nommu.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/mm/nommu.c
   4 *
   5 *  Replacement code for mm functions to support CPU's that don't
   6 *  have any form of memory management unit (thus no virtual memory).
   7 *
   8 *  See Documentation/admin-guide/mm/nommu-mmap.rst
   9 *
  10 *  Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
  11 *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
  12 *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
  13 *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  14 *  Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
  15 */
  16
  17#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  18
  19#include <linux/export.h>
  20#include <linux/mm.h>
  21#include <linux/sched/mm.h>
  22#include <linux/mman.h>
  23#include <linux/swap.h>
  24#include <linux/file.h>
  25#include <linux/highmem.h>
  26#include <linux/pagemap.h>
  27#include <linux/slab.h>
  28#include <linux/vmalloc.h>
  29#include <linux/backing-dev.h>
  30#include <linux/compiler.h>
  31#include <linux/mount.h>
  32#include <linux/personality.h>
  33#include <linux/security.h>
  34#include <linux/syscalls.h>
  35#include <linux/audit.h>
  36#include <linux/printk.h>
  37
  38#include <linux/uaccess.h>
  39#include <linux/uio.h>
  40#include <asm/tlb.h>
  41#include <asm/tlbflush.h>
  42#include <asm/mmu_context.h>
  43#include "internal.h"
  44
  45unsigned long highest_memmap_pfn;
  46int heap_stack_gap = 0;
  47
  48atomic_long_t mmap_pages_allocated;
  49
  50
  51/* list of mapped, potentially shareable regions */
  52static struct kmem_cache *vm_region_jar;
  53struct rb_root nommu_region_tree = RB_ROOT;
  54DECLARE_RWSEM(nommu_region_sem);
  55
  56const struct vm_operations_struct generic_file_vm_ops = {
  57};
  58
  59/*
  60 * Return the total memory allocated for this pointer, not
  61 * just what the caller asked for.
  62 *
  63 * Doesn't have to be accurate, i.e. may have races.
  64 */
  65unsigned int kobjsize(const void *objp)
  66{
  67	struct folio *folio;
  68
  69	/*
  70	 * If the object we have should not have ksize performed on it,
  71	 * return size of 0
  72	 */
  73	if (!objp || !virt_addr_valid(objp))
  74		return 0;
  75
  76	folio = virt_to_folio(objp);
  77
  78	/*
  79	 * If the allocator sets PageSlab, we know the pointer came from
  80	 * kmalloc().
  81	 */
  82	if (folio_test_slab(folio))
  83		return ksize(objp);
  84
  85	/*
  86	 * If it's not a large folio, see if we have a matching VMA
  87	 * region. This test is intentionally done in reverse order,
  88	 * so if there's no VMA, we still fall through and hand back
  89	 * PAGE_SIZE for 0-order folios.
  90	 */
  91	if (!folio_test_large(folio)) {
  92		struct vm_area_struct *vma;
  93
  94		vma = find_vma(current->mm, (unsigned long)objp);
  95		if (vma)
  96			return vma->vm_end - vma->vm_start;
  97	}
  98
  99	/*
 100	 * The ksize() function is only guaranteed to work for pointers
 101	 * returned by kmalloc(). So handle arbitrary pointers here.
 102	 */
 103	return folio_size(folio);
 104}
 105
 106void vfree(const void *addr)
 107{
 108	kfree(addr);
 109}
 110EXPORT_SYMBOL(vfree);
 111
 112void *__vmalloc_noprof(unsigned long size, gfp_t gfp_mask)
 113{
 114	/*
 115	 *  You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
 116	 * returns only a logical address.
 117	 */
 118	return kmalloc_noprof(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
 119}
 120EXPORT_SYMBOL(__vmalloc_noprof);
 121
 122void *vrealloc_node_align_noprof(const void *p, size_t size, unsigned long align,
 123				 gfp_t flags, int node)
 124{
 125	return krealloc_noprof(p, size, (flags | __GFP_COMP) & ~__GFP_HIGHMEM);
 126}
 127
 128void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,
 129		unsigned long start, unsigned long end, gfp_t gfp_mask,
 130		pgprot_t prot, unsigned long vm_flags, int node,
 131		const void *caller)
 132{
 133	return __vmalloc_noprof(size, gfp_mask);
 134}
 135
 136void *__vmalloc_node_noprof(unsigned long size, unsigned long align, gfp_t gfp_mask,
 137		int node, const void *caller)
 138{
 139	return __vmalloc_noprof(size, gfp_mask);
 140}
 141
 142static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
 143{
 144	void *ret;
 145
 146	ret = __vmalloc(size, flags);
 147	if (ret) {
 148		struct vm_area_struct *vma;
 149
 150		mmap_write_lock(current->mm);
 151		vma = find_vma(current->mm, (unsigned long)ret);
 152		if (vma)
 153			vm_flags_set(vma, VM_USERMAP);
 154		mmap_write_unlock(current->mm);
 155	}
 156
 157	return ret;
 158}
 159
 160void *vmalloc_user_noprof(unsigned long size)
 161{
 162	return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
 163}
 164EXPORT_SYMBOL(vmalloc_user_noprof);
 165
 166struct page *vmalloc_to_page(const void *addr)
 167{
 168	return virt_to_page(addr);
 169}
 170EXPORT_SYMBOL(vmalloc_to_page);
 171
 172unsigned long vmalloc_to_pfn(const void *addr)
 173{
 174	return page_to_pfn(virt_to_page(addr));
 175}
 176EXPORT_SYMBOL(vmalloc_to_pfn);
 177
 178long vread_iter(struct iov_iter *iter, const char *addr, size_t count)
 179{
 180	/* Don't allow overflow */
 181	if ((unsigned long) addr + count < count)
 182		count = -(unsigned long) addr;
 183
 184	return copy_to_iter(addr, count, iter);
 185}
 186
 187/*
 188 *	vmalloc  -  allocate virtually contiguous memory
 189 *
 190 *	@size:		allocation size
 191 *
 192 *	Allocate enough pages to cover @size from the page level
 193 *	allocator and map them into contiguous kernel virtual space.
 194 *
 195 *	For tight control over page level allocator and protection flags
 196 *	use __vmalloc() instead.
 197 */
 198void *vmalloc_noprof(unsigned long size)
 199{
 200	return __vmalloc_noprof(size, GFP_KERNEL);
 201}
 202EXPORT_SYMBOL(vmalloc_noprof);
 203
 204/*
 205 *	vmalloc_huge_node  -  allocate virtually contiguous memory, on a node
 206 *
 207 *	@size:		allocation size
 208 *	@gfp_mask:	flags for the page level allocator
 209 *	@node:          node to use for allocation or NUMA_NO_NODE
 210 *
 211 *	Allocate enough pages to cover @size from the page level
 212 *	allocator and map them into contiguous kernel virtual space.
 213 *
 214 *	Due to NOMMU implications the node argument and HUGE page attribute is
 215 *	ignored.
 216 */
 217void *vmalloc_huge_node_noprof(unsigned long size, gfp_t gfp_mask, int node)
 218{
 219	return __vmalloc_noprof(size, gfp_mask);
 220}
 221
 222/*
 223 *	vzalloc - allocate virtually contiguous memory with zero fill
 224 *
 225 *	@size:		allocation size
 226 *
 227 *	Allocate enough pages to cover @size from the page level
 228 *	allocator and map them into contiguous kernel virtual space.
 229 *	The memory allocated is set to zero.
 230 *
 231 *	For tight control over page level allocator and protection flags
 232 *	use __vmalloc() instead.
 233 */
 234void *vzalloc_noprof(unsigned long size)
 235{
 236	return __vmalloc_noprof(size, GFP_KERNEL | __GFP_ZERO);
 237}
 238EXPORT_SYMBOL(vzalloc_noprof);
 239
 240/**
 241 * vmalloc_node - allocate memory on a specific node
 242 * @size:	allocation size
 243 * @node:	numa node
 244 *
 245 * Allocate enough pages to cover @size from the page level
 246 * allocator and map them into contiguous kernel virtual space.
 247 *
 248 * For tight control over page level allocator and protection flags
 249 * use __vmalloc() instead.
 250 */
 251void *vmalloc_node_noprof(unsigned long size, int node)
 252{
 253	return vmalloc_noprof(size);
 254}
 255EXPORT_SYMBOL(vmalloc_node_noprof);
 256
 257/**
 258 * vzalloc_node - allocate memory on a specific node with zero fill
 259 * @size:	allocation size
 260 * @node:	numa node
 261 *
 262 * Allocate enough pages to cover @size from the page level
 263 * allocator and map them into contiguous kernel virtual space.
 264 * The memory allocated is set to zero.
 265 *
 266 * For tight control over page level allocator and protection flags
 267 * use __vmalloc() instead.
 268 */
 269void *vzalloc_node_noprof(unsigned long size, int node)
 270{
 271	return vzalloc_noprof(size);
 272}
 273EXPORT_SYMBOL(vzalloc_node_noprof);
 274
 275/**
 276 * vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 277 *	@size:		allocation size
 278 *
 279 *	Allocate enough 32bit PA addressable pages to cover @size from the
 280 *	page level allocator and map them into contiguous kernel virtual space.
 281 */
 282void *vmalloc_32_noprof(unsigned long size)
 283{
 284	return __vmalloc_noprof(size, GFP_KERNEL);
 285}
 286EXPORT_SYMBOL(vmalloc_32_noprof);
 287
 288/**
 289 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 290 *	@size:		allocation size
 291 *
 292 * The resulting memory area is 32bit addressable and zeroed so it can be
 293 * mapped to userspace without leaking data.
 294 *
 295 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
 296 * remap_vmalloc_range() are permissible.
 297 */
 298void *vmalloc_32_user_noprof(unsigned long size)
 299{
 300	/*
 301	 * We'll have to sort out the ZONE_DMA bits for 64-bit,
 302	 * but for now this can simply use vmalloc_user() directly.
 303	 */
 304	return vmalloc_user_noprof(size);
 305}
 306EXPORT_SYMBOL(vmalloc_32_user_noprof);
 307
 308void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 309{
 310	BUG();
 311	return NULL;
 312}
 313EXPORT_SYMBOL(vmap);
 314
 315void vunmap(const void *addr)
 316{
 317	BUG();
 318}
 319EXPORT_SYMBOL(vunmap);
 320
 321void *vm_map_ram(struct page **pages, unsigned int count, int node)
 322{
 323	BUG();
 324	return NULL;
 325}
 326EXPORT_SYMBOL(vm_map_ram);
 327
 328void vm_unmap_ram(const void *mem, unsigned int count)
 329{
 330	BUG();
 331}
 332EXPORT_SYMBOL(vm_unmap_ram);
 333
 334void vm_unmap_aliases(void)
 335{
 336}
 337EXPORT_SYMBOL_GPL(vm_unmap_aliases);
 338
 339void free_vm_area(struct vm_struct *area)
 340{
 341	BUG();
 342}
 343EXPORT_SYMBOL_GPL(free_vm_area);
 344
 345int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
 346		   struct page *page)
 347{
 348	return -EINVAL;
 349}
 350EXPORT_SYMBOL(vm_insert_page);
 351
 352int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
 353			struct page **pages, unsigned long *num)
 354{
 355	return -EINVAL;
 356}
 357EXPORT_SYMBOL(vm_insert_pages);
 358
 359int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
 360			unsigned long num)
 361{
 362	return -EINVAL;
 363}
 364EXPORT_SYMBOL(vm_map_pages);
 365
 366int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
 367				unsigned long num)
 368{
 369	return -EINVAL;
 370}
 371EXPORT_SYMBOL(vm_map_pages_zero);
 372
 373/*
 374 *  sys_brk() for the most part doesn't need the global kernel
 375 *  lock, except when an application is doing something nasty
 376 *  like trying to un-brk an area that has already been mapped
 377 *  to a regular file.  in this case, the unmapping will need
 378 *  to invoke file system routines that need the global lock.
 379 */
 380SYSCALL_DEFINE1(brk, unsigned long, brk)
 381{
 382	struct mm_struct *mm = current->mm;
 383
 384	if (brk < mm->start_brk || brk > mm->context.end_brk)
 385		return mm->brk;
 386
 387	if (mm->brk == brk)
 388		return mm->brk;
 389
 390	/*
 391	 * Always allow shrinking brk
 392	 */
 393	if (brk <= mm->brk) {
 394		mm->brk = brk;
 395		return brk;
 396	}
 397
 398	/*
 399	 * Ok, looks good - let it rip.
 400	 */
 401	flush_icache_user_range(mm->brk, brk);
 402	return mm->brk = brk;
 403}
 404
 405static int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
 406
 407static const struct ctl_table nommu_table[] = {
 408	{
 409		.procname	= "nr_trim_pages",
 410		.data		= &sysctl_nr_trim_pages,
 411		.maxlen		= sizeof(sysctl_nr_trim_pages),
 412		.mode		= 0644,
 413		.proc_handler	= proc_dointvec_minmax,
 414		.extra1		= SYSCTL_ZERO,
 415	},
 416};
 417
 418/*
 419 * initialise the percpu counter for VM and region record slabs, initialise VMA
 420 * state.
 421 */
 422void __init mmap_init(void)
 423{
 424	int ret;
 425
 426	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
 427	VM_BUG_ON(ret);
 428	vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
 429	register_sysctl_init("vm", nommu_table);
 430	vma_state_init();
 431}
 432
 433/*
 434 * validate the region tree
 435 * - the caller must hold the region lock
 436 */
 437#ifdef CONFIG_DEBUG_NOMMU_REGIONS
 438static noinline void validate_nommu_regions(void)
 439{
 440	struct vm_region *region, *last;
 441	struct rb_node *p, *lastp;
 442
 443	lastp = rb_first(&nommu_region_tree);
 444	if (!lastp)
 445		return;
 446
 447	last = rb_entry(lastp, struct vm_region, vm_rb);
 448	BUG_ON(last->vm_end <= last->vm_start);
 449	BUG_ON(last->vm_top < last->vm_end);
 450
 451	while ((p = rb_next(lastp))) {
 452		region = rb_entry(p, struct vm_region, vm_rb);
 453		last = rb_entry(lastp, struct vm_region, vm_rb);
 454
 455		BUG_ON(region->vm_end <= region->vm_start);
 456		BUG_ON(region->vm_top < region->vm_end);
 457		BUG_ON(region->vm_start < last->vm_top);
 458
 459		lastp = p;
 460	}
 461}
 462#else
 463static void validate_nommu_regions(void)
 464{
 465}
 466#endif
 467
 468/*
 469 * add a region into the global tree
 470 */
 471static void add_nommu_region(struct vm_region *region)
 472{
 473	struct vm_region *pregion;
 474	struct rb_node **p, *parent;
 475
 476	validate_nommu_regions();
 477
 478	parent = NULL;
 479	p = &nommu_region_tree.rb_node;
 480	while (*p) {
 481		parent = *p;
 482		pregion = rb_entry(parent, struct vm_region, vm_rb);
 483		if (region->vm_start < pregion->vm_start)
 484			p = &(*p)->rb_left;
 485		else if (region->vm_start > pregion->vm_start)
 486			p = &(*p)->rb_right;
 487		else if (pregion == region)
 488			return;
 489		else
 490			BUG();
 491	}
 492
 493	rb_link_node(&region->vm_rb, parent, p);
 494	rb_insert_color(&region->vm_rb, &nommu_region_tree);
 495
 496	validate_nommu_regions();
 497}
 498
 499/*
 500 * delete a region from the global tree
 501 */
 502static void delete_nommu_region(struct vm_region *region)
 503{
 504	BUG_ON(!nommu_region_tree.rb_node);
 505
 506	validate_nommu_regions();
 507	rb_erase(&region->vm_rb, &nommu_region_tree);
 508	validate_nommu_regions();
 509}
 510
 511/*
 512 * free a contiguous series of pages
 513 */
 514static void free_page_series(unsigned long from, unsigned long to)
 515{
 516	for (; from < to; from += PAGE_SIZE) {
 517		struct page *page = virt_to_page((void *)from);
 518
 519		atomic_long_dec(&mmap_pages_allocated);
 520		put_page(page);
 521	}
 522}
 523
 524/*
 525 * release a reference to a region
 526 * - the caller must hold the region semaphore for writing, which this releases
 527 * - the region may not have been added to the tree yet, in which case vm_top
 528 *   will equal vm_start
 529 */
 530static void __put_nommu_region(struct vm_region *region)
 531	__releases(nommu_region_sem)
 532{
 533	BUG_ON(!nommu_region_tree.rb_node);
 534
 535	if (--region->vm_usage == 0) {
 536		if (region->vm_top > region->vm_start)
 537			delete_nommu_region(region);
 538		up_write(&nommu_region_sem);
 539
 540		if (region->vm_file)
 541			fput(region->vm_file);
 542
 543		/* IO memory and memory shared directly out of the pagecache
 544		 * from ramfs/tmpfs mustn't be released here */
 545		if (region->vm_flags & VM_MAPPED_COPY)
 546			free_page_series(region->vm_start, region->vm_top);
 547		kmem_cache_free(vm_region_jar, region);
 548	} else {
 549		up_write(&nommu_region_sem);
 550	}
 551}
 552
 553/*
 554 * release a reference to a region
 555 */
 556static void put_nommu_region(struct vm_region *region)
 557{
 558	down_write(&nommu_region_sem);
 559	__put_nommu_region(region);
 560}
 561
 562static void setup_vma_to_mm(struct vm_area_struct *vma, struct mm_struct *mm)
 563{
 564	vma->vm_mm = mm;
 565
 566	/* add the VMA to the mapping */
 567	if (vma->vm_file) {
 568		struct address_space *mapping = vma->vm_file->f_mapping;
 569
 570		i_mmap_lock_write(mapping);
 571		flush_dcache_mmap_lock(mapping);
 572		vma_interval_tree_insert(vma, &mapping->i_mmap);
 573		flush_dcache_mmap_unlock(mapping);
 574		i_mmap_unlock_write(mapping);
 575	}
 576}
 577
 578static void cleanup_vma_from_mm(struct vm_area_struct *vma)
 579{
 580	vma->vm_mm->map_count--;
 581	/* remove the VMA from the mapping */
 582	if (vma->vm_file) {
 583		struct address_space *mapping;
 584		mapping = vma->vm_file->f_mapping;
 585
 586		i_mmap_lock_write(mapping);
 587		flush_dcache_mmap_lock(mapping);
 588		vma_interval_tree_remove(vma, &mapping->i_mmap);
 589		flush_dcache_mmap_unlock(mapping);
 590		i_mmap_unlock_write(mapping);
 591	}
 592}
 593
 594/*
 595 * delete a VMA from its owning mm_struct and address space
 596 */
 597static int delete_vma_from_mm(struct vm_area_struct *vma)
 598{
 599	VMA_ITERATOR(vmi, vma->vm_mm, vma->vm_start);
 600
 601	vma_iter_config(&vmi, vma->vm_start, vma->vm_end);
 602	if (vma_iter_prealloc(&vmi, NULL)) {
 603		pr_warn("Allocation of vma tree for process %d failed\n",
 604		       current->pid);
 605		return -ENOMEM;
 606	}
 607	cleanup_vma_from_mm(vma);
 608
 609	/* remove from the MM's tree and list */
 610	vma_iter_clear(&vmi);
 611	return 0;
 612}
 613/*
 614 * destroy a VMA record
 615 */
 616static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 617{
 618	vma_close(vma);
 619	if (vma->vm_file)
 620		fput(vma->vm_file);
 621	put_nommu_region(vma->vm_region);
 622	vm_area_free(vma);
 623}
 624
 625struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
 626					     unsigned long start_addr,
 627					     unsigned long end_addr)
 628{
 629	unsigned long index = start_addr;
 630
 631	mmap_assert_locked(mm);
 632	return mt_find(&mm->mm_mt, &index, end_addr - 1);
 633}
 634EXPORT_SYMBOL(find_vma_intersection);
 635
 636/*
 637 * look up the first VMA in which addr resides, NULL if none
 638 * - should be called with mm->mmap_lock at least held readlocked
 639 */
 640struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 641{
 642	VMA_ITERATOR(vmi, mm, addr);
 643
 644	return vma_iter_load(&vmi);
 645}
 646EXPORT_SYMBOL(find_vma);
 647
 648/*
 649 * expand a stack to a given address
 650 * - not supported under NOMMU conditions
 651 */
 652int expand_stack_locked(struct vm_area_struct *vma, unsigned long addr)
 653{
 654	return -ENOMEM;
 655}
 656
 657struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr)
 658{
 659	mmap_read_unlock(mm);
 660	return NULL;
 661}
 662
 663/*
 664 * look up the first VMA exactly that exactly matches addr
 665 * - should be called with mm->mmap_lock at least held readlocked
 666 */
 667static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
 668					     unsigned long addr,
 669					     unsigned long len)
 670{
 671	struct vm_area_struct *vma;
 672	unsigned long end = addr + len;
 673	VMA_ITERATOR(vmi, mm, addr);
 674
 675	vma = vma_iter_load(&vmi);
 676	if (!vma)
 677		return NULL;
 678	if (vma->vm_start != addr)
 679		return NULL;
 680	if (vma->vm_end != end)
 681		return NULL;
 682
 683	return vma;
 684}
 685
 686/*
 687 * determine whether a mapping should be permitted and, if so, what sort of
 688 * mapping we're capable of supporting
 689 */
 690static int validate_mmap_request(struct file *file,
 691				 unsigned long addr,
 692				 unsigned long len,
 693				 unsigned long prot,
 694				 unsigned long flags,
 695				 unsigned long pgoff,
 696				 unsigned long *_capabilities)
 697{
 698	unsigned long capabilities, rlen;
 699	int ret;
 700
 701	/* do the simple checks first */
 702	if (flags & MAP_FIXED)
 703		return -EINVAL;
 704
 705	if ((flags & MAP_TYPE) != MAP_PRIVATE &&
 706	    (flags & MAP_TYPE) != MAP_SHARED)
 707		return -EINVAL;
 708
 709	if (!len)
 710		return -EINVAL;
 711
 712	/* Careful about overflows.. */
 713	rlen = PAGE_ALIGN(len);
 714	if (!rlen || rlen > TASK_SIZE)
 715		return -ENOMEM;
 716
 717	/* offset overflow? */
 718	if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
 719		return -EOVERFLOW;
 720
 721	if (file) {
 722		/* files must support mmap */
 723		if (!can_mmap_file(file))
 724			return -ENODEV;
 725
 726		/* work out if what we've got could possibly be shared
 727		 * - we support chardevs that provide their own "memory"
 728		 * - we support files/blockdevs that are memory backed
 729		 */
 730		if (file->f_op->mmap_capabilities) {
 731			capabilities = file->f_op->mmap_capabilities(file);
 732		} else {
 733			/* no explicit capabilities set, so assume some
 734			 * defaults */
 735			switch (file_inode(file)->i_mode & S_IFMT) {
 736			case S_IFREG:
 737			case S_IFBLK:
 738				capabilities = NOMMU_MAP_COPY;
 739				break;
 740
 741			case S_IFCHR:
 742				capabilities =
 743					NOMMU_MAP_DIRECT |
 744					NOMMU_MAP_READ |
 745					NOMMU_MAP_WRITE;
 746				break;
 747
 748			default:
 749				return -EINVAL;
 750			}
 751		}
 752
 753		/* eliminate any capabilities that we can't support on this
 754		 * device */
 755		if (!file->f_op->get_unmapped_area)
 756			capabilities &= ~NOMMU_MAP_DIRECT;
 757		if (!(file->f_mode & FMODE_CAN_READ))
 758			capabilities &= ~NOMMU_MAP_COPY;
 759
 760		/* The file shall have been opened with read permission. */
 761		if (!(file->f_mode & FMODE_READ))
 762			return -EACCES;
 763
 764		if (flags & MAP_SHARED) {
 765			/* do checks for writing, appending and locking */
 766			if ((prot & PROT_WRITE) &&
 767			    !(file->f_mode & FMODE_WRITE))
 768				return -EACCES;
 769
 770			if (IS_APPEND(file_inode(file)) &&
 771			    (file->f_mode & FMODE_WRITE))
 772				return -EACCES;
 773
 774			if (!(capabilities & NOMMU_MAP_DIRECT))
 775				return -ENODEV;
 776
 777			/* we mustn't privatise shared mappings */
 778			capabilities &= ~NOMMU_MAP_COPY;
 779		} else {
 780			/* we're going to read the file into private memory we
 781			 * allocate */
 782			if (!(capabilities & NOMMU_MAP_COPY))
 783				return -ENODEV;
 784
 785			/* we don't permit a private writable mapping to be
 786			 * shared with the backing device */
 787			if (prot & PROT_WRITE)
 788				capabilities &= ~NOMMU_MAP_DIRECT;
 789		}
 790
 791		if (capabilities & NOMMU_MAP_DIRECT) {
 792			if (((prot & PROT_READ)  && !(capabilities & NOMMU_MAP_READ))  ||
 793			    ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
 794			    ((prot & PROT_EXEC)  && !(capabilities & NOMMU_MAP_EXEC))
 795			    ) {
 796				capabilities &= ~NOMMU_MAP_DIRECT;
 797				if (flags & MAP_SHARED) {
 798					pr_warn("MAP_SHARED not completely supported on !MMU\n");
 799					return -EINVAL;
 800				}
 801			}
 802		}
 803
 804		/* handle executable mappings and implied executable
 805		 * mappings */
 806		if (path_noexec(&file->f_path)) {
 807			if (prot & PROT_EXEC)
 808				return -EPERM;
 809		} else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
 810			/* handle implication of PROT_EXEC by PROT_READ */
 811			if (current->personality & READ_IMPLIES_EXEC) {
 812				if (capabilities & NOMMU_MAP_EXEC)
 813					prot |= PROT_EXEC;
 814			}
 815		} else if ((prot & PROT_READ) &&
 816			 (prot & PROT_EXEC) &&
 817			 !(capabilities & NOMMU_MAP_EXEC)
 818			 ) {
 819			/* backing file is not executable, try to copy */
 820			capabilities &= ~NOMMU_MAP_DIRECT;
 821		}
 822	} else {
 823		/* anonymous mappings are always memory backed and can be
 824		 * privately mapped
 825		 */
 826		capabilities = NOMMU_MAP_COPY;
 827
 828		/* handle PROT_EXEC implication by PROT_READ */
 829		if ((prot & PROT_READ) &&
 830		    (current->personality & READ_IMPLIES_EXEC))
 831			prot |= PROT_EXEC;
 832	}
 833
 834	/* allow the security API to have its say */
 835	ret = security_mmap_addr(addr);
 836	if (ret < 0)
 837		return ret;
 838
 839	/* looks okay */
 840	*_capabilities = capabilities;
 841	return 0;
 842}
 843
 844/*
 845 * we've determined that we can make the mapping, now translate what we
 846 * now know into VMA flags
 847 */
 848static vm_flags_t determine_vm_flags(struct file *file,
 849		unsigned long prot,
 850		unsigned long flags,
 851		unsigned long capabilities)
 852{
 853	vm_flags_t vm_flags;
 854
 855	vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(file, flags);
 856
 857	if (!file) {
 858		/*
 859		 * MAP_ANONYMOUS. MAP_SHARED is mapped to MAP_PRIVATE, because
 860		 * there is no fork().
 861		 */
 862		vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 863	} else if (flags & MAP_PRIVATE) {
 864		/* MAP_PRIVATE file mapping */
 865		if (capabilities & NOMMU_MAP_DIRECT)
 866			vm_flags |= (capabilities & NOMMU_VMFLAGS);
 867		else
 868			vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 869
 870		if (!(prot & PROT_WRITE) && !current->ptrace)
 871			/*
 872			 * R/O private file mapping which cannot be used to
 873			 * modify memory, especially also not via active ptrace
 874			 * (e.g., set breakpoints) or later by upgrading
 875			 * permissions (no mprotect()). We can try overlaying
 876			 * the file mapping, which will work e.g., on chardevs,
 877			 * ramfs/tmpfs/shmfs and romfs/cramf.
 878			 */
 879			vm_flags |= VM_MAYOVERLAY;
 880	} else {
 881		/* MAP_SHARED file mapping: NOMMU_MAP_DIRECT is set. */
 882		vm_flags |= VM_SHARED | VM_MAYSHARE |
 883			    (capabilities & NOMMU_VMFLAGS);
 884	}
 885
 886	return vm_flags;
 887}
 888
 889/*
 890 * set up a shared mapping on a file (the driver or filesystem provides and
 891 * pins the storage)
 892 */
 893static int do_mmap_shared_file(struct vm_area_struct *vma)
 894{
 895	int ret;
 896
 897	ret = mmap_file(vma->vm_file, vma);
 898	if (ret == 0) {
 899		vma->vm_region->vm_top = vma->vm_region->vm_end;
 900		return 0;
 901	}
 902	if (ret != -ENOSYS)
 903		return ret;
 904
 905	/* getting -ENOSYS indicates that direct mmap isn't possible (as
 906	 * opposed to tried but failed) so we can only give a suitable error as
 907	 * it's not possible to make a private copy if MAP_SHARED was given */
 908	return -ENODEV;
 909}
 910
 911/*
 912 * set up a private mapping or an anonymous shared mapping
 913 */
 914static int do_mmap_private(struct vm_area_struct *vma,
 915			   struct vm_region *region,
 916			   unsigned long len,
 917			   unsigned long capabilities)
 918{
 919	unsigned long total, point;
 920	void *base;
 921	int ret, order;
 922
 923	/*
 924	 * Invoke the file's mapping function so that it can keep track of
 925	 * shared mappings on devices or memory. VM_MAYOVERLAY will be set if
 926	 * it may attempt to share, which will make is_nommu_shared_mapping()
 927	 * happy.
 928	 */
 929	if (capabilities & NOMMU_MAP_DIRECT) {
 930		ret = mmap_file(vma->vm_file, vma);
 931		/* shouldn't return success if we're not sharing */
 932		if (WARN_ON_ONCE(!is_nommu_shared_mapping(vma->vm_flags)))
 933			ret = -ENOSYS;
 934		if (ret == 0) {
 935			vma->vm_region->vm_top = vma->vm_region->vm_end;
 936			return 0;
 937		}
 938		if (ret != -ENOSYS)
 939			return ret;
 940
 941		/* getting an ENOSYS error indicates that direct mmap isn't
 942		 * possible (as opposed to tried but failed) so we'll try to
 943		 * make a private copy of the data and map that instead */
 944	}
 945
 946
 947	/* allocate some memory to hold the mapping
 948	 * - note that this may not return a page-aligned address if the object
 949	 *   we're allocating is smaller than a page
 950	 */
 951	order = get_order(len);
 952	total = 1 << order;
 953	point = len >> PAGE_SHIFT;
 954
 955	/* we don't want to allocate a power-of-2 sized page set */
 956	if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
 957		total = point;
 958
 959	base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
 960	if (!base)
 961		goto enomem;
 962
 963	atomic_long_add(total, &mmap_pages_allocated);
 964
 965	vm_flags_set(vma, VM_MAPPED_COPY);
 966	region->vm_flags = vma->vm_flags;
 967	region->vm_start = (unsigned long) base;
 968	region->vm_end   = region->vm_start + len;
 969	region->vm_top   = region->vm_start + (total << PAGE_SHIFT);
 970
 971	vma->vm_start = region->vm_start;
 972	vma->vm_end   = region->vm_start + len;
 973
 974	if (vma->vm_file) {
 975		/* read the contents of a file into the copy */
 976		loff_t fpos;
 977
 978		fpos = vma->vm_pgoff;
 979		fpos <<= PAGE_SHIFT;
 980
 981		ret = kernel_read(vma->vm_file, base, len, &fpos);
 982		if (ret < 0)
 983			goto error_free;
 984
 985		/* clear the last little bit */
 986		if (ret < len)
 987			memset(base + ret, 0, len - ret);
 988
 989	} else {
 990		vma_set_anonymous(vma);
 991	}
 992
 993	return 0;
 994
 995error_free:
 996	free_page_series(region->vm_start, region->vm_top);
 997	region->vm_start = vma->vm_start = 0;
 998	region->vm_end   = vma->vm_end = 0;
 999	region->vm_top   = 0;
1000	return ret;
1001
1002enomem:
1003	pr_err("Allocation of length %lu from process %d (%s) failed\n",
1004	       len, current->pid, current->comm);
1005	show_mem();
1006	return -ENOMEM;
1007}
1008
1009/*
1010 * handle mapping creation for uClinux
1011 */
1012unsigned long do_mmap(struct file *file,
1013			unsigned long addr,
1014			unsigned long len,
1015			unsigned long prot,
1016			unsigned long flags,
1017			vm_flags_t vm_flags,
1018			unsigned long pgoff,
1019			unsigned long *populate,
1020			struct list_head *uf)
1021{
1022	struct vm_area_struct *vma;
1023	struct vm_region *region;
1024	struct rb_node *rb;
1025	unsigned long capabilities, result;
1026	int ret;
1027	VMA_ITERATOR(vmi, current->mm, 0);
1028
1029	*populate = 0;
1030
1031	/* decide whether we should attempt the mapping, and if so what sort of
1032	 * mapping */
1033	ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1034				    &capabilities);
1035	if (ret < 0)
1036		return ret;
1037
1038	/* we ignore the address hint */
1039	addr = 0;
1040	len = PAGE_ALIGN(len);
1041
1042	/* we've determined that we can make the mapping, now translate what we
1043	 * now know into VMA flags */
1044	vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1045
1046
1047	/* we're going to need to record the mapping */
1048	region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1049	if (!region)
1050		goto error_getting_region;
1051
1052	vma = vm_area_alloc(current->mm);
1053	if (!vma)
1054		goto error_getting_vma;
1055
1056	region->vm_usage = 1;
1057	region->vm_flags = vm_flags;
1058	region->vm_pgoff = pgoff;
1059
1060	vm_flags_init(vma, vm_flags);
1061	vma->vm_pgoff = pgoff;
1062
1063	if (file) {
1064		region->vm_file = get_file(file);
1065		vma->vm_file = get_file(file);
1066	}
1067
1068	down_write(&nommu_region_sem);
1069
1070	/* if we want to share, we need to check for regions created by other
1071	 * mmap() calls that overlap with our proposed mapping
1072	 * - we can only share with a superset match on most regular files
1073	 * - shared mappings on character devices and memory backed files are
1074	 *   permitted to overlap inexactly as far as we are concerned for in
1075	 *   these cases, sharing is handled in the driver or filesystem rather
1076	 *   than here
1077	 */
1078	if (is_nommu_shared_mapping(vm_flags)) {
1079		struct vm_region *pregion;
1080		unsigned long pglen, rpglen, pgend, rpgend, start;
1081
1082		pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1083		pgend = pgoff + pglen;
1084
1085		for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1086			pregion = rb_entry(rb, struct vm_region, vm_rb);
1087
1088			if (!is_nommu_shared_mapping(pregion->vm_flags))
1089				continue;
1090
1091			/* search for overlapping mappings on the same file */
1092			if (file_inode(pregion->vm_file) !=
1093			    file_inode(file))
1094				continue;
1095
1096			if (pregion->vm_pgoff >= pgend)
1097				continue;
1098
1099			rpglen = pregion->vm_end - pregion->vm_start;
1100			rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1101			rpgend = pregion->vm_pgoff + rpglen;
1102			if (pgoff >= rpgend)
1103				continue;
1104
1105			/* handle inexactly overlapping matches between
1106			 * mappings */
1107			if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1108			    !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1109				/* new mapping is not a subset of the region */
1110				if (!(capabilities & NOMMU_MAP_DIRECT))
1111					goto sharing_violation;
1112				continue;
1113			}
1114
1115			/* we've found a region we can share */
1116			pregion->vm_usage++;
1117			vma->vm_region = pregion;
1118			start = pregion->vm_start;
1119			start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1120			vma->vm_start = start;
1121			vma->vm_end = start + len;
1122
1123			if (pregion->vm_flags & VM_MAPPED_COPY)
1124				vm_flags_set(vma, VM_MAPPED_COPY);
1125			else {
1126				ret = do_mmap_shared_file(vma);
1127				if (ret < 0) {
1128					vma->vm_region = NULL;
1129					vma->vm_start = 0;
1130					vma->vm_end = 0;
1131					pregion->vm_usage--;
1132					pregion = NULL;
1133					goto error_just_free;
1134				}
1135			}
1136			fput(region->vm_file);
1137			kmem_cache_free(vm_region_jar, region);
1138			region = pregion;
1139			result = start;
1140			goto share;
1141		}
1142
1143		/* obtain the address at which to make a shared mapping
1144		 * - this is the hook for quasi-memory character devices to
1145		 *   tell us the location of a shared mapping
1146		 */
1147		if (capabilities & NOMMU_MAP_DIRECT) {
1148			addr = file->f_op->get_unmapped_area(file, addr, len,
1149							     pgoff, flags);
1150			if (IS_ERR_VALUE(addr)) {
1151				ret = addr;
1152				if (ret != -ENOSYS)
1153					goto error_just_free;
1154
1155				/* the driver refused to tell us where to site
1156				 * the mapping so we'll have to attempt to copy
1157				 * it */
1158				ret = -ENODEV;
1159				if (!(capabilities & NOMMU_MAP_COPY))
1160					goto error_just_free;
1161
1162				capabilities &= ~NOMMU_MAP_DIRECT;
1163			} else {
1164				vma->vm_start = region->vm_start = addr;
1165				vma->vm_end = region->vm_end = addr + len;
1166			}
1167		}
1168	}
1169
1170	vma->vm_region = region;
1171
1172	/* set up the mapping
1173	 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1174	 */
1175	if (file && vma->vm_flags & VM_SHARED)
1176		ret = do_mmap_shared_file(vma);
1177	else
1178		ret = do_mmap_private(vma, region, len, capabilities);
1179	if (ret < 0)
1180		goto error_just_free;
1181	add_nommu_region(region);
1182
1183	/* clear anonymous mappings that don't ask for uninitialized data */
1184	if (!vma->vm_file &&
1185	    (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1186	     !(flags & MAP_UNINITIALIZED)))
1187		memset((void *)region->vm_start, 0,
1188		       region->vm_end - region->vm_start);
1189
1190	/* okay... we have a mapping; now we have to register it */
1191	result = vma->vm_start;
1192
1193	current->mm->total_vm += len >> PAGE_SHIFT;
1194
1195share:
1196	BUG_ON(!vma->vm_region);
1197	vma_iter_config(&vmi, vma->vm_start, vma->vm_end);
1198	if (vma_iter_prealloc(&vmi, vma))
1199		goto error_just_free;
1200
1201	setup_vma_to_mm(vma, current->mm);
1202	current->mm->map_count++;
1203	/* add the VMA to the tree */
1204	vma_iter_store_new(&vmi, vma);
1205
1206	/* we flush the region from the icache only when the first executable
1207	 * mapping of it is made  */
1208	if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1209		flush_icache_user_range(region->vm_start, region->vm_end);
1210		region->vm_icache_flushed = true;
1211	}
1212
1213	up_write(&nommu_region_sem);
1214
1215	return result;
1216
1217error_just_free:
1218	up_write(&nommu_region_sem);
1219error:
1220	vma_iter_free(&vmi);
1221	if (region->vm_file)
1222		fput(region->vm_file);
1223	kmem_cache_free(vm_region_jar, region);
1224	if (vma->vm_file)
1225		fput(vma->vm_file);
1226	vm_area_free(vma);
1227	return ret;
1228
1229sharing_violation:
1230	up_write(&nommu_region_sem);
1231	pr_warn("Attempt to share mismatched mappings\n");
1232	ret = -EINVAL;
1233	goto error;
1234
1235error_getting_vma:
1236	kmem_cache_free(vm_region_jar, region);
1237	pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1238			len, current->pid);
1239	show_mem();
1240	return -ENOMEM;
1241
1242error_getting_region:
1243	pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1244			len, current->pid);
1245	show_mem();
1246	return -ENOMEM;
1247}
1248
1249unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1250			      unsigned long prot, unsigned long flags,
1251			      unsigned long fd, unsigned long pgoff)
1252{
1253	struct file *file = NULL;
1254	unsigned long retval = -EBADF;
1255
1256	audit_mmap_fd(fd, flags);
1257	if (!(flags & MAP_ANONYMOUS)) {
1258		file = fget(fd);
1259		if (!file)
1260			goto out;
1261	}
1262
1263	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1264
1265	if (file)
1266		fput(file);
1267out:
1268	return retval;
1269}
1270
1271SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1272		unsigned long, prot, unsigned long, flags,
1273		unsigned long, fd, unsigned long, pgoff)
1274{
1275	return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1276}
1277
1278#ifdef __ARCH_WANT_SYS_OLD_MMAP
1279struct mmap_arg_struct {
1280	unsigned long addr;
1281	unsigned long len;
1282	unsigned long prot;
1283	unsigned long flags;
1284	unsigned long fd;
1285	unsigned long offset;
1286};
1287
1288SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1289{
1290	struct mmap_arg_struct a;
1291
1292	if (copy_from_user(&a, arg, sizeof(a)))
1293		return -EFAULT;
1294	if (offset_in_page(a.offset))
1295		return -EINVAL;
1296
1297	return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1298			       a.offset >> PAGE_SHIFT);
1299}
1300#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1301
1302/*
1303 * split a vma into two pieces at address 'addr', a new vma is allocated either
1304 * for the first part or the tail.
1305 */
1306static int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
1307		     unsigned long addr, int new_below)
1308{
1309	struct vm_area_struct *new;
1310	struct vm_region *region;
1311	unsigned long npages;
1312	struct mm_struct *mm;
1313
1314	/* we're only permitted to split anonymous regions (these should have
1315	 * only a single usage on the region) */
1316	if (vma->vm_file)
1317		return -ENOMEM;
1318
1319	mm = vma->vm_mm;
1320	if (mm->map_count >= sysctl_max_map_count)
1321		return -ENOMEM;
1322
1323	region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1324	if (!region)
1325		return -ENOMEM;
1326
1327	new = vm_area_dup(vma);
1328	if (!new)
1329		goto err_vma_dup;
1330
1331	/* most fields are the same, copy all, and then fixup */
1332	*region = *vma->vm_region;
1333	new->vm_region = region;
1334
1335	npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1336
1337	if (new_below) {
1338		region->vm_top = region->vm_end = new->vm_end = addr;
1339	} else {
1340		region->vm_start = new->vm_start = addr;
1341		region->vm_pgoff = new->vm_pgoff += npages;
1342	}
1343
1344	vma_iter_config(vmi, new->vm_start, new->vm_end);
1345	if (vma_iter_prealloc(vmi, vma)) {
1346		pr_warn("Allocation of vma tree for process %d failed\n",
1347			current->pid);
1348		goto err_vmi_preallocate;
1349	}
1350
1351	if (new->vm_ops && new->vm_ops->open)
1352		new->vm_ops->open(new);
1353
1354	down_write(&nommu_region_sem);
1355	delete_nommu_region(vma->vm_region);
1356	if (new_below) {
1357		vma->vm_region->vm_start = vma->vm_start = addr;
1358		vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1359	} else {
1360		vma->vm_region->vm_end = vma->vm_end = addr;
1361		vma->vm_region->vm_top = addr;
1362	}
1363	add_nommu_region(vma->vm_region);
1364	add_nommu_region(new->vm_region);
1365	up_write(&nommu_region_sem);
1366
1367	setup_vma_to_mm(vma, mm);
1368	setup_vma_to_mm(new, mm);
1369	vma_iter_store_new(vmi, new);
1370	mm->map_count++;
1371	return 0;
1372
1373err_vmi_preallocate:
1374	vm_area_free(new);
1375err_vma_dup:
1376	kmem_cache_free(vm_region_jar, region);
1377	return -ENOMEM;
1378}
1379
1380/*
1381 * shrink a VMA by removing the specified chunk from either the beginning or
1382 * the end
1383 */
1384static int vmi_shrink_vma(struct vma_iterator *vmi,
1385		      struct vm_area_struct *vma,
1386		      unsigned long from, unsigned long to)
1387{
1388	struct vm_region *region;
1389
1390	/* adjust the VMA's pointers, which may reposition it in the MM's tree
1391	 * and list */
1392	if (from > vma->vm_start) {
1393		if (vma_iter_clear_gfp(vmi, from, vma->vm_end, GFP_KERNEL))
1394			return -ENOMEM;
1395		vma->vm_end = from;
1396	} else {
1397		if (vma_iter_clear_gfp(vmi, vma->vm_start, to, GFP_KERNEL))
1398			return -ENOMEM;
1399		vma->vm_start = to;
1400	}
1401
1402	/* cut the backing region down to size */
1403	region = vma->vm_region;
1404	BUG_ON(region->vm_usage != 1);
1405
1406	down_write(&nommu_region_sem);
1407	delete_nommu_region(region);
1408	if (from > region->vm_start) {
1409		to = region->vm_top;
1410		region->vm_top = region->vm_end = from;
1411	} else {
1412		region->vm_start = to;
1413	}
1414	add_nommu_region(region);
1415	up_write(&nommu_region_sem);
1416
1417	free_page_series(from, to);
1418	return 0;
1419}
1420
1421/*
1422 * release a mapping
1423 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1424 *   VMA, though it need not cover the whole VMA
1425 */
1426int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
1427{
1428	VMA_ITERATOR(vmi, mm, start);
1429	struct vm_area_struct *vma;
1430	unsigned long end;
1431	int ret = 0;
1432
1433	len = PAGE_ALIGN(len);
1434	if (len == 0)
1435		return -EINVAL;
1436
1437	end = start + len;
1438
1439	/* find the first potentially overlapping VMA */
1440	vma = vma_find(&vmi, end);
1441	if (!vma) {
1442		static int limit;
1443		if (limit < 5) {
1444			pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1445					current->pid, current->comm,
1446					start, start + len - 1);
1447			limit++;
1448		}
1449		return -EINVAL;
1450	}
1451
1452	/* we're allowed to split an anonymous VMA but not a file-backed one */
1453	if (vma->vm_file) {
1454		do {
1455			if (start > vma->vm_start)
1456				return -EINVAL;
1457			if (end == vma->vm_end)
1458				goto erase_whole_vma;
1459			vma = vma_find(&vmi, end);
1460		} while (vma);
1461		return -EINVAL;
1462	} else {
1463		/* the chunk must be a subset of the VMA found */
1464		if (start == vma->vm_start && end == vma->vm_end)
1465			goto erase_whole_vma;
1466		if (start < vma->vm_start || end > vma->vm_end)
1467			return -EINVAL;
1468		if (offset_in_page(start))
1469			return -EINVAL;
1470		if (end != vma->vm_end && offset_in_page(end))
1471			return -EINVAL;
1472		if (start != vma->vm_start && end != vma->vm_end) {
1473			ret = split_vma(&vmi, vma, start, 1);
1474			if (ret < 0)
1475				return ret;
1476		}
1477		return vmi_shrink_vma(&vmi, vma, start, end);
1478	}
1479
1480erase_whole_vma:
1481	if (delete_vma_from_mm(vma))
1482		ret = -ENOMEM;
1483	else
1484		delete_vma(mm, vma);
1485	return ret;
1486}
1487
1488int vm_munmap(unsigned long addr, size_t len)
1489{
1490	struct mm_struct *mm = current->mm;
1491	int ret;
1492
1493	mmap_write_lock(mm);
1494	ret = do_munmap(mm, addr, len, NULL);
1495	mmap_write_unlock(mm);
1496	return ret;
1497}
1498EXPORT_SYMBOL(vm_munmap);
1499
1500SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1501{
1502	return vm_munmap(addr, len);
1503}
1504
1505/*
1506 * release all the mappings made in a process's VM space
1507 */
1508void exit_mmap(struct mm_struct *mm)
1509{
1510	VMA_ITERATOR(vmi, mm, 0);
1511	struct vm_area_struct *vma;
1512
1513	if (!mm)
1514		return;
1515
1516	mm->total_vm = 0;
1517
1518	/*
1519	 * Lock the mm to avoid assert complaining even though this is the only
1520	 * user of the mm
1521	 */
1522	mmap_write_lock(mm);
1523	for_each_vma(vmi, vma) {
1524		cleanup_vma_from_mm(vma);
1525		delete_vma(mm, vma);
1526		cond_resched();
1527	}
1528	__mt_destroy(&mm->mm_mt);
1529	mmap_write_unlock(mm);
1530}
1531
1532/*
1533 * expand (or shrink) an existing mapping, potentially moving it at the same
1534 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1535 *
1536 * under NOMMU conditions, we only permit changing a mapping's size, and only
1537 * as long as it stays within the region allocated by do_mmap_private() and the
1538 * block is not shareable
1539 *
1540 * MREMAP_FIXED is not supported under NOMMU conditions
1541 */
1542static unsigned long do_mremap(unsigned long addr,
1543			unsigned long old_len, unsigned long new_len,
1544			unsigned long flags, unsigned long new_addr)
1545{
1546	struct vm_area_struct *vma;
1547
1548	/* insanity checks first */
1549	old_len = PAGE_ALIGN(old_len);
1550	new_len = PAGE_ALIGN(new_len);
1551	if (old_len == 0 || new_len == 0)
1552		return (unsigned long) -EINVAL;
1553
1554	if (offset_in_page(addr))
1555		return -EINVAL;
1556
1557	if (flags & MREMAP_FIXED && new_addr != addr)
1558		return (unsigned long) -EINVAL;
1559
1560	vma = find_vma_exact(current->mm, addr, old_len);
1561	if (!vma)
1562		return (unsigned long) -EINVAL;
1563
1564	if (vma->vm_end != vma->vm_start + old_len)
1565		return (unsigned long) -EFAULT;
1566
1567	if (is_nommu_shared_mapping(vma->vm_flags))
1568		return (unsigned long) -EPERM;
1569
1570	if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1571		return (unsigned long) -ENOMEM;
1572
1573	/* all checks complete - do it */
1574	vma->vm_end = vma->vm_start + new_len;
1575	return vma->vm_start;
1576}
1577
1578SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1579		unsigned long, new_len, unsigned long, flags,
1580		unsigned long, new_addr)
1581{
1582	unsigned long ret;
1583
1584	mmap_write_lock(current->mm);
1585	ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1586	mmap_write_unlock(current->mm);
1587	return ret;
1588}
1589
1590int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1591		unsigned long pfn, unsigned long size, pgprot_t prot)
1592{
1593	if (addr != (pfn << PAGE_SHIFT))
1594		return -EINVAL;
1595
1596	vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
1597	return 0;
1598}
1599EXPORT_SYMBOL(remap_pfn_range);
1600
1601int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1602{
1603	unsigned long pfn = start >> PAGE_SHIFT;
1604	unsigned long vm_len = vma->vm_end - vma->vm_start;
1605
1606	pfn += vma->vm_pgoff;
1607	return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1608}
1609EXPORT_SYMBOL(vm_iomap_memory);
1610
1611int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1612			unsigned long pgoff)
1613{
1614	unsigned int size = vma->vm_end - vma->vm_start;
1615
1616	if (!(vma->vm_flags & VM_USERMAP))
1617		return -EINVAL;
1618
1619	vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1620	vma->vm_end = vma->vm_start + size;
1621
1622	return 0;
1623}
1624EXPORT_SYMBOL(remap_vmalloc_range);
1625
1626vm_fault_t filemap_fault(struct vm_fault *vmf)
1627{
1628	BUG();
1629	return 0;
1630}
1631EXPORT_SYMBOL(filemap_fault);
1632
1633vm_fault_t filemap_map_pages(struct vm_fault *vmf,
1634		pgoff_t start_pgoff, pgoff_t end_pgoff)
1635{
1636	BUG();
1637	return 0;
1638}
1639EXPORT_SYMBOL(filemap_map_pages);
1640
1641static int __access_remote_vm(struct mm_struct *mm, unsigned long addr,
1642			      void *buf, int len, unsigned int gup_flags)
1643{
1644	struct vm_area_struct *vma;
1645	int write = gup_flags & FOLL_WRITE;
1646
1647	if (mmap_read_lock_killable(mm))
1648		return 0;
1649
1650	/* the access must start within one of the target process's mappings */
1651	vma = find_vma(mm, addr);
1652	if (vma) {
1653		/* don't overrun this mapping */
1654		if (addr + len >= vma->vm_end)
1655			len = vma->vm_end - addr;
1656
1657		/* only read or write mappings where it is permitted */
1658		if (write && vma->vm_flags & VM_MAYWRITE)
1659			copy_to_user_page(vma, NULL, addr,
1660					 (void *) addr, buf, len);
1661		else if (!write && vma->vm_flags & VM_MAYREAD)
1662			copy_from_user_page(vma, NULL, addr,
1663					    buf, (void *) addr, len);
1664		else
1665			len = 0;
1666	} else {
1667		len = 0;
1668	}
1669
1670	mmap_read_unlock(mm);
1671
1672	return len;
1673}
1674
1675/**
1676 * access_remote_vm - access another process' address space
1677 * @mm:		the mm_struct of the target address space
1678 * @addr:	start address to access
1679 * @buf:	source or destination buffer
1680 * @len:	number of bytes to transfer
1681 * @gup_flags:	flags modifying lookup behaviour
1682 *
1683 * The caller must hold a reference on @mm.
1684 */
1685int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1686		void *buf, int len, unsigned int gup_flags)
1687{
1688	return __access_remote_vm(mm, addr, buf, len, gup_flags);
1689}
1690
1691/*
1692 * Access another process' address space.
1693 * - source/target buffer must be kernel space
1694 */
1695int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1696		unsigned int gup_flags)
1697{
1698	struct mm_struct *mm;
1699
1700	if (addr + len < addr)
1701		return 0;
1702
1703	mm = get_task_mm(tsk);
1704	if (!mm)
1705		return 0;
1706
1707	len = __access_remote_vm(mm, addr, buf, len, gup_flags);
1708
1709	mmput(mm);
1710	return len;
1711}
1712EXPORT_SYMBOL_GPL(access_process_vm);
1713
1714#ifdef CONFIG_BPF_SYSCALL
1715/*
1716 * Copy a string from another process's address space as given in mm.
1717 * If there is any error return -EFAULT.
1718 */
1719static int __copy_remote_vm_str(struct mm_struct *mm, unsigned long addr,
1720				void *buf, int len)
1721{
1722	unsigned long addr_end;
1723	struct vm_area_struct *vma;
1724	int ret = -EFAULT;
1725
1726	*(char *)buf = '\0';
1727
1728	if (mmap_read_lock_killable(mm))
1729		return ret;
1730
1731	/* the access must start within one of the target process's mappings */
1732	vma = find_vma(mm, addr);
1733	if (!vma)
1734		goto out;
1735
1736	if (check_add_overflow(addr, len, &addr_end))
1737		goto out;
1738
1739	/* don't overrun this mapping */
1740	if (addr_end > vma->vm_end)
1741		len = vma->vm_end - addr;
1742
1743	/* only read mappings where it is permitted */
1744	if (vma->vm_flags & VM_MAYREAD) {
1745		ret = strscpy(buf, (char *)addr, len);
1746		if (ret < 0)
1747			ret = len - 1;
1748	}
1749
1750out:
1751	mmap_read_unlock(mm);
1752	return ret;
1753}
1754
1755/**
1756 * copy_remote_vm_str - copy a string from another process's address space.
1757 * @tsk:	the task of the target address space
1758 * @addr:	start address to read from
1759 * @buf:	destination buffer
1760 * @len:	number of bytes to copy
1761 * @gup_flags:	flags modifying lookup behaviour (unused)
1762 *
1763 * The caller must hold a reference on @mm.
1764 *
1765 * Return: number of bytes copied from @addr (source) to @buf (destination);
1766 * not including the trailing NUL. Always guaranteed to leave NUL-terminated
1767 * buffer. On any error, return -EFAULT.
1768 */
1769int copy_remote_vm_str(struct task_struct *tsk, unsigned long addr,
1770		       void *buf, int len, unsigned int gup_flags)
1771{
1772	struct mm_struct *mm;
1773	int ret;
1774
1775	if (unlikely(len == 0))
1776		return 0;
1777
1778	mm = get_task_mm(tsk);
1779	if (!mm) {
1780		*(char *)buf = '\0';
1781		return -EFAULT;
1782	}
1783
1784	ret = __copy_remote_vm_str(mm, addr, buf, len);
1785
1786	mmput(mm);
1787
1788	return ret;
1789}
1790EXPORT_SYMBOL_GPL(copy_remote_vm_str);
1791#endif /* CONFIG_BPF_SYSCALL */
1792
1793/**
1794 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1795 * @inode: The inode to check
1796 * @size: The current filesize of the inode
1797 * @newsize: The proposed filesize of the inode
1798 *
1799 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1800 * make sure that any outstanding VMAs aren't broken and then shrink the
1801 * vm_regions that extend beyond so that do_mmap() doesn't
1802 * automatically grant mappings that are too large.
1803 */
1804int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1805				size_t newsize)
1806{
1807	struct vm_area_struct *vma;
1808	struct vm_region *region;
1809	pgoff_t low, high;
1810	size_t r_size, r_top;
1811
1812	low = newsize >> PAGE_SHIFT;
1813	high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1814
1815	down_write(&nommu_region_sem);
1816	i_mmap_lock_read(inode->i_mapping);
1817
1818	/* search for VMAs that fall within the dead zone */
1819	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
1820		/* found one - only interested if it's shared out of the page
1821		 * cache */
1822		if (vma->vm_flags & VM_SHARED) {
1823			i_mmap_unlock_read(inode->i_mapping);
1824			up_write(&nommu_region_sem);
1825			return -ETXTBSY; /* not quite true, but near enough */
1826		}
1827	}
1828
1829	/* reduce any regions that overlap the dead zone - if in existence,
1830	 * these will be pointed to by VMAs that don't overlap the dead zone
1831	 *
1832	 * we don't check for any regions that start beyond the EOF as there
1833	 * shouldn't be any
1834	 */
1835	vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
1836		if (!(vma->vm_flags & VM_SHARED))
1837			continue;
1838
1839		region = vma->vm_region;
1840		r_size = region->vm_top - region->vm_start;
1841		r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1842
1843		if (r_top > newsize) {
1844			region->vm_top -= r_top - newsize;
1845			if (region->vm_end > region->vm_top)
1846				region->vm_end = region->vm_top;
1847		}
1848	}
1849
1850	i_mmap_unlock_read(inode->i_mapping);
1851	up_write(&nommu_region_sem);
1852	return 0;
1853}
1854
1855/*
1856 * Initialise sysctl_user_reserve_kbytes.
1857 *
1858 * This is intended to prevent a user from starting a single memory hogging
1859 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1860 * mode.
1861 *
1862 * The default value is min(3% of free memory, 128MB)
1863 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1864 */
1865static int __meminit init_user_reserve(void)
1866{
1867	unsigned long free_kbytes;
1868
1869	free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
1870
1871	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1872	return 0;
1873}
1874subsys_initcall(init_user_reserve);
1875
1876/*
1877 * Initialise sysctl_admin_reserve_kbytes.
1878 *
1879 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1880 * to log in and kill a memory hogging process.
1881 *
1882 * Systems with more than 256MB will reserve 8MB, enough to recover
1883 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1884 * only reserve 3% of free pages by default.
1885 */
1886static int __meminit init_admin_reserve(void)
1887{
1888	unsigned long free_kbytes;
1889
1890	free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
1891
1892	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1893	return 0;
1894}
1895subsys_initcall(init_admin_reserve);
1896
1897int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
1898{
1899	mmap_write_lock(oldmm);
1900	dup_mm_exe_file(mm, oldmm);
1901	mmap_write_unlock(oldmm);
1902	return 0;
1903}