maccess: rename strnlen_unsafe_user to strnlen_user_nofault
[linux-2.6-block.git] / mm / nommu.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
1da177e4
LT
2/*
3 * linux/mm/nommu.c
4 *
5 * Replacement code for mm functions to support CPU's that don't
6 * have any form of memory management unit (thus no virtual memory).
7 *
8 * See Documentation/nommu-mmap.txt
9 *
8feae131 10 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
1da177e4
LT
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>
29c185e5 14 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
1da177e4
LT
15 */
16
b1de0d13
MH
17#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18
b95f1b31 19#include <linux/export.h>
1da177e4 20#include <linux/mm.h>
6e84f315 21#include <linux/sched/mm.h>
615d6e87 22#include <linux/vmacache.h>
1da177e4
LT
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>
1da177e4
LT
30#include <linux/blkdev.h>
31#include <linux/backing-dev.h>
3b32123d 32#include <linux/compiler.h>
1da177e4
LT
33#include <linux/mount.h>
34#include <linux/personality.h>
35#include <linux/security.h>
36#include <linux/syscalls.h>
120a795d 37#include <linux/audit.h>
b1de0d13 38#include <linux/printk.h>
1da177e4 39
7c0f6ba6 40#include <linux/uaccess.h>
1da177e4
LT
41#include <asm/tlb.h>
42#include <asm/tlbflush.h>
eb8cdec4 43#include <asm/mmu_context.h>
8feae131
DH
44#include "internal.h"
45
1da177e4 46void *high_memory;
944b6874 47EXPORT_SYMBOL(high_memory);
1da177e4
LT
48struct page *mem_map;
49unsigned long max_mapnr;
5b8bf307 50EXPORT_SYMBOL(max_mapnr);
4266c97a 51unsigned long highest_memmap_pfn;
fc4d5c29 52int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
1da177e4
LT
53int heap_stack_gap = 0;
54
33e5d769 55atomic_long_t mmap_pages_allocated;
8feae131 56
1da177e4 57EXPORT_SYMBOL(mem_map);
1da177e4 58
8feae131
DH
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);
1da177e4 63
f0f37e2f 64const struct vm_operations_struct generic_file_vm_ops = {
1da177e4
LT
65};
66
1da177e4
LT
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
4016a139
MH
77 /*
78 * If the object we have should not have ksize performed on it,
79 * return size of 0
80 */
5a1603be 81 if (!objp || !virt_addr_valid(objp))
6cfd53fc
PM
82 return 0;
83
84 page = virt_to_head_page(objp);
6cfd53fc
PM
85
86 /*
87 * If the allocator sets PageSlab, we know the pointer came from
88 * kmalloc().
89 */
1da177e4
LT
90 if (PageSlab(page))
91 return ksize(objp);
92
ab2e83ea
PM
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
6cfd53fc
PM
107 /*
108 * The ksize() function is only guaranteed to work for pointers
5a1603be 109 * returned by kmalloc(). So handle arbitrary pointers here.
6cfd53fc 110 */
a50b854e 111 return page_size(page);
1da177e4
LT
112}
113
dfc2f91a
PM
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
f1c4069e 135LIST_HEAD(vmap_area_list);
1da177e4 136
b3bdda02 137void vfree(const void *addr)
1da177e4
LT
138{
139 kfree(addr);
140}
b5073173 141EXPORT_SYMBOL(vfree);
1da177e4 142
88dca4ca 143void *__vmalloc(unsigned long size, gfp_t gfp_mask)
1da177e4
LT
144{
145 /*
8518609d
RD
146 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
147 * returns only a logical address.
1da177e4 148 */
84097518 149 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
1da177e4 150}
b5073173 151EXPORT_SYMBOL(__vmalloc);
1da177e4 152
041de93f
CH
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
2b905948
CH
161void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
162 int node, const void *caller)
a7c3e901 163{
2b905948 164 return __vmalloc(size, gfp_mask);
a7c3e901
MH
165}
166
ed81745a 167static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
f905bc44
PM
168{
169 void *ret;
170
88dca4ca 171 ret = __vmalloc(size, flags);
f905bc44
PM
172 if (ret) {
173 struct vm_area_struct *vma;
174
d8ed45c5 175 mmap_write_lock(current->mm);
f905bc44
PM
176 vma = find_vma(current->mm, (unsigned long)ret);
177 if (vma)
178 vma->vm_flags |= VM_USERMAP;
d8ed45c5 179 mmap_write_unlock(current->mm);
f905bc44
PM
180 }
181
182 return ret;
183}
ed81745a
AN
184
185void *vmalloc_user(unsigned long size)
186{
187 return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
188}
f905bc44
PM
189EXPORT_SYMBOL(vmalloc_user);
190
b3bdda02 191struct page *vmalloc_to_page(const void *addr)
1da177e4
LT
192{
193 return virt_to_page(addr);
194}
b5073173 195EXPORT_SYMBOL(vmalloc_to_page);
1da177e4 196
b3bdda02 197unsigned long vmalloc_to_pfn(const void *addr)
1da177e4
LT
198{
199 return page_to_pfn(virt_to_page(addr));
200}
b5073173 201EXPORT_SYMBOL(vmalloc_to_pfn);
1da177e4
LT
202
203long vread(char *buf, char *addr, unsigned long count)
204{
9bde916b
CG
205 /* Don't allow overflow */
206 if ((unsigned long) buf + count < count)
207 count = -(unsigned long) buf;
208
1da177e4
LT
209 memcpy(buf, addr, count);
210 return count;
211}
212
213long vwrite(char *buf, char *addr, unsigned long count)
214{
215 /* Don't allow overflow */
216 if ((unsigned long) addr + count < count)
217 count = -(unsigned long) addr;
218
219 memcpy(addr, buf, count);
ac714904 220 return count;
1da177e4
LT
221}
222
223/*
e1c05067 224 * vmalloc - allocate virtually contiguous memory
1da177e4
LT
225 *
226 * @size: allocation size
227 *
228 * Allocate enough pages to cover @size from the page level
e1c05067 229 * allocator and map them into contiguous kernel virtual space.
1da177e4 230 *
c1c8897f 231 * For tight control over page level allocator and protection flags
1da177e4
LT
232 * use __vmalloc() instead.
233 */
234void *vmalloc(unsigned long size)
235{
88dca4ca 236 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM);
1da177e4 237}
f6138882
AM
238EXPORT_SYMBOL(vmalloc);
239
e1ca7788 240/*
e1c05067 241 * vzalloc - allocate virtually contiguous memory with zero fill
e1ca7788
DY
242 *
243 * @size: allocation size
244 *
245 * Allocate enough pages to cover @size from the page level
e1c05067 246 * allocator and map them into contiguous kernel virtual space.
e1ca7788
DY
247 * The memory allocated is set to zero.
248 *
249 * For tight control over page level allocator and protection flags
250 * use __vmalloc() instead.
251 */
252void *vzalloc(unsigned long size)
253{
88dca4ca 254 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
e1ca7788
DY
255}
256EXPORT_SYMBOL(vzalloc);
257
258/**
259 * vmalloc_node - allocate memory on a specific node
260 * @size: allocation size
261 * @node: numa node
262 *
263 * Allocate enough pages to cover @size from the page level
264 * allocator and map them into contiguous kernel virtual space.
265 *
266 * For tight control over page level allocator and protection flags
267 * use __vmalloc() instead.
268 */
f6138882
AM
269void *vmalloc_node(unsigned long size, int node)
270{
271 return vmalloc(size);
272}
9a14f653 273EXPORT_SYMBOL(vmalloc_node);
e1ca7788
DY
274
275/**
276 * vzalloc_node - allocate memory on a specific node with zero fill
277 * @size: allocation size
278 * @node: numa node
279 *
280 * Allocate enough pages to cover @size from the page level
281 * allocator and map them into contiguous kernel virtual space.
282 * The memory allocated is set to zero.
283 *
284 * For tight control over page level allocator and protection flags
285 * use __vmalloc() instead.
286 */
287void *vzalloc_node(unsigned long size, int node)
288{
289 return vzalloc(size);
290}
291EXPORT_SYMBOL(vzalloc_node);
1da177e4 292
1af446ed
PM
293/**
294 * vmalloc_exec - allocate virtually contiguous, executable memory
295 * @size: allocation size
296 *
297 * Kernel-internal function to allocate enough pages to cover @size
298 * the page level allocator and map them into contiguous and
299 * executable kernel virtual space.
300 *
301 * For tight control over page level allocator and protection flags
302 * use __vmalloc() instead.
303 */
304
305void *vmalloc_exec(unsigned long size)
306{
88dca4ca 307 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM);
1af446ed
PM
308}
309
b5073173
PM
310/**
311 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1da177e4
LT
312 * @size: allocation size
313 *
314 * Allocate enough 32bit PA addressable pages to cover @size from the
e1c05067 315 * page level allocator and map them into contiguous kernel virtual space.
1da177e4
LT
316 */
317void *vmalloc_32(unsigned long size)
318{
88dca4ca 319 return __vmalloc(size, GFP_KERNEL);
1da177e4 320}
b5073173
PM
321EXPORT_SYMBOL(vmalloc_32);
322
323/**
324 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
325 * @size: allocation size
326 *
327 * The resulting memory area is 32bit addressable and zeroed so it can be
328 * mapped to userspace without leaking data.
f905bc44
PM
329 *
330 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
331 * remap_vmalloc_range() are permissible.
b5073173
PM
332 */
333void *vmalloc_32_user(unsigned long size)
334{
f905bc44
PM
335 /*
336 * We'll have to sort out the ZONE_DMA bits for 64-bit,
337 * but for now this can simply use vmalloc_user() directly.
338 */
339 return vmalloc_user(size);
b5073173
PM
340}
341EXPORT_SYMBOL(vmalloc_32_user);
1da177e4
LT
342
343void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
344{
345 BUG();
346 return NULL;
347}
b5073173 348EXPORT_SYMBOL(vmap);
1da177e4 349
b3bdda02 350void vunmap(const void *addr)
1da177e4
LT
351{
352 BUG();
353}
b5073173 354EXPORT_SYMBOL(vunmap);
1da177e4 355
d4efd79a 356void *vm_map_ram(struct page **pages, unsigned int count, int node)
eb6434d9
PM
357{
358 BUG();
359 return NULL;
360}
361EXPORT_SYMBOL(vm_map_ram);
362
363void vm_unmap_ram(const void *mem, unsigned int count)
364{
365 BUG();
366}
367EXPORT_SYMBOL(vm_unmap_ram);
368
369void vm_unmap_aliases(void)
370{
371}
372EXPORT_SYMBOL_GPL(vm_unmap_aliases);
373
cd12909c 374struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
29c185e5
PM
375{
376 BUG();
377 return NULL;
378}
379EXPORT_SYMBOL_GPL(alloc_vm_area);
380
381void free_vm_area(struct vm_struct *area)
382{
383 BUG();
384}
385EXPORT_SYMBOL_GPL(free_vm_area);
386
b5073173
PM
387int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
388 struct page *page)
389{
390 return -EINVAL;
391}
392EXPORT_SYMBOL(vm_insert_page);
393
a667d745
SJ
394int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
395 unsigned long num)
396{
397 return -EINVAL;
398}
399EXPORT_SYMBOL(vm_map_pages);
400
401int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
402 unsigned long num)
403{
404 return -EINVAL;
405}
406EXPORT_SYMBOL(vm_map_pages_zero);
407
1da177e4
LT
408/*
409 * sys_brk() for the most part doesn't need the global kernel
410 * lock, except when an application is doing something nasty
411 * like trying to un-brk an area that has already been mapped
412 * to a regular file. in this case, the unmapping will need
413 * to invoke file system routines that need the global lock.
414 */
6a6160a7 415SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4
LT
416{
417 struct mm_struct *mm = current->mm;
418
419 if (brk < mm->start_brk || brk > mm->context.end_brk)
420 return mm->brk;
421
422 if (mm->brk == brk)
423 return mm->brk;
424
425 /*
426 * Always allow shrinking brk
427 */
428 if (brk <= mm->brk) {
429 mm->brk = brk;
430 return brk;
431 }
432
433 /*
434 * Ok, looks good - let it rip.
435 */
a75a2df6 436 flush_icache_user_range(mm->brk, brk);
1da177e4
LT
437 return mm->brk = brk;
438}
439
8feae131 440/*
3edf41d8 441 * initialise the percpu counter for VM and region record slabs
8feae131
DH
442 */
443void __init mmap_init(void)
1da177e4 444{
00a62ce9
KM
445 int ret;
446
908c7f19 447 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
00a62ce9 448 VM_BUG_ON(ret);
5d097056 449 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
1da177e4 450}
1da177e4 451
3034097a 452/*
8feae131
DH
453 * validate the region tree
454 * - the caller must hold the region lock
3034097a 455 */
8feae131
DH
456#ifdef CONFIG_DEBUG_NOMMU_REGIONS
457static noinline void validate_nommu_regions(void)
3034097a 458{
8feae131
DH
459 struct vm_region *region, *last;
460 struct rb_node *p, *lastp;
3034097a 461
8feae131
DH
462 lastp = rb_first(&nommu_region_tree);
463 if (!lastp)
464 return;
465
466 last = rb_entry(lastp, struct vm_region, vm_rb);
c9427bc0
GT
467 BUG_ON(last->vm_end <= last->vm_start);
468 BUG_ON(last->vm_top < last->vm_end);
8feae131
DH
469
470 while ((p = rb_next(lastp))) {
471 region = rb_entry(p, struct vm_region, vm_rb);
472 last = rb_entry(lastp, struct vm_region, vm_rb);
473
c9427bc0
GT
474 BUG_ON(region->vm_end <= region->vm_start);
475 BUG_ON(region->vm_top < region->vm_end);
476 BUG_ON(region->vm_start < last->vm_top);
3034097a 477
8feae131
DH
478 lastp = p;
479 }
3034097a 480}
8feae131 481#else
33e5d769
DH
482static void validate_nommu_regions(void)
483{
484}
8feae131 485#endif
3034097a
DH
486
487/*
8feae131 488 * add a region into the global tree
3034097a 489 */
8feae131 490static void add_nommu_region(struct vm_region *region)
3034097a 491{
8feae131
DH
492 struct vm_region *pregion;
493 struct rb_node **p, *parent;
3034097a 494
8feae131
DH
495 validate_nommu_regions();
496
8feae131
DH
497 parent = NULL;
498 p = &nommu_region_tree.rb_node;
499 while (*p) {
500 parent = *p;
501 pregion = rb_entry(parent, struct vm_region, vm_rb);
502 if (region->vm_start < pregion->vm_start)
503 p = &(*p)->rb_left;
504 else if (region->vm_start > pregion->vm_start)
505 p = &(*p)->rb_right;
506 else if (pregion == region)
507 return;
508 else
509 BUG();
3034097a
DH
510 }
511
8feae131
DH
512 rb_link_node(&region->vm_rb, parent, p);
513 rb_insert_color(&region->vm_rb, &nommu_region_tree);
3034097a 514
8feae131 515 validate_nommu_regions();
3034097a 516}
3034097a 517
930e652a 518/*
8feae131 519 * delete a region from the global tree
930e652a 520 */
8feae131 521static void delete_nommu_region(struct vm_region *region)
930e652a 522{
8feae131 523 BUG_ON(!nommu_region_tree.rb_node);
930e652a 524
8feae131
DH
525 validate_nommu_regions();
526 rb_erase(&region->vm_rb, &nommu_region_tree);
527 validate_nommu_regions();
57c8f63e
GU
528}
529
6fa5f80b 530/*
8feae131 531 * free a contiguous series of pages
6fa5f80b 532 */
8feae131 533static void free_page_series(unsigned long from, unsigned long to)
6fa5f80b 534{
8feae131
DH
535 for (; from < to; from += PAGE_SIZE) {
536 struct page *page = virt_to_page(from);
537
33e5d769 538 atomic_long_dec(&mmap_pages_allocated);
8feae131 539 put_page(page);
6fa5f80b 540 }
6fa5f80b
DH
541}
542
3034097a 543/*
8feae131 544 * release a reference to a region
33e5d769 545 * - the caller must hold the region semaphore for writing, which this releases
dd8632a1 546 * - the region may not have been added to the tree yet, in which case vm_top
8feae131 547 * will equal vm_start
3034097a 548 */
8feae131
DH
549static void __put_nommu_region(struct vm_region *region)
550 __releases(nommu_region_sem)
1da177e4 551{
8feae131 552 BUG_ON(!nommu_region_tree.rb_node);
1da177e4 553
1e2ae599 554 if (--region->vm_usage == 0) {
dd8632a1 555 if (region->vm_top > region->vm_start)
8feae131
DH
556 delete_nommu_region(region);
557 up_write(&nommu_region_sem);
558
559 if (region->vm_file)
560 fput(region->vm_file);
561
562 /* IO memory and memory shared directly out of the pagecache
563 * from ramfs/tmpfs mustn't be released here */
22cc877b 564 if (region->vm_flags & VM_MAPPED_COPY)
dd8632a1 565 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
566 kmem_cache_free(vm_region_jar, region);
567 } else {
568 up_write(&nommu_region_sem);
1da177e4 569 }
8feae131 570}
1da177e4 571
8feae131
DH
572/*
573 * release a reference to a region
574 */
575static void put_nommu_region(struct vm_region *region)
576{
577 down_write(&nommu_region_sem);
578 __put_nommu_region(region);
1da177e4
LT
579}
580
3034097a 581/*
8feae131
DH
582 * add a VMA into a process's mm_struct in the appropriate place in the list
583 * and tree and add to the address space's page tree also if not an anonymous
584 * page
c1e8d7c6 585 * - should be called with mm->mmap_lock held writelocked
3034097a 586 */
8feae131 587static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 588{
6038def0 589 struct vm_area_struct *pvma, *prev;
1da177e4 590 struct address_space *mapping;
6038def0 591 struct rb_node **p, *parent, *rb_prev;
8feae131 592
8feae131
DH
593 BUG_ON(!vma->vm_region);
594
595 mm->map_count++;
596 vma->vm_mm = mm;
1da177e4
LT
597
598 /* add the VMA to the mapping */
599 if (vma->vm_file) {
600 mapping = vma->vm_file->f_mapping;
601
83cde9e8 602 i_mmap_lock_write(mapping);
1da177e4 603 flush_dcache_mmap_lock(mapping);
6b2dbba8 604 vma_interval_tree_insert(vma, &mapping->i_mmap);
1da177e4 605 flush_dcache_mmap_unlock(mapping);
83cde9e8 606 i_mmap_unlock_write(mapping);
1da177e4
LT
607 }
608
8feae131 609 /* add the VMA to the tree */
6038def0 610 parent = rb_prev = NULL;
8feae131 611 p = &mm->mm_rb.rb_node;
1da177e4
LT
612 while (*p) {
613 parent = *p;
614 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
615
8feae131
DH
616 /* sort by: start addr, end addr, VMA struct addr in that order
617 * (the latter is necessary as we may get identical VMAs) */
618 if (vma->vm_start < pvma->vm_start)
1da177e4 619 p = &(*p)->rb_left;
6038def0
NK
620 else if (vma->vm_start > pvma->vm_start) {
621 rb_prev = parent;
1da177e4 622 p = &(*p)->rb_right;
6038def0 623 } else if (vma->vm_end < pvma->vm_end)
8feae131 624 p = &(*p)->rb_left;
6038def0
NK
625 else if (vma->vm_end > pvma->vm_end) {
626 rb_prev = parent;
8feae131 627 p = &(*p)->rb_right;
6038def0 628 } else if (vma < pvma)
8feae131 629 p = &(*p)->rb_left;
6038def0
NK
630 else if (vma > pvma) {
631 rb_prev = parent;
8feae131 632 p = &(*p)->rb_right;
6038def0 633 } else
8feae131 634 BUG();
1da177e4
LT
635 }
636
637 rb_link_node(&vma->vm_rb, parent, p);
8feae131
DH
638 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
639
640 /* add VMA to the VMA list also */
6038def0
NK
641 prev = NULL;
642 if (rb_prev)
643 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
8feae131 644
aba6dfb7 645 __vma_link_list(mm, vma, prev);
1da177e4
LT
646}
647
3034097a 648/*
8feae131 649 * delete a VMA from its owning mm_struct and address space
3034097a 650 */
8feae131 651static void delete_vma_from_mm(struct vm_area_struct *vma)
1da177e4 652{
615d6e87 653 int i;
1da177e4 654 struct address_space *mapping;
8feae131 655 struct mm_struct *mm = vma->vm_mm;
615d6e87 656 struct task_struct *curr = current;
8feae131 657
8feae131 658 mm->map_count--;
615d6e87
DB
659 for (i = 0; i < VMACACHE_SIZE; i++) {
660 /* if the vma is cached, invalidate the entire cache */
314ff785 661 if (curr->vmacache.vmas[i] == vma) {
e020d5bd 662 vmacache_invalidate(mm);
615d6e87
DB
663 break;
664 }
665 }
1da177e4
LT
666
667 /* remove the VMA from the mapping */
668 if (vma->vm_file) {
669 mapping = vma->vm_file->f_mapping;
670
83cde9e8 671 i_mmap_lock_write(mapping);
1da177e4 672 flush_dcache_mmap_lock(mapping);
6b2dbba8 673 vma_interval_tree_remove(vma, &mapping->i_mmap);
1da177e4 674 flush_dcache_mmap_unlock(mapping);
83cde9e8 675 i_mmap_unlock_write(mapping);
1da177e4
LT
676 }
677
8feae131
DH
678 /* remove from the MM's tree and list */
679 rb_erase(&vma->vm_rb, &mm->mm_rb);
b951bf2c 680
1b9fc5b2 681 __vma_unlink_list(mm, vma);
8feae131
DH
682}
683
684/*
685 * destroy a VMA record
686 */
687static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
688{
8feae131
DH
689 if (vma->vm_ops && vma->vm_ops->close)
690 vma->vm_ops->close(vma);
e9714acf 691 if (vma->vm_file)
8feae131 692 fput(vma->vm_file);
8feae131 693 put_nommu_region(vma->vm_region);
3928d4f5 694 vm_area_free(vma);
8feae131
DH
695}
696
697/*
698 * look up the first VMA in which addr resides, NULL if none
c1e8d7c6 699 * - should be called with mm->mmap_lock at least held readlocked
8feae131
DH
700 */
701struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
702{
703 struct vm_area_struct *vma;
8feae131
DH
704
705 /* check the cache first */
615d6e87
DB
706 vma = vmacache_find(mm, addr);
707 if (likely(vma))
8feae131
DH
708 return vma;
709
e922c4c5 710 /* trawl the list (there may be multiple mappings in which addr
8feae131 711 * resides) */
e922c4c5 712 for (vma = mm->mmap; vma; vma = vma->vm_next) {
8feae131
DH
713 if (vma->vm_start > addr)
714 return NULL;
715 if (vma->vm_end > addr) {
615d6e87 716 vmacache_update(addr, vma);
8feae131
DH
717 return vma;
718 }
719 }
720
721 return NULL;
722}
723EXPORT_SYMBOL(find_vma);
724
725/*
726 * find a VMA
727 * - we don't extend stack VMAs under NOMMU conditions
728 */
729struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
730{
7561e8ca 731 return find_vma(mm, addr);
8feae131
DH
732}
733
734/*
735 * expand a stack to a given address
736 * - not supported under NOMMU conditions
737 */
738int expand_stack(struct vm_area_struct *vma, unsigned long address)
739{
740 return -ENOMEM;
741}
742
743/*
744 * look up the first VMA exactly that exactly matches addr
c1e8d7c6 745 * - should be called with mm->mmap_lock at least held readlocked
8feae131
DH
746 */
747static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
748 unsigned long addr,
749 unsigned long len)
750{
751 struct vm_area_struct *vma;
8feae131
DH
752 unsigned long end = addr + len;
753
754 /* check the cache first */
615d6e87
DB
755 vma = vmacache_find_exact(mm, addr, end);
756 if (vma)
8feae131
DH
757 return vma;
758
e922c4c5 759 /* trawl the list (there may be multiple mappings in which addr
8feae131 760 * resides) */
e922c4c5 761 for (vma = mm->mmap; vma; vma = vma->vm_next) {
8feae131
DH
762 if (vma->vm_start < addr)
763 continue;
764 if (vma->vm_start > addr)
765 return NULL;
766 if (vma->vm_end == end) {
615d6e87 767 vmacache_update(addr, vma);
8feae131
DH
768 return vma;
769 }
770 }
771
772 return NULL;
1da177e4
LT
773}
774
775/*
776 * determine whether a mapping should be permitted and, if so, what sort of
777 * mapping we're capable of supporting
778 */
779static int validate_mmap_request(struct file *file,
780 unsigned long addr,
781 unsigned long len,
782 unsigned long prot,
783 unsigned long flags,
784 unsigned long pgoff,
785 unsigned long *_capabilities)
786{
8feae131 787 unsigned long capabilities, rlen;
1da177e4
LT
788 int ret;
789
790 /* do the simple checks first */
22cc877b 791 if (flags & MAP_FIXED)
1da177e4 792 return -EINVAL;
1da177e4
LT
793
794 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
795 (flags & MAP_TYPE) != MAP_SHARED)
796 return -EINVAL;
797
f81cff0d 798 if (!len)
1da177e4
LT
799 return -EINVAL;
800
f81cff0d 801 /* Careful about overflows.. */
8feae131
DH
802 rlen = PAGE_ALIGN(len);
803 if (!rlen || rlen > TASK_SIZE)
f81cff0d
MF
804 return -ENOMEM;
805
1da177e4 806 /* offset overflow? */
8feae131 807 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
f81cff0d 808 return -EOVERFLOW;
1da177e4
LT
809
810 if (file) {
1da177e4 811 /* files must support mmap */
72c2d531 812 if (!file->f_op->mmap)
1da177e4
LT
813 return -ENODEV;
814
815 /* work out if what we've got could possibly be shared
816 * - we support chardevs that provide their own "memory"
817 * - we support files/blockdevs that are memory backed
818 */
b4caecd4
CH
819 if (file->f_op->mmap_capabilities) {
820 capabilities = file->f_op->mmap_capabilities(file);
821 } else {
1da177e4
LT
822 /* no explicit capabilities set, so assume some
823 * defaults */
496ad9aa 824 switch (file_inode(file)->i_mode & S_IFMT) {
1da177e4
LT
825 case S_IFREG:
826 case S_IFBLK:
b4caecd4 827 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
828 break;
829
830 case S_IFCHR:
831 capabilities =
b4caecd4
CH
832 NOMMU_MAP_DIRECT |
833 NOMMU_MAP_READ |
834 NOMMU_MAP_WRITE;
1da177e4
LT
835 break;
836
837 default:
838 return -EINVAL;
839 }
840 }
841
842 /* eliminate any capabilities that we can't support on this
843 * device */
844 if (!file->f_op->get_unmapped_area)
b4caecd4 845 capabilities &= ~NOMMU_MAP_DIRECT;
6e242a1c 846 if (!(file->f_mode & FMODE_CAN_READ))
b4caecd4 847 capabilities &= ~NOMMU_MAP_COPY;
1da177e4 848
28d7a6ae
GY
849 /* The file shall have been opened with read permission. */
850 if (!(file->f_mode & FMODE_READ))
851 return -EACCES;
852
1da177e4
LT
853 if (flags & MAP_SHARED) {
854 /* do checks for writing, appending and locking */
855 if ((prot & PROT_WRITE) &&
856 !(file->f_mode & FMODE_WRITE))
857 return -EACCES;
858
496ad9aa 859 if (IS_APPEND(file_inode(file)) &&
1da177e4
LT
860 (file->f_mode & FMODE_WRITE))
861 return -EACCES;
862
d7a06983 863 if (locks_verify_locked(file))
1da177e4
LT
864 return -EAGAIN;
865
b4caecd4 866 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
867 return -ENODEV;
868
1da177e4 869 /* we mustn't privatise shared mappings */
b4caecd4 870 capabilities &= ~NOMMU_MAP_COPY;
ac714904 871 } else {
1da177e4
LT
872 /* we're going to read the file into private memory we
873 * allocate */
b4caecd4 874 if (!(capabilities & NOMMU_MAP_COPY))
1da177e4
LT
875 return -ENODEV;
876
877 /* we don't permit a private writable mapping to be
878 * shared with the backing device */
879 if (prot & PROT_WRITE)
b4caecd4 880 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4
LT
881 }
882
b4caecd4
CH
883 if (capabilities & NOMMU_MAP_DIRECT) {
884 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
885 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
886 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
3c7b2045 887 ) {
b4caecd4 888 capabilities &= ~NOMMU_MAP_DIRECT;
3c7b2045 889 if (flags & MAP_SHARED) {
22cc877b 890 pr_warn("MAP_SHARED not completely supported on !MMU\n");
3c7b2045
BS
891 return -EINVAL;
892 }
893 }
894 }
895
1da177e4
LT
896 /* handle executable mappings and implied executable
897 * mappings */
90f8572b 898 if (path_noexec(&file->f_path)) {
1da177e4
LT
899 if (prot & PROT_EXEC)
900 return -EPERM;
ac714904 901 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1da177e4
LT
902 /* handle implication of PROT_EXEC by PROT_READ */
903 if (current->personality & READ_IMPLIES_EXEC) {
b4caecd4 904 if (capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
905 prot |= PROT_EXEC;
906 }
ac714904 907 } else if ((prot & PROT_READ) &&
1da177e4 908 (prot & PROT_EXEC) &&
b4caecd4 909 !(capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
910 ) {
911 /* backing file is not executable, try to copy */
b4caecd4 912 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4 913 }
ac714904 914 } else {
1da177e4
LT
915 /* anonymous mappings are always memory backed and can be
916 * privately mapped
917 */
b4caecd4 918 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
919
920 /* handle PROT_EXEC implication by PROT_READ */
921 if ((prot & PROT_READ) &&
922 (current->personality & READ_IMPLIES_EXEC))
923 prot |= PROT_EXEC;
924 }
925
926 /* allow the security API to have its say */
e5467859 927 ret = security_mmap_addr(addr);
1da177e4
LT
928 if (ret < 0)
929 return ret;
930
931 /* looks okay */
932 *_capabilities = capabilities;
933 return 0;
934}
935
936/*
937 * we've determined that we can make the mapping, now translate what we
938 * now know into VMA flags
939 */
940static unsigned long determine_vm_flags(struct file *file,
941 unsigned long prot,
942 unsigned long flags,
943 unsigned long capabilities)
944{
945 unsigned long vm_flags;
946
e6bfb709 947 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1da177e4
LT
948 /* vm_flags |= mm->def_flags; */
949
b4caecd4 950 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1da177e4 951 /* attempt to share read-only copies of mapped file chunks */
3c7b2045 952 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1da177e4
LT
953 if (file && !(prot & PROT_WRITE))
954 vm_flags |= VM_MAYSHARE;
3c7b2045 955 } else {
1da177e4
LT
956 /* overlay a shareable mapping on the backing device or inode
957 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
958 * romfs/cramfs */
b4caecd4 959 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1da177e4 960 if (flags & MAP_SHARED)
3c7b2045 961 vm_flags |= VM_SHARED;
1da177e4
LT
962 }
963
964 /* refuse to let anyone share private mappings with this process if
965 * it's being traced - otherwise breakpoints set in it may interfere
966 * with another untraced process
967 */
a288eecc 968 if ((flags & MAP_PRIVATE) && current->ptrace)
1da177e4
LT
969 vm_flags &= ~VM_MAYSHARE;
970
971 return vm_flags;
972}
973
974/*
8feae131
DH
975 * set up a shared mapping on a file (the driver or filesystem provides and
976 * pins the storage)
1da177e4 977 */
8feae131 978static int do_mmap_shared_file(struct vm_area_struct *vma)
1da177e4
LT
979{
980 int ret;
981
f74ac015 982 ret = call_mmap(vma->vm_file, vma);
dd8632a1
PM
983 if (ret == 0) {
984 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 985 return 0;
dd8632a1 986 }
1da177e4
LT
987 if (ret != -ENOSYS)
988 return ret;
989
3fa30460
DH
990 /* getting -ENOSYS indicates that direct mmap isn't possible (as
991 * opposed to tried but failed) so we can only give a suitable error as
992 * it's not possible to make a private copy if MAP_SHARED was given */
1da177e4
LT
993 return -ENODEV;
994}
995
996/*
997 * set up a private mapping or an anonymous shared mapping
998 */
8feae131
DH
999static int do_mmap_private(struct vm_area_struct *vma,
1000 struct vm_region *region,
645d83c5
DH
1001 unsigned long len,
1002 unsigned long capabilities)
1da177e4 1003{
dbc8358c 1004 unsigned long total, point;
1da177e4 1005 void *base;
8feae131 1006 int ret, order;
1da177e4
LT
1007
1008 /* invoke the file's mapping function so that it can keep track of
1009 * shared mappings on devices or memory
1010 * - VM_MAYSHARE will be set if it may attempt to share
1011 */
b4caecd4 1012 if (capabilities & NOMMU_MAP_DIRECT) {
f74ac015 1013 ret = call_mmap(vma->vm_file, vma);
dd8632a1 1014 if (ret == 0) {
1da177e4 1015 /* shouldn't return success if we're not sharing */
dd8632a1
PM
1016 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1017 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 1018 return 0;
1da177e4 1019 }
dd8632a1
PM
1020 if (ret != -ENOSYS)
1021 return ret;
1da177e4
LT
1022
1023 /* getting an ENOSYS error indicates that direct mmap isn't
1024 * possible (as opposed to tried but failed) so we'll try to
1025 * make a private copy of the data and map that instead */
1026 }
1027
8feae131 1028
1da177e4
LT
1029 /* allocate some memory to hold the mapping
1030 * - note that this may not return a page-aligned address if the object
1031 * we're allocating is smaller than a page
1032 */
f67d9b15 1033 order = get_order(len);
8feae131 1034 total = 1 << order;
f67d9b15 1035 point = len >> PAGE_SHIFT;
dd8632a1 1036
dbc8358c 1037 /* we don't want to allocate a power-of-2 sized page set */
22cc877b 1038 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
dbc8358c 1039 total = point;
8feae131 1040
da616534 1041 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
dbc8358c
JK
1042 if (!base)
1043 goto enomem;
1044
1045 atomic_long_add(total, &mmap_pages_allocated);
1da177e4 1046
8feae131
DH
1047 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1048 region->vm_start = (unsigned long) base;
f67d9b15 1049 region->vm_end = region->vm_start + len;
dd8632a1 1050 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
8feae131
DH
1051
1052 vma->vm_start = region->vm_start;
1053 vma->vm_end = region->vm_start + len;
1da177e4
LT
1054
1055 if (vma->vm_file) {
1056 /* read the contents of a file into the copy */
1da177e4
LT
1057 loff_t fpos;
1058
1059 fpos = vma->vm_pgoff;
1060 fpos <<= PAGE_SHIFT;
1061
b4bf802a 1062 ret = kernel_read(vma->vm_file, base, len, &fpos);
1da177e4
LT
1063 if (ret < 0)
1064 goto error_free;
1065
1066 /* clear the last little bit */
f67d9b15
BL
1067 if (ret < len)
1068 memset(base + ret, 0, len - ret);
1da177e4 1069
bfd40eaf
KS
1070 } else {
1071 vma_set_anonymous(vma);
1da177e4
LT
1072 }
1073
1074 return 0;
1075
1076error_free:
7223bb4a 1077 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
1078 region->vm_start = vma->vm_start = 0;
1079 region->vm_end = vma->vm_end = 0;
dd8632a1 1080 region->vm_top = 0;
1da177e4
LT
1081 return ret;
1082
1083enomem:
b1de0d13 1084 pr_err("Allocation of length %lu from process %d (%s) failed\n",
05ae6fa3 1085 len, current->pid, current->comm);
9af744d7 1086 show_free_areas(0, NULL);
1da177e4
LT
1087 return -ENOMEM;
1088}
1089
1090/*
1091 * handle mapping creation for uClinux
1092 */
1fcfd8db
ON
1093unsigned long do_mmap(struct file *file,
1094 unsigned long addr,
1095 unsigned long len,
1096 unsigned long prot,
1097 unsigned long flags,
1098 vm_flags_t vm_flags,
1099 unsigned long pgoff,
897ab3e0
MR
1100 unsigned long *populate,
1101 struct list_head *uf)
1da177e4 1102{
8feae131
DH
1103 struct vm_area_struct *vma;
1104 struct vm_region *region;
1da177e4 1105 struct rb_node *rb;
1fcfd8db 1106 unsigned long capabilities, result;
1da177e4
LT
1107 int ret;
1108
41badc15 1109 *populate = 0;
bebeb3d6 1110
1da177e4
LT
1111 /* decide whether we should attempt the mapping, and if so what sort of
1112 * mapping */
1113 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1114 &capabilities);
22cc877b 1115 if (ret < 0)
1da177e4
LT
1116 return ret;
1117
06aab5a3
DH
1118 /* we ignore the address hint */
1119 addr = 0;
f67d9b15 1120 len = PAGE_ALIGN(len);
06aab5a3 1121
1da177e4
LT
1122 /* we've determined that we can make the mapping, now translate what we
1123 * now know into VMA flags */
1fcfd8db 1124 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1da177e4 1125
8feae131
DH
1126 /* we're going to need to record the mapping */
1127 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1128 if (!region)
1129 goto error_getting_region;
1130
490fc053 1131 vma = vm_area_alloc(current->mm);
8feae131
DH
1132 if (!vma)
1133 goto error_getting_vma;
1da177e4 1134
1e2ae599 1135 region->vm_usage = 1;
8feae131
DH
1136 region->vm_flags = vm_flags;
1137 region->vm_pgoff = pgoff;
1138
8feae131
DH
1139 vma->vm_flags = vm_flags;
1140 vma->vm_pgoff = pgoff;
1da177e4 1141
8feae131 1142 if (file) {
cb0942b8
AV
1143 region->vm_file = get_file(file);
1144 vma->vm_file = get_file(file);
8feae131
DH
1145 }
1146
1147 down_write(&nommu_region_sem);
1148
1149 /* if we want to share, we need to check for regions created by other
1da177e4 1150 * mmap() calls that overlap with our proposed mapping
8feae131 1151 * - we can only share with a superset match on most regular files
1da177e4
LT
1152 * - shared mappings on character devices and memory backed files are
1153 * permitted to overlap inexactly as far as we are concerned for in
1154 * these cases, sharing is handled in the driver or filesystem rather
1155 * than here
1156 */
1157 if (vm_flags & VM_MAYSHARE) {
8feae131
DH
1158 struct vm_region *pregion;
1159 unsigned long pglen, rpglen, pgend, rpgend, start;
1da177e4 1160
8feae131
DH
1161 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1162 pgend = pgoff + pglen;
165b2392 1163
8feae131
DH
1164 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1165 pregion = rb_entry(rb, struct vm_region, vm_rb);
1da177e4 1166
8feae131 1167 if (!(pregion->vm_flags & VM_MAYSHARE))
1da177e4
LT
1168 continue;
1169
1170 /* search for overlapping mappings on the same file */
496ad9aa
AV
1171 if (file_inode(pregion->vm_file) !=
1172 file_inode(file))
1da177e4
LT
1173 continue;
1174
8feae131 1175 if (pregion->vm_pgoff >= pgend)
1da177e4
LT
1176 continue;
1177
8feae131
DH
1178 rpglen = pregion->vm_end - pregion->vm_start;
1179 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1180 rpgend = pregion->vm_pgoff + rpglen;
1181 if (pgoff >= rpgend)
1da177e4
LT
1182 continue;
1183
8feae131
DH
1184 /* handle inexactly overlapping matches between
1185 * mappings */
1186 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1187 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1188 /* new mapping is not a subset of the region */
b4caecd4 1189 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
1190 goto sharing_violation;
1191 continue;
1192 }
1193
8feae131 1194 /* we've found a region we can share */
1e2ae599 1195 pregion->vm_usage++;
8feae131
DH
1196 vma->vm_region = pregion;
1197 start = pregion->vm_start;
1198 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1199 vma->vm_start = start;
1200 vma->vm_end = start + len;
1201
22cc877b 1202 if (pregion->vm_flags & VM_MAPPED_COPY)
8feae131 1203 vma->vm_flags |= VM_MAPPED_COPY;
22cc877b 1204 else {
8feae131
DH
1205 ret = do_mmap_shared_file(vma);
1206 if (ret < 0) {
1207 vma->vm_region = NULL;
1208 vma->vm_start = 0;
1209 vma->vm_end = 0;
1e2ae599 1210 pregion->vm_usage--;
8feae131
DH
1211 pregion = NULL;
1212 goto error_just_free;
1213 }
1214 }
1215 fput(region->vm_file);
1216 kmem_cache_free(vm_region_jar, region);
1217 region = pregion;
1218 result = start;
1219 goto share;
1da177e4
LT
1220 }
1221
1da177e4
LT
1222 /* obtain the address at which to make a shared mapping
1223 * - this is the hook for quasi-memory character devices to
1224 * tell us the location of a shared mapping
1225 */
b4caecd4 1226 if (capabilities & NOMMU_MAP_DIRECT) {
1da177e4
LT
1227 addr = file->f_op->get_unmapped_area(file, addr, len,
1228 pgoff, flags);
bb005a59 1229 if (IS_ERR_VALUE(addr)) {
1da177e4 1230 ret = addr;
bb005a59 1231 if (ret != -ENOSYS)
8feae131 1232 goto error_just_free;
1da177e4
LT
1233
1234 /* the driver refused to tell us where to site
1235 * the mapping so we'll have to attempt to copy
1236 * it */
bb005a59 1237 ret = -ENODEV;
b4caecd4 1238 if (!(capabilities & NOMMU_MAP_COPY))
8feae131 1239 goto error_just_free;
1da177e4 1240
b4caecd4 1241 capabilities &= ~NOMMU_MAP_DIRECT;
8feae131
DH
1242 } else {
1243 vma->vm_start = region->vm_start = addr;
1244 vma->vm_end = region->vm_end = addr + len;
1da177e4
LT
1245 }
1246 }
1247 }
1248
8feae131 1249 vma->vm_region = region;
1da177e4 1250
645d83c5 1251 /* set up the mapping
b4caecd4 1252 * - the region is filled in if NOMMU_MAP_DIRECT is still set
645d83c5 1253 */
1da177e4 1254 if (file && vma->vm_flags & VM_SHARED)
8feae131 1255 ret = do_mmap_shared_file(vma);
1da177e4 1256 else
645d83c5 1257 ret = do_mmap_private(vma, region, len, capabilities);
1da177e4 1258 if (ret < 0)
645d83c5
DH
1259 goto error_just_free;
1260 add_nommu_region(region);
8feae131 1261
ea637639 1262 /* clear anonymous mappings that don't ask for uninitialized data */
0bf5f949
CH
1263 if (!vma->vm_file &&
1264 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1265 !(flags & MAP_UNINITIALIZED)))
ea637639
JZ
1266 memset((void *)region->vm_start, 0,
1267 region->vm_end - region->vm_start);
1268
1da177e4 1269 /* okay... we have a mapping; now we have to register it */
8feae131 1270 result = vma->vm_start;
1da177e4 1271
1da177e4
LT
1272 current->mm->total_vm += len >> PAGE_SHIFT;
1273
8feae131
DH
1274share:
1275 add_vma_to_mm(current->mm, vma);
1da177e4 1276
cfe79c00
MF
1277 /* we flush the region from the icache only when the first executable
1278 * mapping of it is made */
1279 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
a75a2df6 1280 flush_icache_user_range(region->vm_start, region->vm_end);
cfe79c00
MF
1281 region->vm_icache_flushed = true;
1282 }
1da177e4 1283
cfe79c00 1284 up_write(&nommu_region_sem);
1da177e4 1285
8feae131 1286 return result;
1da177e4 1287
8feae131
DH
1288error_just_free:
1289 up_write(&nommu_region_sem);
1290error:
89a86402
DH
1291 if (region->vm_file)
1292 fput(region->vm_file);
8feae131 1293 kmem_cache_free(vm_region_jar, region);
89a86402
DH
1294 if (vma->vm_file)
1295 fput(vma->vm_file);
3928d4f5 1296 vm_area_free(vma);
8feae131
DH
1297 return ret;
1298
1299sharing_violation:
1300 up_write(&nommu_region_sem);
22cc877b 1301 pr_warn("Attempt to share mismatched mappings\n");
8feae131
DH
1302 ret = -EINVAL;
1303 goto error;
1da177e4 1304
8feae131
DH
1305error_getting_vma:
1306 kmem_cache_free(vm_region_jar, region);
22cc877b
LR
1307 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1308 len, current->pid);
9af744d7 1309 show_free_areas(0, NULL);
1da177e4
LT
1310 return -ENOMEM;
1311
8feae131 1312error_getting_region:
22cc877b
LR
1313 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1314 len, current->pid);
9af744d7 1315 show_free_areas(0, NULL);
1da177e4
LT
1316 return -ENOMEM;
1317}
6be5ceb0 1318
a90f590a
DB
1319unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1320 unsigned long prot, unsigned long flags,
1321 unsigned long fd, unsigned long pgoff)
66f0dc48
HD
1322{
1323 struct file *file = NULL;
1324 unsigned long retval = -EBADF;
1325
120a795d 1326 audit_mmap_fd(fd, flags);
66f0dc48
HD
1327 if (!(flags & MAP_ANONYMOUS)) {
1328 file = fget(fd);
1329 if (!file)
1330 goto out;
1331 }
1332
1333 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1334
ad1ed293 1335 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
66f0dc48
HD
1336
1337 if (file)
1338 fput(file);
1339out:
1340 return retval;
1341}
1342
a90f590a
DB
1343SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1344 unsigned long, prot, unsigned long, flags,
1345 unsigned long, fd, unsigned long, pgoff)
1346{
1347 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1348}
1349
a4679373
CH
1350#ifdef __ARCH_WANT_SYS_OLD_MMAP
1351struct mmap_arg_struct {
1352 unsigned long addr;
1353 unsigned long len;
1354 unsigned long prot;
1355 unsigned long flags;
1356 unsigned long fd;
1357 unsigned long offset;
1358};
1359
1360SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1361{
1362 struct mmap_arg_struct a;
1363
1364 if (copy_from_user(&a, arg, sizeof(a)))
1365 return -EFAULT;
1824cb75 1366 if (offset_in_page(a.offset))
a4679373
CH
1367 return -EINVAL;
1368
a90f590a
DB
1369 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1370 a.offset >> PAGE_SHIFT);
a4679373
CH
1371}
1372#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1373
1da177e4 1374/*
8feae131
DH
1375 * split a vma into two pieces at address 'addr', a new vma is allocated either
1376 * for the first part or the tail.
1da177e4 1377 */
8feae131
DH
1378int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1379 unsigned long addr, int new_below)
1da177e4 1380{
8feae131
DH
1381 struct vm_area_struct *new;
1382 struct vm_region *region;
1383 unsigned long npages;
1da177e4 1384
779c1023
DH
1385 /* we're only permitted to split anonymous regions (these should have
1386 * only a single usage on the region) */
1387 if (vma->vm_file)
8feae131 1388 return -ENOMEM;
1da177e4 1389
8feae131
DH
1390 if (mm->map_count >= sysctl_max_map_count)
1391 return -ENOMEM;
1da177e4 1392
8feae131
DH
1393 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1394 if (!region)
1395 return -ENOMEM;
1da177e4 1396
3928d4f5 1397 new = vm_area_dup(vma);
8feae131
DH
1398 if (!new) {
1399 kmem_cache_free(vm_region_jar, region);
1400 return -ENOMEM;
1401 }
1402
1403 /* most fields are the same, copy all, and then fixup */
8feae131
DH
1404 *region = *vma->vm_region;
1405 new->vm_region = region;
1406
1407 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1408
1409 if (new_below) {
dd8632a1 1410 region->vm_top = region->vm_end = new->vm_end = addr;
8feae131
DH
1411 } else {
1412 region->vm_start = new->vm_start = addr;
1413 region->vm_pgoff = new->vm_pgoff += npages;
1da177e4 1414 }
8feae131
DH
1415
1416 if (new->vm_ops && new->vm_ops->open)
1417 new->vm_ops->open(new);
1418
1419 delete_vma_from_mm(vma);
1420 down_write(&nommu_region_sem);
1421 delete_nommu_region(vma->vm_region);
1422 if (new_below) {
1423 vma->vm_region->vm_start = vma->vm_start = addr;
1424 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1425 } else {
1426 vma->vm_region->vm_end = vma->vm_end = addr;
dd8632a1 1427 vma->vm_region->vm_top = addr;
8feae131
DH
1428 }
1429 add_nommu_region(vma->vm_region);
1430 add_nommu_region(new->vm_region);
1431 up_write(&nommu_region_sem);
1432 add_vma_to_mm(mm, vma);
1433 add_vma_to_mm(mm, new);
1434 return 0;
1da177e4
LT
1435}
1436
3034097a 1437/*
8feae131
DH
1438 * shrink a VMA by removing the specified chunk from either the beginning or
1439 * the end
3034097a 1440 */
8feae131
DH
1441static int shrink_vma(struct mm_struct *mm,
1442 struct vm_area_struct *vma,
1443 unsigned long from, unsigned long to)
1da177e4 1444{
8feae131 1445 struct vm_region *region;
1da177e4 1446
8feae131
DH
1447 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1448 * and list */
1449 delete_vma_from_mm(vma);
1450 if (from > vma->vm_start)
1451 vma->vm_end = from;
1452 else
1453 vma->vm_start = to;
1454 add_vma_to_mm(mm, vma);
1da177e4 1455
8feae131
DH
1456 /* cut the backing region down to size */
1457 region = vma->vm_region;
1e2ae599 1458 BUG_ON(region->vm_usage != 1);
8feae131
DH
1459
1460 down_write(&nommu_region_sem);
1461 delete_nommu_region(region);
dd8632a1
PM
1462 if (from > region->vm_start) {
1463 to = region->vm_top;
1464 region->vm_top = region->vm_end = from;
1465 } else {
8feae131 1466 region->vm_start = to;
dd8632a1 1467 }
8feae131
DH
1468 add_nommu_region(region);
1469 up_write(&nommu_region_sem);
1470
1471 free_page_series(from, to);
1472 return 0;
1473}
1da177e4 1474
8feae131
DH
1475/*
1476 * release a mapping
1477 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1478 * VMA, though it need not cover the whole VMA
1479 */
897ab3e0 1480int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
8feae131
DH
1481{
1482 struct vm_area_struct *vma;
f67d9b15 1483 unsigned long end;
8feae131 1484 int ret;
1da177e4 1485
f67d9b15 1486 len = PAGE_ALIGN(len);
8feae131
DH
1487 if (len == 0)
1488 return -EINVAL;
365e9c87 1489
f67d9b15
BL
1490 end = start + len;
1491
8feae131
DH
1492 /* find the first potentially overlapping VMA */
1493 vma = find_vma(mm, start);
1494 if (!vma) {
ac714904 1495 static int limit;
33e5d769 1496 if (limit < 5) {
22cc877b
LR
1497 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1498 current->pid, current->comm,
1499 start, start + len - 1);
33e5d769
DH
1500 limit++;
1501 }
8feae131
DH
1502 return -EINVAL;
1503 }
1da177e4 1504
8feae131
DH
1505 /* we're allowed to split an anonymous VMA but not a file-backed one */
1506 if (vma->vm_file) {
1507 do {
22cc877b 1508 if (start > vma->vm_start)
8feae131 1509 return -EINVAL;
8feae131
DH
1510 if (end == vma->vm_end)
1511 goto erase_whole_vma;
d75a310c
NK
1512 vma = vma->vm_next;
1513 } while (vma);
8feae131
DH
1514 return -EINVAL;
1515 } else {
1516 /* the chunk must be a subset of the VMA found */
1517 if (start == vma->vm_start && end == vma->vm_end)
1518 goto erase_whole_vma;
22cc877b 1519 if (start < vma->vm_start || end > vma->vm_end)
8feae131 1520 return -EINVAL;
1824cb75 1521 if (offset_in_page(start))
8feae131 1522 return -EINVAL;
1824cb75 1523 if (end != vma->vm_end && offset_in_page(end))
8feae131 1524 return -EINVAL;
8feae131
DH
1525 if (start != vma->vm_start && end != vma->vm_end) {
1526 ret = split_vma(mm, vma, start, 1);
22cc877b 1527 if (ret < 0)
8feae131 1528 return ret;
8feae131
DH
1529 }
1530 return shrink_vma(mm, vma, start, end);
1531 }
1da177e4 1532
8feae131
DH
1533erase_whole_vma:
1534 delete_vma_from_mm(vma);
1535 delete_vma(mm, vma);
1da177e4
LT
1536 return 0;
1537}
b5073173 1538EXPORT_SYMBOL(do_munmap);
1da177e4 1539
bfce281c 1540int vm_munmap(unsigned long addr, size_t len)
3034097a 1541{
bfce281c 1542 struct mm_struct *mm = current->mm;
3034097a 1543 int ret;
3034097a 1544
d8ed45c5 1545 mmap_write_lock(mm);
897ab3e0 1546 ret = do_munmap(mm, addr, len, NULL);
d8ed45c5 1547 mmap_write_unlock(mm);
3034097a
DH
1548 return ret;
1549}
a46ef99d
LT
1550EXPORT_SYMBOL(vm_munmap);
1551
1552SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1553{
bfce281c 1554 return vm_munmap(addr, len);
a46ef99d 1555}
3034097a
DH
1556
1557/*
8feae131 1558 * release all the mappings made in a process's VM space
3034097a 1559 */
8feae131 1560void exit_mmap(struct mm_struct *mm)
1da177e4 1561{
8feae131 1562 struct vm_area_struct *vma;
1da177e4 1563
8feae131
DH
1564 if (!mm)
1565 return;
1da177e4 1566
8feae131 1567 mm->total_vm = 0;
1da177e4 1568
8feae131
DH
1569 while ((vma = mm->mmap)) {
1570 mm->mmap = vma->vm_next;
1571 delete_vma_from_mm(vma);
1572 delete_vma(mm, vma);
04c34961 1573 cond_resched();
1da177e4
LT
1574 }
1575}
1576
5d22fc25 1577int vm_brk(unsigned long addr, unsigned long len)
1da177e4
LT
1578{
1579 return -ENOMEM;
1580}
1581
1582/*
6fa5f80b
DH
1583 * expand (or shrink) an existing mapping, potentially moving it at the same
1584 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1da177e4 1585 *
6fa5f80b 1586 * under NOMMU conditions, we only permit changing a mapping's size, and only
8feae131
DH
1587 * as long as it stays within the region allocated by do_mmap_private() and the
1588 * block is not shareable
1da177e4 1589 *
6fa5f80b 1590 * MREMAP_FIXED is not supported under NOMMU conditions
1da177e4 1591 */
4b377bab 1592static unsigned long do_mremap(unsigned long addr,
1da177e4
LT
1593 unsigned long old_len, unsigned long new_len,
1594 unsigned long flags, unsigned long new_addr)
1595{
6fa5f80b 1596 struct vm_area_struct *vma;
1da177e4
LT
1597
1598 /* insanity checks first */
f67d9b15
BL
1599 old_len = PAGE_ALIGN(old_len);
1600 new_len = PAGE_ALIGN(new_len);
8feae131 1601 if (old_len == 0 || new_len == 0)
1da177e4
LT
1602 return (unsigned long) -EINVAL;
1603
1824cb75 1604 if (offset_in_page(addr))
8feae131
DH
1605 return -EINVAL;
1606
1da177e4
LT
1607 if (flags & MREMAP_FIXED && new_addr != addr)
1608 return (unsigned long) -EINVAL;
1609
8feae131 1610 vma = find_vma_exact(current->mm, addr, old_len);
6fa5f80b
DH
1611 if (!vma)
1612 return (unsigned long) -EINVAL;
1da177e4 1613
6fa5f80b 1614 if (vma->vm_end != vma->vm_start + old_len)
1da177e4
LT
1615 return (unsigned long) -EFAULT;
1616
6fa5f80b 1617 if (vma->vm_flags & VM_MAYSHARE)
1da177e4
LT
1618 return (unsigned long) -EPERM;
1619
8feae131 1620 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1da177e4
LT
1621 return (unsigned long) -ENOMEM;
1622
1623 /* all checks complete - do it */
6fa5f80b 1624 vma->vm_end = vma->vm_start + new_len;
6fa5f80b
DH
1625 return vma->vm_start;
1626}
1627
6a6160a7
HC
1628SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1629 unsigned long, new_len, unsigned long, flags,
1630 unsigned long, new_addr)
6fa5f80b
DH
1631{
1632 unsigned long ret;
1633
d8ed45c5 1634 mmap_write_lock(current->mm);
6fa5f80b 1635 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
d8ed45c5 1636 mmap_write_unlock(current->mm);
6fa5f80b 1637 return ret;
1da177e4
LT
1638}
1639
df06b37f
KB
1640struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1641 unsigned int foll_flags)
1da177e4
LT
1642{
1643 return NULL;
1644}
1645
8f3b1327
BL
1646int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1647 unsigned long pfn, unsigned long size, pgprot_t prot)
1da177e4 1648{
8f3b1327
BL
1649 if (addr != (pfn << PAGE_SHIFT))
1650 return -EINVAL;
1651
314e51b9 1652 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
66aa2b4b 1653 return 0;
1da177e4 1654}
22c4af40 1655EXPORT_SYMBOL(remap_pfn_range);
1da177e4 1656
3c0b9de6
LT
1657int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1658{
1659 unsigned long pfn = start >> PAGE_SHIFT;
1660 unsigned long vm_len = vma->vm_end - vma->vm_start;
1661
1662 pfn += vma->vm_pgoff;
1663 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1664}
1665EXPORT_SYMBOL(vm_iomap_memory);
1666
f905bc44
PM
1667int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1668 unsigned long pgoff)
1669{
1670 unsigned int size = vma->vm_end - vma->vm_start;
1671
1672 if (!(vma->vm_flags & VM_USERMAP))
1673 return -EINVAL;
1674
1675 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1676 vma->vm_end = vma->vm_start + size;
1677
1678 return 0;
1679}
1680EXPORT_SYMBOL(remap_vmalloc_range);
1681
1da177e4
LT
1682unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1683 unsigned long len, unsigned long pgoff, unsigned long flags)
1684{
1685 return -ENOMEM;
1686}
1687
2bcd6454 1688vm_fault_t filemap_fault(struct vm_fault *vmf)
b0e15190
DH
1689{
1690 BUG();
d0217ac0 1691 return 0;
b0e15190 1692}
b5073173 1693EXPORT_SYMBOL(filemap_fault);
0ec76a11 1694
82b0f8c3 1695void filemap_map_pages(struct vm_fault *vmf,
bae473a4 1696 pgoff_t start_pgoff, pgoff_t end_pgoff)
f1820361
KS
1697{
1698 BUG();
1699}
1700EXPORT_SYMBOL(filemap_map_pages);
1701
84d77d3f 1702int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
442486ec 1703 unsigned long addr, void *buf, int len, unsigned int gup_flags)
0ec76a11 1704{
0ec76a11 1705 struct vm_area_struct *vma;
442486ec 1706 int write = gup_flags & FOLL_WRITE;
0ec76a11 1707
d8ed45c5 1708 if (mmap_read_lock_killable(mm))
1e426fe2 1709 return 0;
0ec76a11
DH
1710
1711 /* the access must start within one of the target process's mappings */
0159b141
DH
1712 vma = find_vma(mm, addr);
1713 if (vma) {
0ec76a11
DH
1714 /* don't overrun this mapping */
1715 if (addr + len >= vma->vm_end)
1716 len = vma->vm_end - addr;
1717
1718 /* only read or write mappings where it is permitted */
d00c7b99 1719 if (write && vma->vm_flags & VM_MAYWRITE)
7959722b
JZ
1720 copy_to_user_page(vma, NULL, addr,
1721 (void *) addr, buf, len);
d00c7b99 1722 else if (!write && vma->vm_flags & VM_MAYREAD)
7959722b
JZ
1723 copy_from_user_page(vma, NULL, addr,
1724 buf, (void *) addr, len);
0ec76a11
DH
1725 else
1726 len = 0;
1727 } else {
1728 len = 0;
1729 }
1730
d8ed45c5 1731 mmap_read_unlock(mm);
f55f199b
MF
1732
1733 return len;
1734}
1735
1736/**
b7701a5f 1737 * access_remote_vm - access another process' address space
f55f199b
MF
1738 * @mm: the mm_struct of the target address space
1739 * @addr: start address to access
1740 * @buf: source or destination buffer
1741 * @len: number of bytes to transfer
6347e8d5 1742 * @gup_flags: flags modifying lookup behaviour
f55f199b
MF
1743 *
1744 * The caller must hold a reference on @mm.
1745 */
1746int access_remote_vm(struct mm_struct *mm, unsigned long addr,
6347e8d5 1747 void *buf, int len, unsigned int gup_flags)
f55f199b 1748{
6347e8d5 1749 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
f55f199b
MF
1750}
1751
1752/*
1753 * Access another process' address space.
1754 * - source/target buffer must be kernel space
1755 */
f307ab6d
LS
1756int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1757 unsigned int gup_flags)
f55f199b
MF
1758{
1759 struct mm_struct *mm;
1760
1761 if (addr + len < addr)
1762 return 0;
1763
1764 mm = get_task_mm(tsk);
1765 if (!mm)
1766 return 0;
1767
f307ab6d 1768 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
f55f199b 1769
0ec76a11
DH
1770 mmput(mm);
1771 return len;
1772}
fcd35857 1773EXPORT_SYMBOL_GPL(access_process_vm);
7e660872
DH
1774
1775/**
1776 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1777 * @inode: The inode to check
1778 * @size: The current filesize of the inode
1779 * @newsize: The proposed filesize of the inode
1780 *
1781 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1782 * make sure that that any outstanding VMAs aren't broken and then shrink the
1783 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1784 * automatically grant mappings that are too large.
1785 */
1786int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1787 size_t newsize)
1788{
1789 struct vm_area_struct *vma;
7e660872
DH
1790 struct vm_region *region;
1791 pgoff_t low, high;
1792 size_t r_size, r_top;
1793
1794 low = newsize >> PAGE_SHIFT;
1795 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1796
1797 down_write(&nommu_region_sem);
1acf2e04 1798 i_mmap_lock_read(inode->i_mapping);
7e660872
DH
1799
1800 /* search for VMAs that fall within the dead zone */
6b2dbba8 1801 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
7e660872
DH
1802 /* found one - only interested if it's shared out of the page
1803 * cache */
1804 if (vma->vm_flags & VM_SHARED) {
1acf2e04 1805 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1806 up_write(&nommu_region_sem);
1807 return -ETXTBSY; /* not quite true, but near enough */
1808 }
1809 }
1810
1811 /* reduce any regions that overlap the dead zone - if in existence,
1812 * these will be pointed to by VMAs that don't overlap the dead zone
1813 *
1814 * we don't check for any regions that start beyond the EOF as there
1815 * shouldn't be any
1816 */
1acf2e04 1817 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
7e660872
DH
1818 if (!(vma->vm_flags & VM_SHARED))
1819 continue;
1820
1821 region = vma->vm_region;
1822 r_size = region->vm_top - region->vm_start;
1823 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1824
1825 if (r_top > newsize) {
1826 region->vm_top -= r_top - newsize;
1827 if (region->vm_end > region->vm_top)
1828 region->vm_end = region->vm_top;
1829 }
1830 }
1831
1acf2e04 1832 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1833 up_write(&nommu_region_sem);
1834 return 0;
1835}
c9b1d098
AS
1836
1837/*
1838 * Initialise sysctl_user_reserve_kbytes.
1839 *
1840 * This is intended to prevent a user from starting a single memory hogging
1841 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1842 * mode.
1843 *
1844 * The default value is min(3% of free memory, 128MB)
1845 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1846 */
1847static int __meminit init_user_reserve(void)
1848{
1849 unsigned long free_kbytes;
1850
c41f012a 1851 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
c9b1d098
AS
1852
1853 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1854 return 0;
1855}
a4bc6fc7 1856subsys_initcall(init_user_reserve);
4eeab4f5
AS
1857
1858/*
1859 * Initialise sysctl_admin_reserve_kbytes.
1860 *
1861 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1862 * to log in and kill a memory hogging process.
1863 *
1864 * Systems with more than 256MB will reserve 8MB, enough to recover
1865 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1866 * only reserve 3% of free pages by default.
1867 */
1868static int __meminit init_admin_reserve(void)
1869{
1870 unsigned long free_kbytes;
1871
c41f012a 1872 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
4eeab4f5
AS
1873
1874 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1875 return 0;
1876}
a4bc6fc7 1877subsys_initcall(init_admin_reserve);