_regions();
 482}
 483
 484/*
 485 * delete a region from the global tree
 486 */
 487static void delete_nommu_region(struct vm_region *region)
 488{
 489	BUG_ON(!nommu_region_tree.rb_node);
 490
 491	validate_nommu_regions();
 492	rb_erase(&region->vm_rb, &nommu_region_tree);
 493	validate_nommu_regions();
 494}
 495
 496/*
 497 * free a contiguous series of pages
 498 */
 499static void free_page_series(unsigned long from, unsigned long to)
 500{
 501	for (; from < to; from += PAGE_SIZE) {
 502		struct page *page = virt_to_page(from);
 503
 
 504		atomic_long_dec(&mmap_pages_allocated);
 
 
 
 505		put_page(page);
 506	}
 507}
 508
 509/*
 510 * release a reference to a region
 511 * - the caller must hold the region semaphore for writing, which this releases
 512 * - the region may not have been added to the tree yet, in which case vm_top
 513 *   will equal vm_start
 514 */
 515static void __put_nommu_region(struct vm_region *region)
 516	__releases(nommu_region_sem)
 517{
 
 
 518	BUG_ON(!nommu_region_tree.rb_node);
 519
 520	if (--region->vm_usage == 0) {
 521		if (region->vm_top > region->vm_start)
 522			delete_nommu_region(region);
 523		up_write(&nommu_region_sem);
 524
 525		if (region->vm_file)
 526			fput(region->vm_file);
 527
 528		/* IO memory and memory shared directly out of the pagecache
 529		 * from ramfs/tmpfs mustn't be released here */
 530		if (region->vm_flags & VM_MAPPED_COPY)
 
 531			free_page_series(region->vm_start, region->vm_top);
 
 532		kmem_cache_free(vm_region_jar, region);
 533	} else {
 534		up_write(&nommu_region_sem);
 535	}
 536}
 537
 538/*
 539 * release a reference to a region
 540 */
 541static void put_nommu_region(struct vm_region *region)
 542{
 543	down_write(&nommu_region_sem);
 544	__put_nommu_region(region);
 545}
 546
 547/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 548 * add a VMA into a process's mm_struct in the appropriate place in the list
 549 * and tree and add to the address space's page tree also if not an anonymous
 550 * page
 551 * - should be called with mm->mmap_lock held writelocked
 552 */
 553static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
 554{
 555	struct vm_area_struct *pvma, *prev;
 556	struct address_space *mapping;
 557	struct rb_node **p, *parent, *rb_prev;
 558
 
 
 559	BUG_ON(!vma->vm_region);
 560
 561	mm->map_count++;
 562	vma->vm_mm = mm;
 563
 
 
 564	/* add the VMA to the mapping */
 565	if (vma->vm_file) {
 566		mapping = vma->vm_file->f_mapping;
 567
 568		i_mmap_lock_write(mapping);
 569		flush_dcache_mmap_lock(mapping);
 570		vma_interval_tree_insert(vma, &mapping->i_mmap);
 571		flush_dcache_mmap_unlock(mapping);
 572		i_mmap_unlock_write(mapping);
 573	}
 574
 575	/* add the VMA to the tree */
 576	parent = rb_prev = NULL;
 577	p = &mm->mm_rb.rb_node;
 578	while (*p) {
 579		parent = *p;
 580		pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
 581
 582		/* sort by: start addr, end addr, VMA struct addr in that order
 583		 * (the latter is necessary as we may get identical VMAs) */
 584		if (vma->vm_start < pvma->vm_start)
 585			p = &(*p)->rb_left;
 586		else if (vma->vm_start > pvma->vm_start) {
 587			rb_prev = parent;
 588			p = &(*p)->rb_right;
 589		} else if (vma->vm_end < pvma->vm_end)
 590			p = &(*p)->rb_left;
 591		else if (vma->vm_end > pvma->vm_end) {
 592			rb_prev = parent;
 593			p = &(*p)->rb_right;
 594		} else if (vma < pvma)
 595			p = &(*p)->rb_left;
 596		else if (vma > pvma) {
 597			rb_prev = parent;
 598			p = &(*p)->rb_right;
 599		} else
 600			BUG();
 601	}
 602
 603	rb_link_node(&vma->vm_rb, parent, p);
 604	rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 605
 606	/* add VMA to the VMA list also */
 607	prev = NULL;
 608	if (rb_prev)
 609		prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 610
 611	__vma_link_list(mm, vma, prev);
 612}
 613
 614/*
 615 * delete a VMA from its owning mm_struct and address space
 616 */
 617static void delete_vma_from_mm(struct vm_area_struct *vma)
 618{
 619	int i;
 620	struct address_space *mapping;
 621	struct mm_struct *mm = vma->vm_mm;
 622	struct task_struct *curr = current;
 
 
 
 623
 624	mm->map_count--;
 625	for (i = 0; i < VMACACHE_SIZE; i++) {
 626		/* if the vma is cached, invalidate the entire cache */
 627		if (curr->vmacache.vmas[i] == vma) {
 628			vmacache_invalidate(mm);
 629			break;
 630		}
 631	}
 632
 633	/* remove the VMA from the mapping */
 634	if (vma->vm_file) {
 635		mapping = vma->vm_file->f_mapping;
 636
 637		i_mmap_lock_write(mapping);
 638		flush_dcache_mmap_lock(mapping);
 639		vma_interval_tree_remove(vma, &mapping->i_mmap);
 640		flush_dcache_mmap_unlock(mapping);
 641		i_mmap_unlock_write(mapping);
 642	}
 643
 644	/* remove from the MM's tree and list */
 645	rb_erase(&vma->vm_rb, &mm->mm_rb);
 646
 647	__vma_unlink_list(mm, vma);
 
 
 
 
 
 
 648}
 649
 650/*
 651 * destroy a VMA record
 652 */
 653static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 654{
 
 655	if (vma->vm_ops && vma->vm_ops->close)
 656		vma->vm_ops->close(vma);
 657	if (vma->vm_file)
 658		fput(vma->vm_file);
 
 
 
 659	put_nommu_region(vma->vm_region);
 660	vm_area_free(vma);
 661}
 662
 663/*
 664 * look up the first VMA in which addr resides, NULL if none
 665 * - should be called with mm->mmap_lock at least held readlocked
 666 */
 667struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 668{
 669	struct vm_area_struct *vma;
 670
 671	/* check the cache first */
 672	vma = vmacache_find(mm, addr);
 673	if (likely(vma))
 674		return vma;
 675
 676	/* trawl the list (there may be multiple mappings in which addr
 677	 * resides) */
 678	for (vma = mm->mmap; vma; vma = vma->vm_next) {
 679		if (vma->vm_start > addr)
 680			return NULL;
 681		if (vma->vm_end > addr) {
 682			vmacache_update(addr, vma);
 683			return vma;
 684		}
 685	}
 686
 687	return NULL;
 688}
 689EXPORT_SYMBOL(find_vma);
 690
 691/*
 692 * find a VMA
 693 * - we don't extend stack VMAs under NOMMU conditions
 694 */
 695struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 696{
 697	return find_vma(mm, addr);
 698}
 699
 700/*
 701 * expand a stack to a given address
 702 * - not supported under NOMMU conditions
 703 */
 704int expand_stack(struct vm_area_struct *vma, unsigned long address)
 705{
 706	return -ENOMEM;
 707}
 708
 709/*
 710 * look up the first VMA exactly that exactly matches addr
 711 * - should be called with mm->mmap_lock at least held readlocked
 712 */
 713static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 714					     unsigned long addr,
 715					     unsigned long len)
 716{
 717	struct vm_area_struct *vma;
 718	unsigned long end = addr + len;
 719
 720	/* check the cache first */
 721	vma = vmacache_find_exact(mm, addr, end);
 722	if (vma)
 723		return vma;
 724
 725	/* trawl the list (there may be multiple mappings in which addr
 726	 * resides) */
 727	for (vma = mm->mmap; vma; vma = vma->vm_next) {
 728		if (vma->vm_start < addr)
 729			continue;
 730		if (vma->vm_start > addr)
 731			return NULL;
 732		if (vma->vm_end == end) {
 733			vmacache_update(addr, vma);
 734			return vma;
 735		}
 736	}
 737
 738	return NULL;
 739}
 740
 741/*
 742 * determine whether a mapping should be permitted and, if so, what sort of
 743 * mapping we're capable of supporting
 744 */
 745static int validate_mmap_request(struct file *file,
 746				 unsigned long addr,
 747				 unsigned long len,
 748				 unsigned long prot,
 749				 unsigned long flags,
 750				 unsigned long pgoff,
 751				 unsigned long *_capabilities)
 752{
 753	unsigned long capabilities, rlen;
 754	int ret;
 755
 756	/* do the simple checks first */
 757	if (flags & MAP_FIXED)
 
 
 
 758		return -EINVAL;
 
 759
 760	if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 761	    (flags & MAP_TYPE) != MAP_SHARED)
 762		return -EINVAL;
 763
 764	if (!len)
 765		return -EINVAL;
 766
 767	/* Careful about overflows.. */
 768	rlen = PAGE_ALIGN(len);
 769	if (!rlen || rlen > TASK_SIZE)
 770		return -ENOMEM;
 771
 772	/* offset overflow? */
 773	if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 774		return -EOVERFLOW;
 775
 776	if (file) {
 
 
 
 777		/* files must support mmap */
 778		if (!file->f_op->mmap)
 779			return -ENODEV;
 780
 781		/* work out if what we've got could possibly be shared
 782		 * - we support chardevs that provide their own "memory"
 783		 * - we support files/blockdevs that are memory backed
 784		 */
 785		if (file->f_op->mmap_capabilities) {
 786			capabilities = file->f_op->mmap_capabilities(file);
 787		} else {
 
 
 
 
 
 
 788			/* no explicit capabilities set, so assume some
 789			 * defaults */
 790			switch (file_inode(file)->i_mode & S_IFMT) {
 791			case S_IFREG:
 792			case S_IFBLK:
 793				capabilities = NOMMU_MAP_COPY;
 794				break;
 795
 796			case S_IFCHR:
 797				capabilities =
 798					NOMMU_MAP_DIRECT |
 799					NOMMU_MAP_READ |
 800					NOMMU_MAP_WRITE;
 801				break;
 802
 803			default:
 804				return -EINVAL;
 805			}
 806		}
 807
 808		/* eliminate any capabilities that we can't support on this
 809		 * device */
 810		if (!file->f_op->get_unmapped_area)
 811			capabilities &= ~NOMMU_MAP_DIRECT;
 812		if (!(file->f_mode & FMODE_CAN_READ))
 813			capabilities &= ~NOMMU_MAP_COPY;
 814
 815		/* The file shall have been opened with read permission. */
 816		if (!(file->f_mode & FMODE_READ))
 817			return -EACCES;
 818
 819		if (flags & MAP_SHARED) {
 820			/* do checks for writing, appending and locking */
 821			if ((prot & PROT_WRITE) &&
 822			    !(file->f_mode & FMODE_WRITE))
 823				return -EACCES;
 824
 825			if (IS_APPEND(file_inode(file)) &&
 826			    (file->f_mode & FMODE_WRITE))
 827				return -EACCES;
 828
 829			if (locks_verify_locked(file))
 830				return -EAGAIN;
 831
 832			if (!(capabilities & NOMMU_MAP_DIRECT))
 833				return -ENODEV;
 834
 835			/* we mustn't privatise shared mappings */
 836			capabilities &= ~NOMMU_MAP_COPY;
 837		} else {
 
 838			/* we're going to read the file into private memory we
 839			 * allocate */
 840			if (!(capabilities & NOMMU_MAP_COPY))
 841				return -ENODEV;
 842
 843			/* we don't permit a private writable mapping to be
 844			 * shared with the backing device */
 845			if (prot & PROT_WRITE)
 846				capabilities &= ~NOMMU_MAP_DIRECT;
 847		}
 848
 849		if (capabilities & NOMMU_MAP_DIRECT) {
 850			if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
 851			    ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
 852			    ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
 853			    ) {
 854				capabilities &= ~NOMMU_MAP_DIRECT;
 855				if (flags & MAP_SHARED) {
 856					pr_warn("MAP_SHARED not completely supported on !MMU\n");
 
 857					return -EINVAL;
 858				}
 859			}
 860		}
 861
 862		/* handle executable mappings and implied executable
 863		 * mappings */
 864		if (path_noexec(&file->f_path)) {
 865			if (prot & PROT_EXEC)
 866				return -EPERM;
 867		} else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 
 868			/* handle implication of PROT_EXEC by PROT_READ */
 869			if (current->personality & READ_IMPLIES_EXEC) {
 870				if (capabilities & NOMMU_MAP_EXEC)
 871					prot |= PROT_EXEC;
 872			}
 873		} else if ((prot & PROT_READ) &&
 
 874			 (prot & PROT_EXEC) &&
 875			 !(capabilities & NOMMU_MAP_EXEC)
 876			 ) {
 877			/* backing file is not executable, try to copy */
 878			capabilities &= ~NOMMU_MAP_DIRECT;
 879		}
 880	} else {
 
 881		/* anonymous mappings are always memory backed and can be
 882		 * privately mapped
 883		 */
 884		capabilities = NOMMU_MAP_COPY;
 885
 886		/* handle PROT_EXEC implication by PROT_READ */
 887		if ((prot & PROT_READ) &&
 888		    (current->personality & READ_IMPLIES_EXEC))
 889			prot |= PROT_EXEC;
 890	}
 891
 892	/* allow the security API to have its say */
 893	ret = security_mmap_addr(addr);
 894	if (ret < 0)
 895		return ret;
 896
 897	/* looks okay */
 898	*_capabilities = capabilities;
 899	return 0;
 900}
 901
 902/*
 903 * we've determined that we can make the mapping, now translate what we
 904 * now know into VMA flags
 905 */
 906static unsigned long determine_vm_flags(struct file *file,
 907					unsigned long prot,
 908					unsigned long flags,
 909					unsigned long capabilities)
 910{
 911	unsigned long vm_flags;
 912
 913	vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
 914	/* vm_flags |= mm->def_flags; */
 915
 916	if (!(capabilities & NOMMU_MAP_DIRECT)) {
 917		/* attempt to share read-only copies of mapped file chunks */
 918		vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 919		if (file && !(prot & PROT_WRITE))
 920			vm_flags |= VM_MAYSHARE;
 921	} else {
 922		/* overlay a shareable mapping on the backing device or inode
 923		 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
 924		 * romfs/cramfs */
 925		vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
 926		if (flags & MAP_SHARED)
 927			vm_flags |= VM_SHARED;
 928	}
 929
 930	/* refuse to let anyone share private mappings with this process if
 931	 * it's being traced - otherwise breakpoints set in it may interfere
 932	 * with another untraced process
 933	 */
 934	if ((flags & MAP_PRIVATE) && current->ptrace)
 935		vm_flags &= ~VM_MAYSHARE;
 936
 937	return vm_flags;
 938}
 939
 940/*
 941 * set up a shared mapping on a file (the driver or filesystem provides and
 942 * pins the storage)
 943 */
 944static int do_mmap_shared_file(struct vm_area_struct *vma)
 945{
 946	int ret;
 947
 948	ret = call_mmap(vma->vm_file, vma);
 949	if (ret == 0) {
 950		vma->vm_region->vm_top = vma->vm_region->vm_end;
 951		return 0;
 952	}
 953	if (ret != -ENOSYS)
 954		return ret;
 955
 956	/* getting -ENOSYS indicates that direct mmap isn't possible (as
 957	 * opposed to tried but failed) so we can only give a suitable error as
 958	 * it's not possible to make a private copy if MAP_SHARED was given */
 959	return -ENODEV;
 960}
 961
 962/*
 963 * set up a private mapping or an anonymous shared mapping
 964 */
 965static int do_mmap_private(struct vm_area_struct *vma,
 966			   struct vm_region *region,
 967			   unsigned long len,
 968			   unsigned long capabilities)
 969{
 970	unsigned long total, point;
 
 971	void *base;
 972	int ret, order;
 973
 974	/* invoke the file's mapping function so that it can keep track of
 975	 * shared mappings on devices or memory
 976	 * - VM_MAYSHARE will be set if it may attempt to share
 977	 */
 978	if (capabilities & NOMMU_MAP_DIRECT) {
 979		ret = call_mmap(vma->vm_file, vma);
 980		if (ret == 0) {
 981			/* shouldn't return success if we're not sharing */
 982			BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
 983			vma->vm_region->vm_top = vma->vm_region->vm_end;
 984			return 0;
 985		}
 986		if (ret != -ENOSYS)
 987			return ret;
 988
 989		/* getting an ENOSYS error indicates that direct mmap isn't
 990		 * possible (as opposed to tried but failed) so we'll try to
 991		 * make a private copy of the data and map that instead */
 992	}
 993
 994
 995	/* allocate some memory to hold the mapping
 996	 * - note that this may not return a page-aligned address if the object
 997	 *   we're allocating is smaller than a page
 998	 */
 999	order = get_order(len);
1000	total = 1 << order;
1001	point = len >> PAGE_SHIFT;
1002
1003	/* we don't want to allocate a power-of-2 sized page set */
1004	if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1005		total = point;
1006
1007	base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1008	if (!base)
1009		goto enomem;
1010
 
1011	atomic_long_add(total, &mmap_pages_allocated);
1012
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1013	region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1014	region->vm_start = (unsigned long) base;
1015	region->vm_end   = region->vm_start + len;
1016	region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
1017
1018	vma->vm_start = region->vm_start;
1019	vma->vm_end   = region->vm_start + len;
1020
1021	if (vma->vm_file) {
1022		/* read the contents of a file into the copy */
 
1023		loff_t fpos;
1024
1025		fpos = vma->vm_pgoff;
1026		fpos <<= PAGE_SHIFT;
1027
1028		ret = kernel_read(vma->vm_file, base, len, &fpos);
 
 
 
 
1029		if (ret < 0)
1030			goto error_free;
1031
1032		/* clear the last little bit */
1033		if (ret < len)
1034			memset(base + ret, 0, len - ret);
1035
1036	} else {
1037		vma_set_anonymous(vma);
1038	}
1039
1040	return 0;
1041
1042error_free:
1043	free_page_series(region->vm_start, region->vm_top);
1044	region->vm_start = vma->vm_start = 0;
1045	region->vm_end   = vma->vm_end = 0;
1046	region->vm_top   = 0;
1047	return ret;
1048
1049enomem:
1050	pr_err("Allocation of length %lu from process %d (%s) failed\n",
1051	       len, current->pid, current->comm);
1052	show_free_areas(0, NULL);
1053	return -ENOMEM;
1054}
1055
1056/*
1057 * handle mapping creation for uClinux
1058 */
1059unsigned long do_mmap(struct file *file,
1060			unsigned long addr,
1061			unsigned long len,
1062			unsigned long prot,
1063			unsigned long flags,
1064			unsigned long pgoff,
1065			unsigned long *populate,
1066			struct list_head *uf)
1067{
1068	struct vm_area_struct *vma;
1069	struct vm_region *region;
1070	struct rb_node *rb;
1071	vm_flags_t vm_flags;
1072	unsigned long capabilities, result;
1073	int ret;
1074
1075	*populate = 0;
1076
1077	/* decide whether we should attempt the mapping, and if so what sort of
1078	 * mapping */
1079	ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1080				    &capabilities);
1081	if (ret < 0)
 
1082		return ret;
 
1083
1084	/* we ignore the address hint */
1085	addr = 0;
1086	len = PAGE_ALIGN(len);
1087
1088	/* we've determined that we can make the mapping, now translate what we
1089	 * now know into VMA flags */
1090	vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1091
1092	/* we're going to need to record the mapping */
1093	region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1094	if (!region)
1095		goto error_getting_region;
1096
1097	vma = vm_area_alloc(current->mm);
1098	if (!vma)
1099		goto error_getting_vma;
1100
1101	region->vm_usage = 1;
1102	region->vm_flags = vm_flags;
1103	region->vm_pgoff = pgoff;
1104
 
1105	vma->vm_flags = vm_flags;
1106	vma->vm_pgoff = pgoff;
1107
1108	if (file) {
1109		region->vm_file = get_file(file);
1110		vma->vm_file = get_file(file);
 
 
 
 
 
 
1111	}
1112
1113	down_write(&nommu_region_sem);
1114
1115	/* if we want to share, we need to check for regions created by other
1116	 * mmap() calls that overlap with our proposed mapping
1117	 * - we can only share with a superset match on most regular files
1118	 * - shared mappings on character devices and memory backed files are
1119	 *   permitted to overlap inexactly as far as we are concerned for in
1120	 *   these cases, sharing is handled in the driver or filesystem rather
1121	 *   than here
1122	 */
1123	if (vm_flags & VM_MAYSHARE) {
1124		struct vm_region *pregion;
1125		unsigned long pglen, rpglen, pgend, rpgend, start;
1126
1127		pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1128		pgend = pgoff + pglen;
1129
1130		for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1131			pregion = rb_entry(rb, struct vm_region, vm_rb);
1132
1133			if (!(pregion->vm_flags & VM_MAYSHARE))
1134				continue;
1135
1136			/* search for overlapping mappings on the same file */
1137			if (file_inode(pregion->vm_file) !=
1138			    file_inode(file))
1139				continue;
1140
1141			if (pregion->vm_pgoff >= pgend)
1142				continue;
1143
1144			rpglen = pregion->vm_end - pregion->vm_start;
1145			rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1146			rpgend = pregion->vm_pgoff + rpglen;
1147			if (pgoff >= rpgend)
1148				continue;
1149
1150			/* handle inexactly overlapping matches between
1151			 * mappings */
1152			if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1153			    !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1154				/* new mapping is not a subset of the region */
1155				if (!(capabilities & NOMMU_MAP_DIRECT))
1156					goto sharing_violation;
1157				continue;
1158			}
1159
1160			/* we've found a region we can share */
1161			pregion->vm_usage++;
1162			vma->vm_region = pregion;
1163			start = pregion->vm_start;
1164			start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1165			vma->vm_start = start;
1166			vma->vm_end = start + len;
1167
1168			if (pregion->vm_flags & VM_MAPPED_COPY)
 
1169				vma->vm_flags |= VM_MAPPED_COPY;
1170			else {
 
1171				ret = do_mmap_shared_file(vma);
1172				if (ret < 0) {
1173					vma->vm_region = NULL;
1174					vma->vm_start = 0;
1175					vma->vm_end = 0;
1176					pregion->vm_usage--;
1177					pregion = NULL;
1178					goto error_just_free;
1179				}
1180			}
1181			fput(region->vm_file);
1182			kmem_cache_free(vm_region_jar, region);
1183			region = pregion;
1184			result = start;
1185			goto share;
1186		}
1187
1188		/* obtain the address at which to make a shared mapping
1189		 * - this is the hook for quasi-memory character devices to
1190		 *   tell us the location of a shared mapping
1191		 */
1192		if (capabilities & NOMMU_MAP_DIRECT) {
1193			addr = file->f_op->get_unmapped_area(file, addr, len,
1194							     pgoff, flags);
1195			if (IS_ERR_VALUE(addr)) {
1196				ret = addr;
1197				if (ret != -ENOSYS)
1198					goto error_just_free;
1199
1200				/* the driver refused to tell us where to site
1201				 * the mapping so we'll have to attempt to copy
1202				 * it */
1203				ret = -ENODEV;
1204				if (!(capabilities & NOMMU_MAP_COPY))
1205					goto error_just_free;
1206
1207				capabilities &= ~NOMMU_MAP_DIRECT;
1208			} else {
1209				vma->vm_start = region->vm_start = addr;
1210				vma->vm_end = region->vm_end = addr + len;
1211			}
1212		}
1213	}
1214
1215	vma->vm_region = region;
1216
1217	/* set up the mapping
1218	 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1219	 */
1220	if (file && vma->vm_flags & VM_SHARED)
1221		ret = do_mmap_shared_file(vma);
1222	else
1223		ret = do_mmap_private(vma, region, len, capabilities);
1224	if (ret < 0)
1225		goto error_just_free;
1226	add_nommu_region(region);
1227
1228	/* clear anonymous mappings that don't ask for uninitialized data */
1229	if (!vma->vm_file &&
1230	    (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1231	     !(flags & MAP_UNINITIALIZED)))
1232		memset((void *)region->vm_start, 0,
1233		       region->vm_end - region->vm_start);
1234
1235	/* okay... we have a mapping; now we have to register it */
1236	result = vma->vm_start;
1237
1238	current->mm->total_vm += len >> PAGE_SHIFT;
1239
1240share:
1241	add_vma_to_mm(current->mm, vma);
1242
1243	/* we flush the region from the icache only when the first executable
1244	 * mapping of it is made  */
1245	if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1246		flush_icache_user_range(region->vm_start, region->vm_end);
1247		region->vm_icache_flushed = true;
1248	}
1249
1250	up_write(&nommu_region_sem);
1251
 
1252	return result;
1253
1254error_just_free:
1255	up_write(&nommu_region_sem);
1256error:
1257	if (region->vm_file)
1258		fput(region->vm_file);
1259	kmem_cache_free(vm_region_jar, region);
1260	if (vma->vm_file)
1261		fput(vma->vm_file);
1262	vm_area_free(vma);
 
 
 
1263	return ret;
1264
1265sharing_violation:
1266	up_write(&nommu_region_sem);
1267	pr_warn("Attempt to share mismatched mappings\n");
1268	ret = -EINVAL;
1269	goto error;
1270
1271error_getting_vma:
1272	kmem_cache_free(vm_region_jar, region);
1273	pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1274			len, current->pid);
1275	show_free_areas(0, NULL);
 
1276	return -ENOMEM;
1277
1278error_getting_region:
1279	pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1280			len, current->pid);
1281	show_free_areas(0, NULL);
 
1282	return -ENOMEM;
1283}
1284
1285unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1286			      unsigned long prot, unsigned long flags,
1287			      unsigned long fd, unsigned long pgoff)
1288{
1289	struct file *file = NULL;
1290	unsigned long retval = -EBADF;
1291
1292	audit_mmap_fd(fd, flags);
1293	if (!(flags & MAP_ANONYMOUS)) {
1294		file = fget(fd);
1295		if (!file)
1296			goto out;
1297	}
1298
1299	flags &= ~MAP_DENYWRITE;
1300
1301	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1302
1303	if (file)
1304		fput(file);
1305out:
1306	return retval;
1307}
1308
1309SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1310		unsigned long, prot, unsigned long, flags,
1311		unsigned long, fd, unsigned long, pgoff)
1312{
1313	return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1314}
1315
1316#ifdef __ARCH_WANT_SYS_OLD_MMAP
1317struct mmap_arg_struct {
1318	unsigned long addr;
1319	unsigned long len;
1320	unsigned long prot;
1321	unsigned long flags;
1322	unsigned long fd;
1323	unsigned long offset;
1324};
1325
1326SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1327{
1328	struct mmap_arg_struct a;
1329
1330	if (copy_from_user(&a, arg, sizeof(a)))
1331		return -EFAULT;
1332	if (offset_in_page(a.offset))
1333		return -EINVAL;
1334
1335	return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1336			       a.offset >> PAGE_SHIFT);
1337}
1338#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1339
1340/*
1341 * split a vma into two pieces at address 'addr', a new vma is allocated either
1342 * for the first part or the tail.
1343 */
1344int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1345	      unsigned long addr, int new_below)
1346{
1347	struct vm_area_struct *new;
1348	struct vm_region *region;
1349	unsigned long npages;
1350
 
 
1351	/* we're only permitted to split anonymous regions (these should have
1352	 * only a single usage on the region) */
1353	if (vma->vm_file)
1354		return -ENOMEM;
1355
1356	if (mm->map_count >= sysctl_max_map_count)
1357		return -ENOMEM;
1358
1359	region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1360	if (!region)
1361		return -ENOMEM;
1362
1363	new = vm_area_dup(vma);
1364	if (!new) {
1365		kmem_cache_free(vm_region_jar, region);
1366		return -ENOMEM;
1367	}
1368
1369	/* most fields are the same, copy all, and then fixup */
 
1370	*region = *vma->vm_region;
1371	new->vm_region = region;
1372
1373	npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1374
1375	if (new_below) {
1376		region->vm_top = region->vm_end = new->vm_end = addr;
1377	} else {
1378		region->vm_start = new->vm_start = addr;
1379		region->vm_pgoff = new->vm_pgoff += npages;
1380	}
1381
1382	if (new->vm_ops && new->vm_ops->open)
1383		new->vm_ops->open(new);
1384
1385	delete_vma_from_mm(vma);
1386	down_write(&nommu_region_sem);
1387	delete_nommu_region(vma->vm_region);
1388	if (new_below) {
1389		vma->vm_region->vm_start = vma->vm_start = addr;
1390		vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1391	} else {
1392		vma->vm_region->vm_end = vma->vm_end = addr;
1393		vma->vm_region->vm_top = addr;
1394	}
1395	add_nommu_region(vma->vm_region);
1396	add_nommu_region(new->vm_region);
1397	up_write(&nommu_region_sem);
1398	add_vma_to_mm(mm, vma);
1399	add_vma_to_mm(mm, new);
1400	return 0;
1401}
1402
1403/*
1404 * shrink a VMA by removing the specified chunk from either the beginning or
1405 * the end
1406 */
1407static int shrink_vma(struct mm_struct *mm,
1408		      struct vm_area_struct *vma,
1409		      unsigned long from, unsigned long to)
1410{
1411	struct vm_region *region;
1412
 
 
1413	/* adjust the VMA's pointers, which may reposition it in the MM's tree
1414	 * and list */
1415	delete_vma_from_mm(vma);
1416	if (from > vma->vm_start)
1417		vma->vm_end = from;
1418	else
1419		vma->vm_start = to;
1420	add_vma_to_mm(mm, vma);
1421
1422	/* cut the backing region down to size */
1423	region = vma->vm_region;
1424	BUG_ON(region->vm_usage != 1);
1425
1426	down_write(&nommu_region_sem);
1427	delete_nommu_region(region);
1428	if (from > region->vm_start) {
1429		to = region->vm_top;
1430		region->vm_top = region->vm_end = from;
1431	} else {
1432		region->vm_start = to;
1433	}
1434	add_nommu_region(region);
1435	up_write(&nommu_region_sem);
1436
1437	free_page_series(from, to);
1438	return 0;
1439}
1440
1441/*
1442 * release a mapping
1443 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1444 *   VMA, though it need not cover the whole VMA
1445 */
1446int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1447{
1448	struct vm_area_struct *vma;
1449	unsigned long end;
1450	int ret;
1451
 
 
1452	len = PAGE_ALIGN(len);
1453	if (len == 0)
1454		return -EINVAL;
1455
1456	end = start + len;
1457
1458	/* find the first potentially overlapping VMA */
1459	vma = find_vma(mm, start);
1460	if (!vma) {
1461		static int limit;
1462		if (limit < 5) {
1463			pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1464					current->pid, current->comm,
1465					start, start + len - 1);
 
 
1466			limit++;
1467		}
1468		return -EINVAL;
1469	}
1470
1471	/* we're allowed to split an anonymous VMA but not a file-backed one */
1472	if (vma->vm_file) {
1473		do {
1474			if (start > vma->vm_start)
 
1475				return -EINVAL;
 
1476			if (end == vma->vm_end)
1477				goto erase_whole_vma;
1478			vma = vma->vm_next;
1479		} while (vma);
 
1480		return -EINVAL;
1481	} else {
1482		/* the chunk must be a subset of the VMA found */
1483		if (start == vma->vm_start && end == vma->vm_end)
1484			goto erase_whole_vma;
1485		if (start < vma->vm_start || end > vma->vm_end)
 
1486			return -EINVAL;
1487		if (offset_in_page(start))
 
 
1488			return -EINVAL;
1489		if (end != vma->vm_end && offset_in_page(end))
 
 
1490			return -EINVAL;
 
1491		if (start != vma->vm_start && end != vma->vm_end) {
1492			ret = split_vma(mm, vma, start, 1);
1493			if (ret < 0)
 
1494				return ret;
 
1495		}
1496		return shrink_vma(mm, vma, start, end);
1497	}
1498
1499erase_whole_vma:
1500	delete_vma_from_mm(vma);
1501	delete_vma(mm, vma);
 
1502	return 0;
1503}
 
1504
1505int vm_munmap(unsigned long addr, size_t len)
1506{
1507	struct mm_struct *mm = current->mm;
1508	int ret;
1509
1510	mmap_write_lock(mm);
1511	ret = do_munmap(mm, addr, len, NULL);
1512	mmap_write_unlock(mm);
1513	return ret;
1514}
1515EXPORT_SYMBOL(vm_munmap);
1516
1517SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1518{
1519	return vm_munmap(addr, len);
1520}
1521
1522/*
1523 * release all the mappings made in a process's VM space
1524 */
1525void exit_mmap(struct mm_struct *mm)
1526{
1527	struct vm_area_struct *vma;
1528
1529	if (!mm)
1530		return;
1531
 
 
1532	mm->total_vm = 0;
1533
1534	while ((vma = mm->mmap)) {
1535		mm->mmap = vma->vm_next;
1536		delete_vma_from_mm(vma);
1537		delete_vma(mm, vma);
1538		cond_resched();
1539	}
 
 
1540}
1541
1542int vm_brk(unsigned long addr, unsigned long len)
1543{
1544	return -ENOMEM;
1545}
1546
1547/*
1548 * expand (or shrink) an existing mapping, potentially moving it at the same
1549 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1550 *
1551 * under NOMMU conditions, we only permit changing a mapping's size, and only
1552 * as long as it stays within the region allocated by do_mmap_private() and the
1553 * block is not shareable
1554 *
1555 * MREMAP_FIXED is not supported under NOMMU conditions
1556 */
1557static unsigned long do_mremap(unsigned long addr,
1558			unsigned long old_len, unsigned long new_len,
1559			unsigned long flags, unsigned long new_addr)
1560{
1561	struct vm_area_struct *vma;
1562
1563	/* insanity checks first */
1564	old_len = PAGE_ALIGN(old_len);
1565	new_len = PAGE_ALIGN(new_len);
1566	if (old_len == 0 || new_len == 0)
1567		return (unsigned long) -EINVAL;
1568
1569	if (offset_in_page(addr))
1570		return -EINVAL;
1571
1572	if (flags & MREMAP_FIXED && new_addr != addr)
1573		return (unsigned long) -EINVAL;
1574
1575	vma = find_vma_exact(current->mm, addr, old_len);
1576	if (!vma)
1577		return (unsigned long) -EINVAL;
1578
1579	if (vma->vm_end != vma->vm_start + old_len)
1580		return (unsigned long) -EFAULT;
1581
1582	if (vma->vm_flags & VM_MAYSHARE)
1583		return (unsigned long) -EPERM;
1584
1585	if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1586		return (unsigned long) -ENOMEM;
1587
1588	/* all checks complete - do it */
1589	vma->vm_end = vma->vm_start + new_len;
1590	return vma->vm_start;
1591}
 
1592
1593SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1594		unsigned long, new_len, unsigned long, flags,
1595		unsigned long, new_addr)
1596{
1597	unsigned long ret;
1598
1599	mmap_write_lock(current->mm);
1600	ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1601	mmap_write_unlock(current->mm);
1602	return ret;
1603}
1604
1605struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1606			 unsigned int foll_flags)
1607{
1608	return NULL;
1609}
1610
1611int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1612		unsigned long pfn, unsigned long size, pgprot_t prot)
1613{
1614	if (addr != (pfn << PAGE_SHIFT))
1615		return -EINVAL;
1616
1617	vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1618	return 0;
1619}
1620EXPORT_SYMBOL(remap_pfn_range);
1621
1622int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1623{
1624	unsigned long pfn = start >> PAGE_SHIFT;
1625	unsigned long vm_len = vma->vm_end - vma->vm_start;
1626
1627	pfn += vma->vm_pgoff;
1628	return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1629}
1630EXPORT_SYMBOL(vm_iomap_memory);
1631
1632int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1633			unsigned long pgoff)
1634{
1635	unsigned int size = vma->vm_end - vma->vm_start;
1636
1637	if (!(vma->vm_flags & VM_USERMAP))
1638		return -EINVAL;
1639
1640	vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1641	vma->vm_end = vma->vm_start + size;
1642
1643	return 0;
1644}
1645EXPORT_SYMBOL(remap_vmalloc_range);
1646
1647unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1648	unsigned long len, unsigned long pgoff, unsigned long flags)
1649{
1650	return -ENOMEM;
1651}
1652
1653vm_fault_t filemap_fault(struct vm_fault *vmf)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1654{
1655	BUG();
1656	return 0;
1657}
1658EXPORT_SYMBOL(filemap_fault);
1659
1660vm_fault_t filemap_map_pages(struct vm_fault *vmf,
1661		pgoff_t start_pgoff, pgoff_t end_pgoff)
1662{
1663	BUG();
1664	return 0;
1665}
1666EXPORT_SYMBOL(filemap_map_pages);
1667
1668int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
1669		       int len, unsigned int gup_flags)
1670{
1671	struct vm_area_struct *vma;
1672	int write = gup_flags & FOLL_WRITE;
1673
1674	if (mmap_read_lock_killable(mm))
1675		return 0;
1676
1677	/* the access must start within one of the target process's mappings */
1678	vma = find_vma(mm, addr);
1679	if (vma) {
1680		/* don't overrun this mapping */
1681		if (addr + len >= vma->vm_end)
1682			len = vma->vm_end - addr;
1683
1684		/* only read or write mappings where it is permitted */
1685		if (write && vma->vm_flags & VM_MAYWRITE)
1686			copy_to_user_page(vma, NULL, addr,
1687					 (void *) addr, buf, len);
1688		else if (!write && vma->vm_flags & VM_MAYREAD)
1689			copy_from_user_page(vma, NULL, addr,
1690					    buf, (void *) addr, len);
1691		else
1692			len = 0;
1693	} else {
1694		len = 0;
1695	}
1696
1697	mmap_read_unlock(mm);
1698
1699	return len;
1700}
1701
1702/**
1703 * access_remote_vm - access another process' address space
1704 * @mm:		the mm_struct of the target address space
1705 * @addr:	start address to access
1706 * @buf:	source or destination buffer
1707 * @len:	number of bytes to transfer
1708 * @gup_flags:	flags modifying lookup behaviour
1709 *
1710 * The caller must hold a reference on @mm.
1711 */
1712int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1713		void *buf, int len, unsigned int gup_flags)
1714{
1715	return __access_remote_vm(mm, addr, buf, len, gup_flags);
1716}
1717
1718/*
1719 * Access another process' address space.
1720 * - source/target buffer must be kernel space
1721 */
1722int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1723		unsigned int gup_flags)
1724{
1725	struct mm_struct *mm;
1726
1727	if (addr + len < addr)
1728		return 0;
1729
1730	mm = get_task_mm(tsk);
1731	if (!mm)
1732		return 0;
1733
1734	len = __access_remote_vm(mm, addr, buf, len, gup_flags);
1735
1736	mmput(mm);
1737	return len;
1738}
1739EXPORT_SYMBOL_GPL(access_process_vm);
1740
1741/**
1742 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1743 * @inode: The inode to check
1744 * @size: The current filesize of the inode
1745 * @newsize: The proposed filesize of the inode
1746 *
1747 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1748 * make sure that any outstanding VMAs aren't broken and then shrink the
1749 * vm_regions that extend beyond so that do_mmap() doesn't
1750 * automatically grant mappings that are too large.
1751 */
1752int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1753				size_t newsize)
1754{
1755	struct vm_area_struct *vma;
 
1756	struct vm_region *region;
1757	pgoff_t low, high;
1758	size_t r_size, r_top;
1759
1760	low = newsize >> PAGE_SHIFT;
1761	high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1762
1763	down_write(&nommu_region_sem);
1764	i_mmap_lock_read(inode->i_mapping);
1765
1766	/* search for VMAs that fall within the dead zone */
1767	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
 
1768		/* found one - only interested if it's shared out of the page
1769		 * cache */
1770		if (vma->vm_flags & VM_SHARED) {
1771			i_mmap_unlock_read(inode->i_mapping);
1772			up_write(&nommu_region_sem);
1773			return -ETXTBSY; /* not quite true, but near enough */
1774		}
1775	}
1776
1777	/* reduce any regions that overlap the dead zone - if in existence,
1778	 * these will be pointed to by VMAs that don't overlap the dead zone
1779	 *
1780	 * we don't check for any regions that start beyond the EOF as there
1781	 * shouldn't be any
1782	 */
1783	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
 
1784		if (!(vma->vm_flags & VM_SHARED))
1785			continue;
1786
1787		region = vma->vm_region;
1788		r_size = region->vm_top - region->vm_start;
1789		r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1790
1791		if (r_top > newsize) {
1792			region->vm_top -= r_top - newsize;
1793			if (region->vm_end > region->vm_top)
1794				region->vm_end = region->vm_top;
1795		}
1796	}
1797
1798	i_mmap_unlock_read(inode->i_mapping);
1799	up_write(&nommu_region_sem);
1800	return 0;
1801}
1802
1803/*
1804 * Initialise sysctl_user_reserve_kbytes.
1805 *
1806 * This is intended to prevent a user from starting a single memory hogging
1807 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1808 * mode.
1809 *
1810 * The default value is min(3% of free memory, 128MB)
1811 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1812 */
1813static int __meminit init_user_reserve(void)
1814{
1815	unsigned long free_kbytes;
1816
1817	free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1818
1819	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1820	return 0;
1821}
1822subsys_initcall(init_user_reserve);
1823
1824/*
1825 * Initialise sysctl_admin_reserve_kbytes.
1826 *
1827 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1828 * to log in and kill a memory hogging process.
1829 *
1830 * Systems with more than 256MB will reserve 8MB, enough to recover
1831 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1832 * only reserve 3% of free pages by default.
1833 */
1834static int __meminit init_admin_reserve(void)
1835{
1836	unsigned long free_kbytes;
1837
1838	free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1839
1840	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1841	return 0;
1842}
1843subsys_initcall(init_admin_reserve);