mm/nommu: factor out check for NOMMU shared mappings into is_nommu_shared_mapping()
[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 *
dd19d293 8 * See Documentation/admin-guide/mm/nommu-mmap.rst
1da177e4 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>
1da177e4
LT
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>
1da177e4 29#include <linux/backing-dev.h>
3b32123d 30#include <linux/compiler.h>
1da177e4
LT
31#include <linux/mount.h>
32#include <linux/personality.h>
33#include <linux/security.h>
34#include <linux/syscalls.h>
120a795d 35#include <linux/audit.h>
b1de0d13 36#include <linux/printk.h>
1da177e4 37
7c0f6ba6 38#include <linux/uaccess.h>
1da177e4
LT
39#include <asm/tlb.h>
40#include <asm/tlbflush.h>
eb8cdec4 41#include <asm/mmu_context.h>
8feae131
DH
42#include "internal.h"
43
1da177e4 44void *high_memory;
944b6874 45EXPORT_SYMBOL(high_memory);
1da177e4
LT
46struct page *mem_map;
47unsigned long max_mapnr;
5b8bf307 48EXPORT_SYMBOL(max_mapnr);
4266c97a 49unsigned long highest_memmap_pfn;
fc4d5c29 50int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
1da177e4
LT
51int heap_stack_gap = 0;
52
33e5d769 53atomic_long_t mmap_pages_allocated;
8feae131 54
1da177e4 55EXPORT_SYMBOL(mem_map);
1da177e4 56
8feae131
DH
57/* list of mapped, potentially shareable regions */
58static struct kmem_cache *vm_region_jar;
59struct rb_root nommu_region_tree = RB_ROOT;
60DECLARE_RWSEM(nommu_region_sem);
1da177e4 61
f0f37e2f 62const struct vm_operations_struct generic_file_vm_ops = {
1da177e4
LT
63};
64
1da177e4
LT
65/*
66 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
68 *
69 * Doesn't have to be accurate, i.e. may have races.
70 */
71unsigned int kobjsize(const void *objp)
72{
73 struct page *page;
74
4016a139
MH
75 /*
76 * If the object we have should not have ksize performed on it,
77 * return size of 0
78 */
5a1603be 79 if (!objp || !virt_addr_valid(objp))
6cfd53fc
PM
80 return 0;
81
82 page = virt_to_head_page(objp);
6cfd53fc
PM
83
84 /*
85 * If the allocator sets PageSlab, we know the pointer came from
86 * kmalloc().
87 */
1da177e4
LT
88 if (PageSlab(page))
89 return ksize(objp);
90
ab2e83ea
PM
91 /*
92 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
96 */
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
99
100 vma = find_vma(current->mm, (unsigned long)objp);
101 if (vma)
102 return vma->vm_end - vma->vm_start;
103 }
104
6cfd53fc
PM
105 /*
106 * The ksize() function is only guaranteed to work for pointers
5a1603be 107 * returned by kmalloc(). So handle arbitrary pointers here.
6cfd53fc 108 */
a50b854e 109 return page_size(page);
1da177e4
LT
110}
111
dfc2f91a
PM
112/**
113 * follow_pfn - look up PFN at a user virtual address
114 * @vma: memory mapping
115 * @address: user virtual address
116 * @pfn: location to store found PFN
117 *
118 * Only IO mappings and raw PFN mappings are allowed.
119 *
120 * Returns zero and the pfn at @pfn on success, -ve otherwise.
121 */
122int follow_pfn(struct vm_area_struct *vma, unsigned long address,
123 unsigned long *pfn)
124{
125 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
126 return -EINVAL;
127
128 *pfn = address >> PAGE_SHIFT;
129 return 0;
130}
131EXPORT_SYMBOL(follow_pfn);
132
f1c4069e 133LIST_HEAD(vmap_area_list);
1da177e4 134
b3bdda02 135void vfree(const void *addr)
1da177e4
LT
136{
137 kfree(addr);
138}
b5073173 139EXPORT_SYMBOL(vfree);
1da177e4 140
88dca4ca 141void *__vmalloc(unsigned long size, gfp_t gfp_mask)
1da177e4
LT
142{
143 /*
8518609d
RD
144 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
145 * returns only a logical address.
1da177e4 146 */
84097518 147 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
1da177e4 148}
b5073173 149EXPORT_SYMBOL(__vmalloc);
1da177e4 150
041de93f
CH
151void *__vmalloc_node_range(unsigned long size, unsigned long align,
152 unsigned long start, unsigned long end, gfp_t gfp_mask,
153 pgprot_t prot, unsigned long vm_flags, int node,
154 const void *caller)
155{
156 return __vmalloc(size, gfp_mask);
157}
158
2b905948
CH
159void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
160 int node, const void *caller)
a7c3e901 161{
2b905948 162 return __vmalloc(size, gfp_mask);
a7c3e901
MH
163}
164
ed81745a 165static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
f905bc44
PM
166{
167 void *ret;
168
88dca4ca 169 ret = __vmalloc(size, flags);
f905bc44
PM
170 if (ret) {
171 struct vm_area_struct *vma;
172
d8ed45c5 173 mmap_write_lock(current->mm);
f905bc44
PM
174 vma = find_vma(current->mm, (unsigned long)ret);
175 if (vma)
176 vma->vm_flags |= VM_USERMAP;
d8ed45c5 177 mmap_write_unlock(current->mm);
f905bc44
PM
178 }
179
180 return ret;
181}
ed81745a
AN
182
183void *vmalloc_user(unsigned long size)
184{
185 return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
186}
f905bc44
PM
187EXPORT_SYMBOL(vmalloc_user);
188
b3bdda02 189struct page *vmalloc_to_page(const void *addr)
1da177e4
LT
190{
191 return virt_to_page(addr);
192}
b5073173 193EXPORT_SYMBOL(vmalloc_to_page);
1da177e4 194
b3bdda02 195unsigned long vmalloc_to_pfn(const void *addr)
1da177e4
LT
196{
197 return page_to_pfn(virt_to_page(addr));
198}
b5073173 199EXPORT_SYMBOL(vmalloc_to_pfn);
1da177e4
LT
200
201long vread(char *buf, char *addr, unsigned long count)
202{
9bde916b
CG
203 /* Don't allow overflow */
204 if ((unsigned long) buf + count < count)
205 count = -(unsigned long) buf;
206
1da177e4
LT
207 memcpy(buf, addr, count);
208 return count;
209}
210
1da177e4 211/*
e1c05067 212 * vmalloc - allocate virtually contiguous memory
1da177e4
LT
213 *
214 * @size: allocation size
215 *
216 * Allocate enough pages to cover @size from the page level
e1c05067 217 * allocator and map them into contiguous kernel virtual space.
1da177e4 218 *
c1c8897f 219 * For tight control over page level allocator and protection flags
1da177e4
LT
220 * use __vmalloc() instead.
221 */
222void *vmalloc(unsigned long size)
223{
176056fd 224 return __vmalloc(size, GFP_KERNEL);
1da177e4 225}
f6138882
AM
226EXPORT_SYMBOL(vmalloc);
227
0fc74d82
LT
228void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) __weak __alias(__vmalloc);
229
e1ca7788 230/*
e1c05067 231 * vzalloc - allocate virtually contiguous memory with zero fill
e1ca7788
DY
232 *
233 * @size: allocation size
234 *
235 * Allocate enough pages to cover @size from the page level
e1c05067 236 * allocator and map them into contiguous kernel virtual space.
e1ca7788
DY
237 * The memory allocated is set to zero.
238 *
239 * For tight control over page level allocator and protection flags
240 * use __vmalloc() instead.
241 */
242void *vzalloc(unsigned long size)
243{
176056fd 244 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
e1ca7788
DY
245}
246EXPORT_SYMBOL(vzalloc);
247
248/**
249 * vmalloc_node - allocate memory on a specific node
250 * @size: allocation size
251 * @node: numa node
252 *
253 * Allocate enough pages to cover @size from the page level
254 * allocator and map them into contiguous kernel virtual space.
255 *
256 * For tight control over page level allocator and protection flags
257 * use __vmalloc() instead.
258 */
f6138882
AM
259void *vmalloc_node(unsigned long size, int node)
260{
261 return vmalloc(size);
262}
9a14f653 263EXPORT_SYMBOL(vmalloc_node);
e1ca7788
DY
264
265/**
266 * vzalloc_node - allocate memory on a specific node with zero fill
267 * @size: allocation size
268 * @node: numa node
269 *
270 * Allocate enough pages to cover @size from the page level
271 * allocator and map them into contiguous kernel virtual space.
272 * The memory allocated is set to zero.
273 *
274 * For tight control over page level allocator and protection flags
275 * use __vmalloc() instead.
276 */
277void *vzalloc_node(unsigned long size, int node)
278{
279 return vzalloc(size);
280}
281EXPORT_SYMBOL(vzalloc_node);
1da177e4 282
b5073173
PM
283/**
284 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1da177e4
LT
285 * @size: allocation size
286 *
287 * Allocate enough 32bit PA addressable pages to cover @size from the
e1c05067 288 * page level allocator and map them into contiguous kernel virtual space.
1da177e4
LT
289 */
290void *vmalloc_32(unsigned long size)
291{
88dca4ca 292 return __vmalloc(size, GFP_KERNEL);
1da177e4 293}
b5073173
PM
294EXPORT_SYMBOL(vmalloc_32);
295
296/**
297 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
298 * @size: allocation size
299 *
300 * The resulting memory area is 32bit addressable and zeroed so it can be
301 * mapped to userspace without leaking data.
f905bc44
PM
302 *
303 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
304 * remap_vmalloc_range() are permissible.
b5073173
PM
305 */
306void *vmalloc_32_user(unsigned long size)
307{
f905bc44
PM
308 /*
309 * We'll have to sort out the ZONE_DMA bits for 64-bit,
310 * but for now this can simply use vmalloc_user() directly.
311 */
312 return vmalloc_user(size);
b5073173
PM
313}
314EXPORT_SYMBOL(vmalloc_32_user);
1da177e4
LT
315
316void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
317{
318 BUG();
319 return NULL;
320}
b5073173 321EXPORT_SYMBOL(vmap);
1da177e4 322
b3bdda02 323void vunmap(const void *addr)
1da177e4
LT
324{
325 BUG();
326}
b5073173 327EXPORT_SYMBOL(vunmap);
1da177e4 328
d4efd79a 329void *vm_map_ram(struct page **pages, unsigned int count, int node)
eb6434d9
PM
330{
331 BUG();
332 return NULL;
333}
334EXPORT_SYMBOL(vm_map_ram);
335
336void vm_unmap_ram(const void *mem, unsigned int count)
337{
338 BUG();
339}
340EXPORT_SYMBOL(vm_unmap_ram);
341
342void vm_unmap_aliases(void)
343{
344}
345EXPORT_SYMBOL_GPL(vm_unmap_aliases);
346
29c185e5
PM
347void free_vm_area(struct vm_struct *area)
348{
349 BUG();
350}
351EXPORT_SYMBOL_GPL(free_vm_area);
352
b5073173
PM
353int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
354 struct page *page)
355{
356 return -EINVAL;
357}
358EXPORT_SYMBOL(vm_insert_page);
359
a667d745
SJ
360int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
361 unsigned long num)
362{
363 return -EINVAL;
364}
365EXPORT_SYMBOL(vm_map_pages);
366
367int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
368 unsigned long num)
369{
370 return -EINVAL;
371}
372EXPORT_SYMBOL(vm_map_pages_zero);
373
1da177e4
LT
374/*
375 * sys_brk() for the most part doesn't need the global kernel
376 * lock, except when an application is doing something nasty
377 * like trying to un-brk an area that has already been mapped
378 * to a regular file. in this case, the unmapping will need
379 * to invoke file system routines that need the global lock.
380 */
6a6160a7 381SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4
LT
382{
383 struct mm_struct *mm = current->mm;
384
385 if (brk < mm->start_brk || brk > mm->context.end_brk)
386 return mm->brk;
387
388 if (mm->brk == brk)
389 return mm->brk;
390
391 /*
392 * Always allow shrinking brk
393 */
394 if (brk <= mm->brk) {
395 mm->brk = brk;
396 return brk;
397 }
398
399 /*
400 * Ok, looks good - let it rip.
401 */
a75a2df6 402 flush_icache_user_range(mm->brk, brk);
1da177e4
LT
403 return mm->brk = brk;
404}
405
8feae131 406/*
3edf41d8 407 * initialise the percpu counter for VM and region record slabs
8feae131
DH
408 */
409void __init mmap_init(void)
1da177e4 410{
00a62ce9
KM
411 int ret;
412
908c7f19 413 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
00a62ce9 414 VM_BUG_ON(ret);
5d097056 415 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
1da177e4 416}
1da177e4 417
3034097a 418/*
8feae131
DH
419 * validate the region tree
420 * - the caller must hold the region lock
3034097a 421 */
8feae131
DH
422#ifdef CONFIG_DEBUG_NOMMU_REGIONS
423static noinline void validate_nommu_regions(void)
3034097a 424{
8feae131
DH
425 struct vm_region *region, *last;
426 struct rb_node *p, *lastp;
3034097a 427
8feae131
DH
428 lastp = rb_first(&nommu_region_tree);
429 if (!lastp)
430 return;
431
432 last = rb_entry(lastp, struct vm_region, vm_rb);
c9427bc0
GT
433 BUG_ON(last->vm_end <= last->vm_start);
434 BUG_ON(last->vm_top < last->vm_end);
8feae131
DH
435
436 while ((p = rb_next(lastp))) {
437 region = rb_entry(p, struct vm_region, vm_rb);
438 last = rb_entry(lastp, struct vm_region, vm_rb);
439
c9427bc0
GT
440 BUG_ON(region->vm_end <= region->vm_start);
441 BUG_ON(region->vm_top < region->vm_end);
442 BUG_ON(region->vm_start < last->vm_top);
3034097a 443
8feae131
DH
444 lastp = p;
445 }
3034097a 446}
8feae131 447#else
33e5d769
DH
448static void validate_nommu_regions(void)
449{
450}
8feae131 451#endif
3034097a
DH
452
453/*
8feae131 454 * add a region into the global tree
3034097a 455 */
8feae131 456static void add_nommu_region(struct vm_region *region)
3034097a 457{
8feae131
DH
458 struct vm_region *pregion;
459 struct rb_node **p, *parent;
3034097a 460
8feae131
DH
461 validate_nommu_regions();
462
8feae131
DH
463 parent = NULL;
464 p = &nommu_region_tree.rb_node;
465 while (*p) {
466 parent = *p;
467 pregion = rb_entry(parent, struct vm_region, vm_rb);
468 if (region->vm_start < pregion->vm_start)
469 p = &(*p)->rb_left;
470 else if (region->vm_start > pregion->vm_start)
471 p = &(*p)->rb_right;
472 else if (pregion == region)
473 return;
474 else
475 BUG();
3034097a
DH
476 }
477
8feae131
DH
478 rb_link_node(&region->vm_rb, parent, p);
479 rb_insert_color(&region->vm_rb, &nommu_region_tree);
3034097a 480
8feae131 481 validate_nommu_regions();
3034097a 482}
3034097a 483
930e652a 484/*
8feae131 485 * delete a region from the global tree
930e652a 486 */
8feae131 487static void delete_nommu_region(struct vm_region *region)
930e652a 488{
8feae131 489 BUG_ON(!nommu_region_tree.rb_node);
930e652a 490
8feae131
DH
491 validate_nommu_regions();
492 rb_erase(&region->vm_rb, &nommu_region_tree);
493 validate_nommu_regions();
57c8f63e
GU
494}
495
6fa5f80b 496/*
8feae131 497 * free a contiguous series of pages
6fa5f80b 498 */
8feae131 499static void free_page_series(unsigned long from, unsigned long to)
6fa5f80b 500{
8feae131 501 for (; from < to; from += PAGE_SIZE) {
9330723c 502 struct page *page = virt_to_page((void *)from);
8feae131 503
33e5d769 504 atomic_long_dec(&mmap_pages_allocated);
8feae131 505 put_page(page);
6fa5f80b 506 }
6fa5f80b
DH
507}
508
3034097a 509/*
8feae131 510 * release a reference to a region
33e5d769 511 * - the caller must hold the region semaphore for writing, which this releases
dd8632a1 512 * - the region may not have been added to the tree yet, in which case vm_top
8feae131 513 * will equal vm_start
3034097a 514 */
8feae131
DH
515static void __put_nommu_region(struct vm_region *region)
516 __releases(nommu_region_sem)
1da177e4 517{
8feae131 518 BUG_ON(!nommu_region_tree.rb_node);
1da177e4 519
1e2ae599 520 if (--region->vm_usage == 0) {
dd8632a1 521 if (region->vm_top > region->vm_start)
8feae131
DH
522 delete_nommu_region(region);
523 up_write(&nommu_region_sem);
524
525 if (region->vm_file)
526 fput(region->vm_file);
527
528 /* IO memory and memory shared directly out of the pagecache
529 * from ramfs/tmpfs mustn't be released here */
22cc877b 530 if (region->vm_flags & VM_MAPPED_COPY)
dd8632a1 531 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
532 kmem_cache_free(vm_region_jar, region);
533 } else {
534 up_write(&nommu_region_sem);
1da177e4 535 }
8feae131 536}
1da177e4 537
8feae131
DH
538/*
539 * release a reference to a region
540 */
541static void put_nommu_region(struct vm_region *region)
542{
543 down_write(&nommu_region_sem);
544 __put_nommu_region(region);
1da177e4
LT
545}
546
d4af56c5
LH
547void vma_mas_store(struct vm_area_struct *vma, struct ma_state *mas)
548{
549 mas_set_range(mas, vma->vm_start, vma->vm_end - 1);
550 mas_store_prealloc(mas, vma);
551}
552
553void vma_mas_remove(struct vm_area_struct *vma, struct ma_state *mas)
554{
555 mas->index = vma->vm_start;
556 mas->last = vma->vm_end - 1;
557 mas_store_prealloc(mas, NULL);
558}
559
8220543d 560static void setup_vma_to_mm(struct vm_area_struct *vma, struct mm_struct *mm)
1da177e4 561{
8feae131 562 vma->vm_mm = mm;
1da177e4
LT
563
564 /* add the VMA to the mapping */
565 if (vma->vm_file) {
8220543d 566 struct address_space *mapping = vma->vm_file->f_mapping;
1da177e4 567
83cde9e8 568 i_mmap_lock_write(mapping);
1da177e4 569 flush_dcache_mmap_lock(mapping);
6b2dbba8 570 vma_interval_tree_insert(vma, &mapping->i_mmap);
1da177e4 571 flush_dcache_mmap_unlock(mapping);
83cde9e8 572 i_mmap_unlock_write(mapping);
1da177e4 573 }
8220543d 574}
1da177e4 575
8220543d
MWO
576/*
577 * mas_add_vma_to_mm() - Maple state variant of add_mas_to_mm().
578 * @mas: The maple state with preallocations.
579 * @mm: The mm_struct
580 * @vma: The vma to add
581 *
582 */
583static void mas_add_vma_to_mm(struct ma_state *mas, struct mm_struct *mm,
584 struct vm_area_struct *vma)
585{
8220543d
MWO
586 BUG_ON(!vma->vm_region);
587
588 setup_vma_to_mm(vma, mm);
fd9edbdb 589 mm->map_count++;
8220543d 590
8feae131 591 /* add the VMA to the tree */
8220543d 592 vma_mas_store(vma, mas);
1da177e4
LT
593}
594
3034097a 595/*
8220543d
MWO
596 * add a VMA into a process's mm_struct in the appropriate place in the list
597 * and tree and add to the address space's page tree also if not an anonymous
598 * page
599 * - should be called with mm->mmap_lock held writelocked
3034097a 600 */
8220543d 601static int add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 602{
8220543d
MWO
603 MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end);
604
605 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
606 pr_warn("Allocation of vma tree for process %d failed\n",
607 current->pid);
608 return -ENOMEM;
609 }
610 mas_add_vma_to_mm(&mas, mm, vma);
611 return 0;
612}
8feae131 613
8220543d
MWO
614static void cleanup_vma_from_mm(struct vm_area_struct *vma)
615{
7964cf8c 616 vma->vm_mm->map_count--;
1da177e4
LT
617 /* remove the VMA from the mapping */
618 if (vma->vm_file) {
7964cf8c 619 struct address_space *mapping;
1da177e4
LT
620 mapping = vma->vm_file->f_mapping;
621
83cde9e8 622 i_mmap_lock_write(mapping);
1da177e4 623 flush_dcache_mmap_lock(mapping);
6b2dbba8 624 vma_interval_tree_remove(vma, &mapping->i_mmap);
1da177e4 625 flush_dcache_mmap_unlock(mapping);
83cde9e8 626 i_mmap_unlock_write(mapping);
1da177e4 627 }
8220543d
MWO
628}
629/*
630 * delete a VMA from its owning mm_struct and address space
631 */
632static int delete_vma_from_mm(struct vm_area_struct *vma)
633{
634 MA_STATE(mas, &vma->vm_mm->mm_mt, 0, 0);
635
636 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
637 pr_warn("Allocation of vma tree for process %d failed\n",
638 current->pid);
639 return -ENOMEM;
640 }
641 cleanup_vma_from_mm(vma);
1da177e4 642
8feae131 643 /* remove from the MM's tree and list */
524e00b3 644 vma_mas_remove(vma, &mas);
8220543d 645 return 0;
8feae131
DH
646}
647
648/*
649 * destroy a VMA record
650 */
651static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
652{
8feae131
DH
653 if (vma->vm_ops && vma->vm_ops->close)
654 vma->vm_ops->close(vma);
e9714acf 655 if (vma->vm_file)
8feae131 656 fput(vma->vm_file);
8feae131 657 put_nommu_region(vma->vm_region);
3928d4f5 658 vm_area_free(vma);
8feae131
DH
659}
660
abdba2dd
LH
661struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
662 unsigned long start_addr,
663 unsigned long end_addr)
664{
665 unsigned long index = start_addr;
666
667 mmap_assert_locked(mm);
668 return mt_find(&mm->mm_mt, &index, end_addr - 1);
669}
670EXPORT_SYMBOL(find_vma_intersection);
671
8feae131
DH
672/*
673 * look up the first VMA in which addr resides, NULL if none
c1e8d7c6 674 * - should be called with mm->mmap_lock at least held readlocked
8feae131
DH
675 */
676struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
677{
524e00b3 678 MA_STATE(mas, &mm->mm_mt, addr, addr);
8feae131 679
7964cf8c 680 return mas_walk(&mas);
8feae131
DH
681}
682EXPORT_SYMBOL(find_vma);
683
684/*
685 * find a VMA
686 * - we don't extend stack VMAs under NOMMU conditions
687 */
688struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
689{
7561e8ca 690 return find_vma(mm, addr);
8feae131
DH
691}
692
693/*
694 * expand a stack to a given address
695 * - not supported under NOMMU conditions
696 */
697int expand_stack(struct vm_area_struct *vma, unsigned long address)
698{
699 return -ENOMEM;
700}
701
702/*
703 * look up the first VMA exactly that exactly matches addr
c1e8d7c6 704 * - should be called with mm->mmap_lock at least held readlocked
8feae131
DH
705 */
706static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
707 unsigned long addr,
708 unsigned long len)
709{
710 struct vm_area_struct *vma;
8feae131 711 unsigned long end = addr + len;
524e00b3 712 MA_STATE(mas, &mm->mm_mt, addr, addr);
8feae131 713
524e00b3
LH
714 vma = mas_walk(&mas);
715 if (!vma)
716 return NULL;
717 if (vma->vm_start != addr)
718 return NULL;
719 if (vma->vm_end != end)
720 return NULL;
721
524e00b3 722 return vma;
1da177e4
LT
723}
724
725/*
726 * determine whether a mapping should be permitted and, if so, what sort of
727 * mapping we're capable of supporting
728 */
729static int validate_mmap_request(struct file *file,
730 unsigned long addr,
731 unsigned long len,
732 unsigned long prot,
733 unsigned long flags,
734 unsigned long pgoff,
735 unsigned long *_capabilities)
736{
8feae131 737 unsigned long capabilities, rlen;
1da177e4
LT
738 int ret;
739
740 /* do the simple checks first */
22cc877b 741 if (flags & MAP_FIXED)
1da177e4 742 return -EINVAL;
1da177e4
LT
743
744 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
745 (flags & MAP_TYPE) != MAP_SHARED)
746 return -EINVAL;
747
f81cff0d 748 if (!len)
1da177e4
LT
749 return -EINVAL;
750
f81cff0d 751 /* Careful about overflows.. */
8feae131
DH
752 rlen = PAGE_ALIGN(len);
753 if (!rlen || rlen > TASK_SIZE)
f81cff0d
MF
754 return -ENOMEM;
755
1da177e4 756 /* offset overflow? */
8feae131 757 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
f81cff0d 758 return -EOVERFLOW;
1da177e4
LT
759
760 if (file) {
1da177e4 761 /* files must support mmap */
72c2d531 762 if (!file->f_op->mmap)
1da177e4
LT
763 return -ENODEV;
764
765 /* work out if what we've got could possibly be shared
766 * - we support chardevs that provide their own "memory"
767 * - we support files/blockdevs that are memory backed
768 */
b4caecd4
CH
769 if (file->f_op->mmap_capabilities) {
770 capabilities = file->f_op->mmap_capabilities(file);
771 } else {
1da177e4
LT
772 /* no explicit capabilities set, so assume some
773 * defaults */
496ad9aa 774 switch (file_inode(file)->i_mode & S_IFMT) {
1da177e4
LT
775 case S_IFREG:
776 case S_IFBLK:
b4caecd4 777 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
778 break;
779
780 case S_IFCHR:
781 capabilities =
b4caecd4
CH
782 NOMMU_MAP_DIRECT |
783 NOMMU_MAP_READ |
784 NOMMU_MAP_WRITE;
1da177e4
LT
785 break;
786
787 default:
788 return -EINVAL;
789 }
790 }
791
792 /* eliminate any capabilities that we can't support on this
793 * device */
794 if (!file->f_op->get_unmapped_area)
b4caecd4 795 capabilities &= ~NOMMU_MAP_DIRECT;
6e242a1c 796 if (!(file->f_mode & FMODE_CAN_READ))
b4caecd4 797 capabilities &= ~NOMMU_MAP_COPY;
1da177e4 798
28d7a6ae
GY
799 /* The file shall have been opened with read permission. */
800 if (!(file->f_mode & FMODE_READ))
801 return -EACCES;
802
1da177e4
LT
803 if (flags & MAP_SHARED) {
804 /* do checks for writing, appending and locking */
805 if ((prot & PROT_WRITE) &&
806 !(file->f_mode & FMODE_WRITE))
807 return -EACCES;
808
496ad9aa 809 if (IS_APPEND(file_inode(file)) &&
1da177e4
LT
810 (file->f_mode & FMODE_WRITE))
811 return -EACCES;
812
b4caecd4 813 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
814 return -ENODEV;
815
1da177e4 816 /* we mustn't privatise shared mappings */
b4caecd4 817 capabilities &= ~NOMMU_MAP_COPY;
ac714904 818 } else {
1da177e4
LT
819 /* we're going to read the file into private memory we
820 * allocate */
b4caecd4 821 if (!(capabilities & NOMMU_MAP_COPY))
1da177e4
LT
822 return -ENODEV;
823
824 /* we don't permit a private writable mapping to be
825 * shared with the backing device */
826 if (prot & PROT_WRITE)
b4caecd4 827 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4
LT
828 }
829
b4caecd4
CH
830 if (capabilities & NOMMU_MAP_DIRECT) {
831 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
832 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
833 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
3c7b2045 834 ) {
b4caecd4 835 capabilities &= ~NOMMU_MAP_DIRECT;
3c7b2045 836 if (flags & MAP_SHARED) {
22cc877b 837 pr_warn("MAP_SHARED not completely supported on !MMU\n");
3c7b2045
BS
838 return -EINVAL;
839 }
840 }
841 }
842
1da177e4
LT
843 /* handle executable mappings and implied executable
844 * mappings */
90f8572b 845 if (path_noexec(&file->f_path)) {
1da177e4
LT
846 if (prot & PROT_EXEC)
847 return -EPERM;
ac714904 848 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1da177e4
LT
849 /* handle implication of PROT_EXEC by PROT_READ */
850 if (current->personality & READ_IMPLIES_EXEC) {
b4caecd4 851 if (capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
852 prot |= PROT_EXEC;
853 }
ac714904 854 } else if ((prot & PROT_READ) &&
1da177e4 855 (prot & PROT_EXEC) &&
b4caecd4 856 !(capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
857 ) {
858 /* backing file is not executable, try to copy */
b4caecd4 859 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4 860 }
ac714904 861 } else {
1da177e4
LT
862 /* anonymous mappings are always memory backed and can be
863 * privately mapped
864 */
b4caecd4 865 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
866
867 /* handle PROT_EXEC implication by PROT_READ */
868 if ((prot & PROT_READ) &&
869 (current->personality & READ_IMPLIES_EXEC))
870 prot |= PROT_EXEC;
871 }
872
873 /* allow the security API to have its say */
e5467859 874 ret = security_mmap_addr(addr);
1da177e4
LT
875 if (ret < 0)
876 return ret;
877
878 /* looks okay */
879 *_capabilities = capabilities;
880 return 0;
881}
882
883/*
884 * we've determined that we can make the mapping, now translate what we
885 * now know into VMA flags
886 */
887static unsigned long determine_vm_flags(struct file *file,
888 unsigned long prot,
889 unsigned long flags,
890 unsigned long capabilities)
891{
892 unsigned long vm_flags;
893
e6bfb709 894 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1da177e4
LT
895 /* vm_flags |= mm->def_flags; */
896
b4caecd4 897 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1da177e4 898 /* attempt to share read-only copies of mapped file chunks */
3c7b2045 899 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1da177e4
LT
900 if (file && !(prot & PROT_WRITE))
901 vm_flags |= VM_MAYSHARE;
3c7b2045 902 } else {
1da177e4
LT
903 /* overlay a shareable mapping on the backing device or inode
904 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
905 * romfs/cramfs */
b4caecd4 906 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1da177e4 907 if (flags & MAP_SHARED)
3c7b2045 908 vm_flags |= VM_SHARED;
1da177e4
LT
909 }
910
911 /* refuse to let anyone share private mappings with this process if
912 * it's being traced - otherwise breakpoints set in it may interfere
913 * with another untraced process
914 */
a288eecc 915 if ((flags & MAP_PRIVATE) && current->ptrace)
1da177e4
LT
916 vm_flags &= ~VM_MAYSHARE;
917
918 return vm_flags;
919}
920
921/*
8feae131
DH
922 * set up a shared mapping on a file (the driver or filesystem provides and
923 * pins the storage)
1da177e4 924 */
8feae131 925static int do_mmap_shared_file(struct vm_area_struct *vma)
1da177e4
LT
926{
927 int ret;
928
f74ac015 929 ret = call_mmap(vma->vm_file, vma);
dd8632a1
PM
930 if (ret == 0) {
931 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 932 return 0;
dd8632a1 933 }
1da177e4
LT
934 if (ret != -ENOSYS)
935 return ret;
936
3fa30460
DH
937 /* getting -ENOSYS indicates that direct mmap isn't possible (as
938 * opposed to tried but failed) so we can only give a suitable error as
939 * it's not possible to make a private copy if MAP_SHARED was given */
1da177e4
LT
940 return -ENODEV;
941}
942
943/*
944 * set up a private mapping or an anonymous shared mapping
945 */
8feae131
DH
946static int do_mmap_private(struct vm_area_struct *vma,
947 struct vm_region *region,
645d83c5
DH
948 unsigned long len,
949 unsigned long capabilities)
1da177e4 950{
dbc8358c 951 unsigned long total, point;
1da177e4 952 void *base;
8feae131 953 int ret, order;
1da177e4
LT
954
955 /* invoke the file's mapping function so that it can keep track of
956 * shared mappings on devices or memory
957 * - VM_MAYSHARE will be set if it may attempt to share
958 */
b4caecd4 959 if (capabilities & NOMMU_MAP_DIRECT) {
f74ac015 960 ret = call_mmap(vma->vm_file, vma);
fc4f4be9
DH
961 /* shouldn't return success if we're not sharing */
962 if (WARN_ON_ONCE(!is_nommu_shared_mapping(vma->vm_flags)))
963 ret = -ENOSYS;
dd8632a1 964 if (ret == 0) {
dd8632a1 965 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 966 return 0;
1da177e4 967 }
dd8632a1
PM
968 if (ret != -ENOSYS)
969 return ret;
1da177e4
LT
970
971 /* getting an ENOSYS error indicates that direct mmap isn't
972 * possible (as opposed to tried but failed) so we'll try to
973 * make a private copy of the data and map that instead */
974 }
975
8feae131 976
1da177e4
LT
977 /* allocate some memory to hold the mapping
978 * - note that this may not return a page-aligned address if the object
979 * we're allocating is smaller than a page
980 */
f67d9b15 981 order = get_order(len);
8feae131 982 total = 1 << order;
f67d9b15 983 point = len >> PAGE_SHIFT;
dd8632a1 984
dbc8358c 985 /* we don't want to allocate a power-of-2 sized page set */
22cc877b 986 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
dbc8358c 987 total = point;
8feae131 988
da616534 989 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
dbc8358c
JK
990 if (!base)
991 goto enomem;
992
993 atomic_long_add(total, &mmap_pages_allocated);
1da177e4 994
8feae131
DH
995 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
996 region->vm_start = (unsigned long) base;
f67d9b15 997 region->vm_end = region->vm_start + len;
dd8632a1 998 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
8feae131
DH
999
1000 vma->vm_start = region->vm_start;
1001 vma->vm_end = region->vm_start + len;
1da177e4
LT
1002
1003 if (vma->vm_file) {
1004 /* read the contents of a file into the copy */
1da177e4
LT
1005 loff_t fpos;
1006
1007 fpos = vma->vm_pgoff;
1008 fpos <<= PAGE_SHIFT;
1009
b4bf802a 1010 ret = kernel_read(vma->vm_file, base, len, &fpos);
1da177e4
LT
1011 if (ret < 0)
1012 goto error_free;
1013
1014 /* clear the last little bit */
f67d9b15
BL
1015 if (ret < len)
1016 memset(base + ret, 0, len - ret);
1da177e4 1017
bfd40eaf
KS
1018 } else {
1019 vma_set_anonymous(vma);
1da177e4
LT
1020 }
1021
1022 return 0;
1023
1024error_free:
7223bb4a 1025 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
1026 region->vm_start = vma->vm_start = 0;
1027 region->vm_end = vma->vm_end = 0;
dd8632a1 1028 region->vm_top = 0;
1da177e4
LT
1029 return ret;
1030
1031enomem:
b1de0d13 1032 pr_err("Allocation of length %lu from process %d (%s) failed\n",
05ae6fa3 1033 len, current->pid, current->comm);
9af744d7 1034 show_free_areas(0, NULL);
1da177e4
LT
1035 return -ENOMEM;
1036}
1037
1038/*
1039 * handle mapping creation for uClinux
1040 */
1fcfd8db
ON
1041unsigned long do_mmap(struct file *file,
1042 unsigned long addr,
1043 unsigned long len,
1044 unsigned long prot,
1045 unsigned long flags,
1fcfd8db 1046 unsigned long pgoff,
897ab3e0
MR
1047 unsigned long *populate,
1048 struct list_head *uf)
1da177e4 1049{
8feae131
DH
1050 struct vm_area_struct *vma;
1051 struct vm_region *region;
1da177e4 1052 struct rb_node *rb;
45e55300 1053 vm_flags_t vm_flags;
1fcfd8db 1054 unsigned long capabilities, result;
1da177e4 1055 int ret;
8220543d 1056 MA_STATE(mas, &current->mm->mm_mt, 0, 0);
1da177e4 1057
41badc15 1058 *populate = 0;
bebeb3d6 1059
1da177e4
LT
1060 /* decide whether we should attempt the mapping, and if so what sort of
1061 * mapping */
1062 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1063 &capabilities);
22cc877b 1064 if (ret < 0)
1da177e4
LT
1065 return ret;
1066
06aab5a3
DH
1067 /* we ignore the address hint */
1068 addr = 0;
f67d9b15 1069 len = PAGE_ALIGN(len);
06aab5a3 1070
1da177e4
LT
1071 /* we've determined that we can make the mapping, now translate what we
1072 * now know into VMA flags */
45e55300 1073 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1da177e4 1074
8220543d 1075
8feae131
DH
1076 /* we're going to need to record the mapping */
1077 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1078 if (!region)
1079 goto error_getting_region;
1080
490fc053 1081 vma = vm_area_alloc(current->mm);
8feae131
DH
1082 if (!vma)
1083 goto error_getting_vma;
1da177e4 1084
8220543d
MWO
1085 if (mas_preallocate(&mas, vma, GFP_KERNEL))
1086 goto error_maple_preallocate;
1087
1e2ae599 1088 region->vm_usage = 1;
8feae131
DH
1089 region->vm_flags = vm_flags;
1090 region->vm_pgoff = pgoff;
1091
8feae131
DH
1092 vma->vm_flags = vm_flags;
1093 vma->vm_pgoff = pgoff;
1da177e4 1094
8feae131 1095 if (file) {
cb0942b8
AV
1096 region->vm_file = get_file(file);
1097 vma->vm_file = get_file(file);
8feae131
DH
1098 }
1099
1100 down_write(&nommu_region_sem);
1101
1102 /* if we want to share, we need to check for regions created by other
1da177e4 1103 * mmap() calls that overlap with our proposed mapping
8feae131 1104 * - we can only share with a superset match on most regular files
1da177e4
LT
1105 * - shared mappings on character devices and memory backed files are
1106 * permitted to overlap inexactly as far as we are concerned for in
1107 * these cases, sharing is handled in the driver or filesystem rather
1108 * than here
1109 */
fc4f4be9 1110 if (is_nommu_shared_mapping(vm_flags)) {
8feae131
DH
1111 struct vm_region *pregion;
1112 unsigned long pglen, rpglen, pgend, rpgend, start;
1da177e4 1113
8feae131
DH
1114 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1115 pgend = pgoff + pglen;
165b2392 1116
8feae131
DH
1117 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1118 pregion = rb_entry(rb, struct vm_region, vm_rb);
1da177e4 1119
fc4f4be9 1120 if (!is_nommu_shared_mapping(pregion->vm_flags))
1da177e4
LT
1121 continue;
1122
1123 /* search for overlapping mappings on the same file */
496ad9aa
AV
1124 if (file_inode(pregion->vm_file) !=
1125 file_inode(file))
1da177e4
LT
1126 continue;
1127
8feae131 1128 if (pregion->vm_pgoff >= pgend)
1da177e4
LT
1129 continue;
1130
8feae131
DH
1131 rpglen = pregion->vm_end - pregion->vm_start;
1132 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1133 rpgend = pregion->vm_pgoff + rpglen;
1134 if (pgoff >= rpgend)
1da177e4
LT
1135 continue;
1136
8feae131
DH
1137 /* handle inexactly overlapping matches between
1138 * mappings */
1139 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1140 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1141 /* new mapping is not a subset of the region */
b4caecd4 1142 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
1143 goto sharing_violation;
1144 continue;
1145 }
1146
8feae131 1147 /* we've found a region we can share */
1e2ae599 1148 pregion->vm_usage++;
8feae131
DH
1149 vma->vm_region = pregion;
1150 start = pregion->vm_start;
1151 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1152 vma->vm_start = start;
1153 vma->vm_end = start + len;
1154
22cc877b 1155 if (pregion->vm_flags & VM_MAPPED_COPY)
8feae131 1156 vma->vm_flags |= VM_MAPPED_COPY;
22cc877b 1157 else {
8feae131
DH
1158 ret = do_mmap_shared_file(vma);
1159 if (ret < 0) {
1160 vma->vm_region = NULL;
1161 vma->vm_start = 0;
1162 vma->vm_end = 0;
1e2ae599 1163 pregion->vm_usage--;
8feae131
DH
1164 pregion = NULL;
1165 goto error_just_free;
1166 }
1167 }
1168 fput(region->vm_file);
1169 kmem_cache_free(vm_region_jar, region);
1170 region = pregion;
1171 result = start;
1172 goto share;
1da177e4
LT
1173 }
1174
1da177e4
LT
1175 /* obtain the address at which to make a shared mapping
1176 * - this is the hook for quasi-memory character devices to
1177 * tell us the location of a shared mapping
1178 */
b4caecd4 1179 if (capabilities & NOMMU_MAP_DIRECT) {
1da177e4
LT
1180 addr = file->f_op->get_unmapped_area(file, addr, len,
1181 pgoff, flags);
bb005a59 1182 if (IS_ERR_VALUE(addr)) {
1da177e4 1183 ret = addr;
bb005a59 1184 if (ret != -ENOSYS)
8feae131 1185 goto error_just_free;
1da177e4
LT
1186
1187 /* the driver refused to tell us where to site
1188 * the mapping so we'll have to attempt to copy
1189 * it */
bb005a59 1190 ret = -ENODEV;
b4caecd4 1191 if (!(capabilities & NOMMU_MAP_COPY))
8feae131 1192 goto error_just_free;
1da177e4 1193
b4caecd4 1194 capabilities &= ~NOMMU_MAP_DIRECT;
8feae131
DH
1195 } else {
1196 vma->vm_start = region->vm_start = addr;
1197 vma->vm_end = region->vm_end = addr + len;
1da177e4
LT
1198 }
1199 }
1200 }
1201
8feae131 1202 vma->vm_region = region;
1da177e4 1203
645d83c5 1204 /* set up the mapping
b4caecd4 1205 * - the region is filled in if NOMMU_MAP_DIRECT is still set
645d83c5 1206 */
1da177e4 1207 if (file && vma->vm_flags & VM_SHARED)
8feae131 1208 ret = do_mmap_shared_file(vma);
1da177e4 1209 else
645d83c5 1210 ret = do_mmap_private(vma, region, len, capabilities);
1da177e4 1211 if (ret < 0)
645d83c5
DH
1212 goto error_just_free;
1213 add_nommu_region(region);
8feae131 1214
ea637639 1215 /* clear anonymous mappings that don't ask for uninitialized data */
0bf5f949
CH
1216 if (!vma->vm_file &&
1217 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1218 !(flags & MAP_UNINITIALIZED)))
ea637639
JZ
1219 memset((void *)region->vm_start, 0,
1220 region->vm_end - region->vm_start);
1221
1da177e4 1222 /* okay... we have a mapping; now we have to register it */
8feae131 1223 result = vma->vm_start;
1da177e4 1224
1da177e4
LT
1225 current->mm->total_vm += len >> PAGE_SHIFT;
1226
8feae131 1227share:
8220543d 1228 mas_add_vma_to_mm(&mas, current->mm, vma);
1da177e4 1229
cfe79c00
MF
1230 /* we flush the region from the icache only when the first executable
1231 * mapping of it is made */
1232 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
a75a2df6 1233 flush_icache_user_range(region->vm_start, region->vm_end);
cfe79c00
MF
1234 region->vm_icache_flushed = true;
1235 }
1da177e4 1236
cfe79c00 1237 up_write(&nommu_region_sem);
1da177e4 1238
8feae131 1239 return result;
1da177e4 1240
8feae131
DH
1241error_just_free:
1242 up_write(&nommu_region_sem);
1243error:
7f31cced 1244 mas_destroy(&mas);
89a86402
DH
1245 if (region->vm_file)
1246 fput(region->vm_file);
8feae131 1247 kmem_cache_free(vm_region_jar, region);
89a86402
DH
1248 if (vma->vm_file)
1249 fput(vma->vm_file);
3928d4f5 1250 vm_area_free(vma);
8feae131
DH
1251 return ret;
1252
1253sharing_violation:
1254 up_write(&nommu_region_sem);
22cc877b 1255 pr_warn("Attempt to share mismatched mappings\n");
8feae131
DH
1256 ret = -EINVAL;
1257 goto error;
1da177e4 1258
8feae131
DH
1259error_getting_vma:
1260 kmem_cache_free(vm_region_jar, region);
22cc877b
LR
1261 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1262 len, current->pid);
9af744d7 1263 show_free_areas(0, NULL);
1da177e4
LT
1264 return -ENOMEM;
1265
8feae131 1266error_getting_region:
22cc877b
LR
1267 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1268 len, current->pid);
9af744d7 1269 show_free_areas(0, NULL);
1da177e4 1270 return -ENOMEM;
8220543d
MWO
1271
1272error_maple_preallocate:
1273 kmem_cache_free(vm_region_jar, region);
1274 vm_area_free(vma);
1275 pr_warn("Allocation of vma tree for process %d failed\n", current->pid);
1276 show_free_areas(0, NULL);
1277 return -ENOMEM;
1278
1da177e4 1279}
6be5ceb0 1280
a90f590a
DB
1281unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1282 unsigned long prot, unsigned long flags,
1283 unsigned long fd, unsigned long pgoff)
66f0dc48
HD
1284{
1285 struct file *file = NULL;
1286 unsigned long retval = -EBADF;
1287
120a795d 1288 audit_mmap_fd(fd, flags);
66f0dc48
HD
1289 if (!(flags & MAP_ANONYMOUS)) {
1290 file = fget(fd);
1291 if (!file)
1292 goto out;
1293 }
1294
ad1ed293 1295 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
66f0dc48
HD
1296
1297 if (file)
1298 fput(file);
1299out:
1300 return retval;
1301}
1302
a90f590a
DB
1303SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1304 unsigned long, prot, unsigned long, flags,
1305 unsigned long, fd, unsigned long, pgoff)
1306{
1307 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1308}
1309
a4679373
CH
1310#ifdef __ARCH_WANT_SYS_OLD_MMAP
1311struct mmap_arg_struct {
1312 unsigned long addr;
1313 unsigned long len;
1314 unsigned long prot;
1315 unsigned long flags;
1316 unsigned long fd;
1317 unsigned long offset;
1318};
1319
1320SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1321{
1322 struct mmap_arg_struct a;
1323
1324 if (copy_from_user(&a, arg, sizeof(a)))
1325 return -EFAULT;
1824cb75 1326 if (offset_in_page(a.offset))
a4679373
CH
1327 return -EINVAL;
1328
a90f590a
DB
1329 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1330 a.offset >> PAGE_SHIFT);
a4679373
CH
1331}
1332#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1333
1da177e4 1334/*
8feae131
DH
1335 * split a vma into two pieces at address 'addr', a new vma is allocated either
1336 * for the first part or the tail.
1da177e4 1337 */
8feae131
DH
1338int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1339 unsigned long addr, int new_below)
1da177e4 1340{
8feae131
DH
1341 struct vm_area_struct *new;
1342 struct vm_region *region;
1343 unsigned long npages;
8220543d 1344 MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end);
1da177e4 1345
779c1023
DH
1346 /* we're only permitted to split anonymous regions (these should have
1347 * only a single usage on the region) */
1348 if (vma->vm_file)
8feae131 1349 return -ENOMEM;
1da177e4 1350
fd9edbdb 1351 mm = vma->vm_mm;
8feae131
DH
1352 if (mm->map_count >= sysctl_max_map_count)
1353 return -ENOMEM;
1da177e4 1354
8feae131
DH
1355 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1356 if (!region)
1357 return -ENOMEM;
1da177e4 1358
3928d4f5 1359 new = vm_area_dup(vma);
8220543d
MWO
1360 if (!new)
1361 goto err_vma_dup;
1362
1363 if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
1364 pr_warn("Allocation of vma tree for process %d failed\n",
1365 current->pid);
1366 goto err_mas_preallocate;
8feae131
DH
1367 }
1368
1369 /* most fields are the same, copy all, and then fixup */
8feae131
DH
1370 *region = *vma->vm_region;
1371 new->vm_region = region;
1372
1373 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1374
1375 if (new_below) {
dd8632a1 1376 region->vm_top = region->vm_end = new->vm_end = addr;
8feae131
DH
1377 } else {
1378 region->vm_start = new->vm_start = addr;
1379 region->vm_pgoff = new->vm_pgoff += npages;
1da177e4 1380 }
8feae131
DH
1381
1382 if (new->vm_ops && new->vm_ops->open)
1383 new->vm_ops->open(new);
1384
8feae131
DH
1385 down_write(&nommu_region_sem);
1386 delete_nommu_region(vma->vm_region);
1387 if (new_below) {
1388 vma->vm_region->vm_start = vma->vm_start = addr;
1389 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1390 } else {
1391 vma->vm_region->vm_end = vma->vm_end = addr;
dd8632a1 1392 vma->vm_region->vm_top = addr;
8feae131
DH
1393 }
1394 add_nommu_region(vma->vm_region);
1395 add_nommu_region(new->vm_region);
1396 up_write(&nommu_region_sem);
8220543d
MWO
1397
1398 setup_vma_to_mm(vma, mm);
1399 setup_vma_to_mm(new, mm);
1400 mas_set_range(&mas, vma->vm_start, vma->vm_end - 1);
1401 mas_store(&mas, vma);
1402 vma_mas_store(new, &mas);
fd9edbdb 1403 mm->map_count++;
8feae131 1404 return 0;
8220543d
MWO
1405
1406err_mas_preallocate:
1407 vm_area_free(new);
1408err_vma_dup:
1409 kmem_cache_free(vm_region_jar, region);
1410 return -ENOMEM;
1da177e4
LT
1411}
1412
3034097a 1413/*
8feae131
DH
1414 * shrink a VMA by removing the specified chunk from either the beginning or
1415 * the end
3034097a 1416 */
8feae131
DH
1417static int shrink_vma(struct mm_struct *mm,
1418 struct vm_area_struct *vma,
1419 unsigned long from, unsigned long to)
1da177e4 1420{
8feae131 1421 struct vm_region *region;
1da177e4 1422
8feae131
DH
1423 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1424 * and list */
8220543d
MWO
1425 if (delete_vma_from_mm(vma))
1426 return -ENOMEM;
8feae131
DH
1427 if (from > vma->vm_start)
1428 vma->vm_end = from;
1429 else
1430 vma->vm_start = to;
8220543d
MWO
1431 if (add_vma_to_mm(mm, vma))
1432 return -ENOMEM;
1da177e4 1433
8feae131
DH
1434 /* cut the backing region down to size */
1435 region = vma->vm_region;
1e2ae599 1436 BUG_ON(region->vm_usage != 1);
8feae131
DH
1437
1438 down_write(&nommu_region_sem);
1439 delete_nommu_region(region);
dd8632a1
PM
1440 if (from > region->vm_start) {
1441 to = region->vm_top;
1442 region->vm_top = region->vm_end = from;
1443 } else {
8feae131 1444 region->vm_start = to;
dd8632a1 1445 }
8feae131
DH
1446 add_nommu_region(region);
1447 up_write(&nommu_region_sem);
1448
1449 free_page_series(from, to);
1450 return 0;
1451}
1da177e4 1452
8feae131
DH
1453/*
1454 * release a mapping
1455 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1456 * VMA, though it need not cover the whole VMA
1457 */
897ab3e0 1458int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
8feae131 1459{
8220543d 1460 MA_STATE(mas, &mm->mm_mt, start, start);
8feae131 1461 struct vm_area_struct *vma;
f67d9b15 1462 unsigned long end;
8220543d 1463 int ret = 0;
1da177e4 1464
f67d9b15 1465 len = PAGE_ALIGN(len);
8feae131
DH
1466 if (len == 0)
1467 return -EINVAL;
365e9c87 1468
f67d9b15
BL
1469 end = start + len;
1470
8feae131 1471 /* find the first potentially overlapping VMA */
8220543d 1472 vma = mas_find(&mas, end - 1);
8feae131 1473 if (!vma) {
ac714904 1474 static int limit;
33e5d769 1475 if (limit < 5) {
22cc877b
LR
1476 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1477 current->pid, current->comm,
1478 start, start + len - 1);
33e5d769
DH
1479 limit++;
1480 }
8feae131
DH
1481 return -EINVAL;
1482 }
1da177e4 1483
8feae131
DH
1484 /* we're allowed to split an anonymous VMA but not a file-backed one */
1485 if (vma->vm_file) {
1486 do {
22cc877b 1487 if (start > vma->vm_start)
8feae131 1488 return -EINVAL;
8feae131
DH
1489 if (end == vma->vm_end)
1490 goto erase_whole_vma;
8220543d 1491 vma = mas_next(&mas, end - 1);
d75a310c 1492 } while (vma);
8feae131
DH
1493 return -EINVAL;
1494 } else {
1495 /* the chunk must be a subset of the VMA found */
1496 if (start == vma->vm_start && end == vma->vm_end)
1497 goto erase_whole_vma;
22cc877b 1498 if (start < vma->vm_start || end > vma->vm_end)
8feae131 1499 return -EINVAL;
1824cb75 1500 if (offset_in_page(start))
8feae131 1501 return -EINVAL;
1824cb75 1502 if (end != vma->vm_end && offset_in_page(end))
8feae131 1503 return -EINVAL;
8feae131
DH
1504 if (start != vma->vm_start && end != vma->vm_end) {
1505 ret = split_vma(mm, vma, start, 1);
22cc877b 1506 if (ret < 0)
8feae131 1507 return ret;
8feae131
DH
1508 }
1509 return shrink_vma(mm, vma, start, end);
1510 }
1da177e4 1511
8feae131 1512erase_whole_vma:
8220543d
MWO
1513 if (delete_vma_from_mm(vma))
1514 ret = -ENOMEM;
80be727e
LH
1515 else
1516 delete_vma(mm, vma);
8220543d 1517 return ret;
1da177e4
LT
1518}
1519
bfce281c 1520int vm_munmap(unsigned long addr, size_t len)
3034097a 1521{
bfce281c 1522 struct mm_struct *mm = current->mm;
3034097a 1523 int ret;
3034097a 1524
d8ed45c5 1525 mmap_write_lock(mm);
897ab3e0 1526 ret = do_munmap(mm, addr, len, NULL);
d8ed45c5 1527 mmap_write_unlock(mm);
3034097a
DH
1528 return ret;
1529}
a46ef99d
LT
1530EXPORT_SYMBOL(vm_munmap);
1531
1532SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1533{
bfce281c 1534 return vm_munmap(addr, len);
a46ef99d 1535}
3034097a
DH
1536
1537/*
8feae131 1538 * release all the mappings made in a process's VM space
3034097a 1539 */
8feae131 1540void exit_mmap(struct mm_struct *mm)
1da177e4 1541{
8220543d 1542 VMA_ITERATOR(vmi, mm, 0);
8feae131 1543 struct vm_area_struct *vma;
1da177e4 1544
8feae131
DH
1545 if (!mm)
1546 return;
1da177e4 1547
8feae131 1548 mm->total_vm = 0;
1da177e4 1549
8220543d
MWO
1550 /*
1551 * Lock the mm to avoid assert complaining even though this is the only
1552 * user of the mm
1553 */
1554 mmap_write_lock(mm);
1555 for_each_vma(vmi, vma) {
1556 cleanup_vma_from_mm(vma);
8feae131 1557 delete_vma(mm, vma);
04c34961 1558 cond_resched();
1da177e4 1559 }
524e00b3 1560 __mt_destroy(&mm->mm_mt);
8220543d 1561 mmap_write_unlock(mm);
1da177e4
LT
1562}
1563
5d22fc25 1564int vm_brk(unsigned long addr, unsigned long len)
1da177e4
LT
1565{
1566 return -ENOMEM;
1567}
1568
1569/*
6fa5f80b
DH
1570 * expand (or shrink) an existing mapping, potentially moving it at the same
1571 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1da177e4 1572 *
6fa5f80b 1573 * under NOMMU conditions, we only permit changing a mapping's size, and only
8feae131
DH
1574 * as long as it stays within the region allocated by do_mmap_private() and the
1575 * block is not shareable
1da177e4 1576 *
6fa5f80b 1577 * MREMAP_FIXED is not supported under NOMMU conditions
1da177e4 1578 */
4b377bab 1579static unsigned long do_mremap(unsigned long addr,
1da177e4
LT
1580 unsigned long old_len, unsigned long new_len,
1581 unsigned long flags, unsigned long new_addr)
1582{
6fa5f80b 1583 struct vm_area_struct *vma;
1da177e4
LT
1584
1585 /* insanity checks first */
f67d9b15
BL
1586 old_len = PAGE_ALIGN(old_len);
1587 new_len = PAGE_ALIGN(new_len);
8feae131 1588 if (old_len == 0 || new_len == 0)
1da177e4
LT
1589 return (unsigned long) -EINVAL;
1590
1824cb75 1591 if (offset_in_page(addr))
8feae131
DH
1592 return -EINVAL;
1593
1da177e4
LT
1594 if (flags & MREMAP_FIXED && new_addr != addr)
1595 return (unsigned long) -EINVAL;
1596
8feae131 1597 vma = find_vma_exact(current->mm, addr, old_len);
6fa5f80b
DH
1598 if (!vma)
1599 return (unsigned long) -EINVAL;
1da177e4 1600
6fa5f80b 1601 if (vma->vm_end != vma->vm_start + old_len)
1da177e4
LT
1602 return (unsigned long) -EFAULT;
1603
fc4f4be9 1604 if (is_nommu_shared_mapping(vma->vm_flags))
1da177e4
LT
1605 return (unsigned long) -EPERM;
1606
8feae131 1607 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1da177e4
LT
1608 return (unsigned long) -ENOMEM;
1609
1610 /* all checks complete - do it */
6fa5f80b 1611 vma->vm_end = vma->vm_start + new_len;
6fa5f80b
DH
1612 return vma->vm_start;
1613}
1614
6a6160a7
HC
1615SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1616 unsigned long, new_len, unsigned long, flags,
1617 unsigned long, new_addr)
6fa5f80b
DH
1618{
1619 unsigned long ret;
1620
d8ed45c5 1621 mmap_write_lock(current->mm);
6fa5f80b 1622 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
d8ed45c5 1623 mmap_write_unlock(current->mm);
6fa5f80b 1624 return ret;
1da177e4
LT
1625}
1626
df06b37f
KB
1627struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1628 unsigned int foll_flags)
1da177e4
LT
1629{
1630 return NULL;
1631}
1632
8f3b1327
BL
1633int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1634 unsigned long pfn, unsigned long size, pgprot_t prot)
1da177e4 1635{
8f3b1327
BL
1636 if (addr != (pfn << PAGE_SHIFT))
1637 return -EINVAL;
1638
314e51b9 1639 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
66aa2b4b 1640 return 0;
1da177e4 1641}
22c4af40 1642EXPORT_SYMBOL(remap_pfn_range);
1da177e4 1643
3c0b9de6
LT
1644int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1645{
1646 unsigned long pfn = start >> PAGE_SHIFT;
1647 unsigned long vm_len = vma->vm_end - vma->vm_start;
1648
1649 pfn += vma->vm_pgoff;
1650 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1651}
1652EXPORT_SYMBOL(vm_iomap_memory);
1653
f905bc44
PM
1654int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1655 unsigned long pgoff)
1656{
1657 unsigned int size = vma->vm_end - vma->vm_start;
1658
1659 if (!(vma->vm_flags & VM_USERMAP))
1660 return -EINVAL;
1661
1662 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1663 vma->vm_end = vma->vm_start + size;
1664
1665 return 0;
1666}
1667EXPORT_SYMBOL(remap_vmalloc_range);
1668
2bcd6454 1669vm_fault_t filemap_fault(struct vm_fault *vmf)
b0e15190
DH
1670{
1671 BUG();
d0217ac0 1672 return 0;
b0e15190 1673}
b5073173 1674EXPORT_SYMBOL(filemap_fault);
0ec76a11 1675
3f98a28c 1676vm_fault_t filemap_map_pages(struct vm_fault *vmf,
bae473a4 1677 pgoff_t start_pgoff, pgoff_t end_pgoff)
f1820361
KS
1678{
1679 BUG();
3f98a28c 1680 return 0;
f1820361
KS
1681}
1682EXPORT_SYMBOL(filemap_map_pages);
1683
d3f5ffca
JH
1684int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
1685 int len, unsigned int gup_flags)
0ec76a11 1686{
0ec76a11 1687 struct vm_area_struct *vma;
442486ec 1688 int write = gup_flags & FOLL_WRITE;
0ec76a11 1689
d8ed45c5 1690 if (mmap_read_lock_killable(mm))
1e426fe2 1691 return 0;
0ec76a11
DH
1692
1693 /* the access must start within one of the target process's mappings */
0159b141
DH
1694 vma = find_vma(mm, addr);
1695 if (vma) {
0ec76a11
DH
1696 /* don't overrun this mapping */
1697 if (addr + len >= vma->vm_end)
1698 len = vma->vm_end - addr;
1699
1700 /* only read or write mappings where it is permitted */
d00c7b99 1701 if (write && vma->vm_flags & VM_MAYWRITE)
7959722b
JZ
1702 copy_to_user_page(vma, NULL, addr,
1703 (void *) addr, buf, len);
d00c7b99 1704 else if (!write && vma->vm_flags & VM_MAYREAD)
7959722b
JZ
1705 copy_from_user_page(vma, NULL, addr,
1706 buf, (void *) addr, len);
0ec76a11
DH
1707 else
1708 len = 0;
1709 } else {
1710 len = 0;
1711 }
1712
d8ed45c5 1713 mmap_read_unlock(mm);
f55f199b
MF
1714
1715 return len;
1716}
1717
1718/**
b7701a5f 1719 * access_remote_vm - access another process' address space
f55f199b
MF
1720 * @mm: the mm_struct of the target address space
1721 * @addr: start address to access
1722 * @buf: source or destination buffer
1723 * @len: number of bytes to transfer
6347e8d5 1724 * @gup_flags: flags modifying lookup behaviour
f55f199b
MF
1725 *
1726 * The caller must hold a reference on @mm.
1727 */
1728int access_remote_vm(struct mm_struct *mm, unsigned long addr,
6347e8d5 1729 void *buf, int len, unsigned int gup_flags)
f55f199b 1730{
d3f5ffca 1731 return __access_remote_vm(mm, addr, buf, len, gup_flags);
f55f199b
MF
1732}
1733
1734/*
1735 * Access another process' address space.
1736 * - source/target buffer must be kernel space
1737 */
f307ab6d
LS
1738int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1739 unsigned int gup_flags)
f55f199b
MF
1740{
1741 struct mm_struct *mm;
1742
1743 if (addr + len < addr)
1744 return 0;
1745
1746 mm = get_task_mm(tsk);
1747 if (!mm)
1748 return 0;
1749
d3f5ffca 1750 len = __access_remote_vm(mm, addr, buf, len, gup_flags);
f55f199b 1751
0ec76a11
DH
1752 mmput(mm);
1753 return len;
1754}
fcd35857 1755EXPORT_SYMBOL_GPL(access_process_vm);
7e660872
DH
1756
1757/**
1758 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1759 * @inode: The inode to check
1760 * @size: The current filesize of the inode
1761 * @newsize: The proposed filesize of the inode
1762 *
1763 * Check the shared mappings on an inode on behalf of a shrinking truncate to
c08b342c
RD
1764 * make sure that any outstanding VMAs aren't broken and then shrink the
1765 * vm_regions that extend beyond so that do_mmap() doesn't
7e660872
DH
1766 * automatically grant mappings that are too large.
1767 */
1768int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1769 size_t newsize)
1770{
1771 struct vm_area_struct *vma;
7e660872
DH
1772 struct vm_region *region;
1773 pgoff_t low, high;
1774 size_t r_size, r_top;
1775
1776 low = newsize >> PAGE_SHIFT;
1777 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1778
1779 down_write(&nommu_region_sem);
1acf2e04 1780 i_mmap_lock_read(inode->i_mapping);
7e660872
DH
1781
1782 /* search for VMAs that fall within the dead zone */
6b2dbba8 1783 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
7e660872
DH
1784 /* found one - only interested if it's shared out of the page
1785 * cache */
1786 if (vma->vm_flags & VM_SHARED) {
1acf2e04 1787 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1788 up_write(&nommu_region_sem);
1789 return -ETXTBSY; /* not quite true, but near enough */
1790 }
1791 }
1792
1793 /* reduce any regions that overlap the dead zone - if in existence,
1794 * these will be pointed to by VMAs that don't overlap the dead zone
1795 *
1796 * we don't check for any regions that start beyond the EOF as there
1797 * shouldn't be any
1798 */
1acf2e04 1799 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
7e660872
DH
1800 if (!(vma->vm_flags & VM_SHARED))
1801 continue;
1802
1803 region = vma->vm_region;
1804 r_size = region->vm_top - region->vm_start;
1805 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1806
1807 if (r_top > newsize) {
1808 region->vm_top -= r_top - newsize;
1809 if (region->vm_end > region->vm_top)
1810 region->vm_end = region->vm_top;
1811 }
1812 }
1813
1acf2e04 1814 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1815 up_write(&nommu_region_sem);
1816 return 0;
1817}
c9b1d098
AS
1818
1819/*
1820 * Initialise sysctl_user_reserve_kbytes.
1821 *
1822 * This is intended to prevent a user from starting a single memory hogging
1823 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1824 * mode.
1825 *
1826 * The default value is min(3% of free memory, 128MB)
1827 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1828 */
1829static int __meminit init_user_reserve(void)
1830{
1831 unsigned long free_kbytes;
1832
c41f012a 1833 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
c9b1d098
AS
1834
1835 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1836 return 0;
1837}
a4bc6fc7 1838subsys_initcall(init_user_reserve);
4eeab4f5
AS
1839
1840/*
1841 * Initialise sysctl_admin_reserve_kbytes.
1842 *
1843 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1844 * to log in and kill a memory hogging process.
1845 *
1846 * Systems with more than 256MB will reserve 8MB, enough to recover
1847 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1848 * only reserve 3% of free pages by default.
1849 */
1850static int __meminit init_admin_reserve(void)
1851{
1852 unsigned long free_kbytes;
1853
c41f012a 1854 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
4eeab4f5
AS
1855
1856 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1857 return 0;
1858}
a4bc6fc7 1859subsys_initcall(init_admin_reserve);