MIPS: generic: Replace <linux/clk-provider.h> by <linux/of_clk.h>
[linux-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
dd0fc66f 143void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
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
a7c3e901
MH
153void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags)
154{
155 return __vmalloc(size, flags, PAGE_KERNEL);
156}
157
ed81745a 158static void *__vmalloc_user_flags(unsigned long size, gfp_t flags)
f905bc44
PM
159{
160 void *ret;
161
ed81745a 162 ret = __vmalloc(size, flags, PAGE_KERNEL);
f905bc44
PM
163 if (ret) {
164 struct vm_area_struct *vma;
165
166 down_write(&current->mm->mmap_sem);
167 vma = find_vma(current->mm, (unsigned long)ret);
168 if (vma)
169 vma->vm_flags |= VM_USERMAP;
170 up_write(&current->mm->mmap_sem);
171 }
172
173 return ret;
174}
ed81745a
AN
175
176void *vmalloc_user(unsigned long size)
177{
178 return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO);
179}
f905bc44
PM
180EXPORT_SYMBOL(vmalloc_user);
181
ed81745a
AN
182void *vmalloc_user_node_flags(unsigned long size, int node, gfp_t flags)
183{
184 return __vmalloc_user_flags(size, flags | __GFP_ZERO);
185}
186EXPORT_SYMBOL(vmalloc_user_node_flags);
187
b3bdda02 188struct page *vmalloc_to_page(const void *addr)
1da177e4
LT
189{
190 return virt_to_page(addr);
191}
b5073173 192EXPORT_SYMBOL(vmalloc_to_page);
1da177e4 193
b3bdda02 194unsigned long vmalloc_to_pfn(const void *addr)
1da177e4
LT
195{
196 return page_to_pfn(virt_to_page(addr));
197}
b5073173 198EXPORT_SYMBOL(vmalloc_to_pfn);
1da177e4
LT
199
200long vread(char *buf, char *addr, unsigned long count)
201{
9bde916b
CG
202 /* Don't allow overflow */
203 if ((unsigned long) buf + count < count)
204 count = -(unsigned long) buf;
205
1da177e4
LT
206 memcpy(buf, addr, count);
207 return count;
208}
209
210long vwrite(char *buf, char *addr, unsigned long count)
211{
212 /* Don't allow overflow */
213 if ((unsigned long) addr + count < count)
214 count = -(unsigned long) addr;
215
216 memcpy(addr, buf, count);
ac714904 217 return count;
1da177e4
LT
218}
219
220/*
e1c05067 221 * vmalloc - allocate virtually contiguous memory
1da177e4
LT
222 *
223 * @size: allocation size
224 *
225 * Allocate enough pages to cover @size from the page level
e1c05067 226 * allocator and map them into contiguous kernel virtual space.
1da177e4 227 *
c1c8897f 228 * For tight control over page level allocator and protection flags
1da177e4
LT
229 * use __vmalloc() instead.
230 */
231void *vmalloc(unsigned long size)
232{
233 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
234}
f6138882
AM
235EXPORT_SYMBOL(vmalloc);
236
e1ca7788 237/*
e1c05067 238 * vzalloc - allocate virtually contiguous memory with zero fill
e1ca7788
DY
239 *
240 * @size: allocation size
241 *
242 * Allocate enough pages to cover @size from the page level
e1c05067 243 * allocator and map them into contiguous kernel virtual space.
e1ca7788
DY
244 * The memory allocated is set to zero.
245 *
246 * For tight control over page level allocator and protection flags
247 * use __vmalloc() instead.
248 */
249void *vzalloc(unsigned long size)
250{
251 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
252 PAGE_KERNEL);
253}
254EXPORT_SYMBOL(vzalloc);
255
256/**
257 * vmalloc_node - allocate memory on a specific node
258 * @size: allocation size
259 * @node: numa node
260 *
261 * Allocate enough pages to cover @size from the page level
262 * allocator and map them into contiguous kernel virtual space.
263 *
264 * For tight control over page level allocator and protection flags
265 * use __vmalloc() instead.
266 */
f6138882
AM
267void *vmalloc_node(unsigned long size, int node)
268{
269 return vmalloc(size);
270}
9a14f653 271EXPORT_SYMBOL(vmalloc_node);
e1ca7788
DY
272
273/**
274 * vzalloc_node - allocate memory on a specific node with zero fill
275 * @size: allocation size
276 * @node: numa node
277 *
278 * Allocate enough pages to cover @size from the page level
279 * allocator and map them into contiguous kernel virtual space.
280 * The memory allocated is set to zero.
281 *
282 * For tight control over page level allocator and protection flags
283 * use __vmalloc() instead.
284 */
285void *vzalloc_node(unsigned long size, int node)
286{
287 return vzalloc(size);
288}
289EXPORT_SYMBOL(vzalloc_node);
1da177e4 290
1af446ed
PM
291/**
292 * vmalloc_exec - allocate virtually contiguous, executable memory
293 * @size: allocation size
294 *
295 * Kernel-internal function to allocate enough pages to cover @size
296 * the page level allocator and map them into contiguous and
297 * executable kernel virtual space.
298 *
299 * For tight control over page level allocator and protection flags
300 * use __vmalloc() instead.
301 */
302
303void *vmalloc_exec(unsigned long size)
304{
305 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
306}
307
b5073173
PM
308/**
309 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1da177e4
LT
310 * @size: allocation size
311 *
312 * Allocate enough 32bit PA addressable pages to cover @size from the
e1c05067 313 * page level allocator and map them into contiguous kernel virtual space.
1da177e4
LT
314 */
315void *vmalloc_32(unsigned long size)
316{
317 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
318}
b5073173
PM
319EXPORT_SYMBOL(vmalloc_32);
320
321/**
322 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
323 * @size: allocation size
324 *
325 * The resulting memory area is 32bit addressable and zeroed so it can be
326 * mapped to userspace without leaking data.
f905bc44
PM
327 *
328 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
329 * remap_vmalloc_range() are permissible.
b5073173
PM
330 */
331void *vmalloc_32_user(unsigned long size)
332{
f905bc44
PM
333 /*
334 * We'll have to sort out the ZONE_DMA bits for 64-bit,
335 * but for now this can simply use vmalloc_user() directly.
336 */
337 return vmalloc_user(size);
b5073173
PM
338}
339EXPORT_SYMBOL(vmalloc_32_user);
1da177e4
LT
340
341void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
342{
343 BUG();
344 return NULL;
345}
b5073173 346EXPORT_SYMBOL(vmap);
1da177e4 347
b3bdda02 348void vunmap(const void *addr)
1da177e4
LT
349{
350 BUG();
351}
b5073173 352EXPORT_SYMBOL(vunmap);
1da177e4 353
eb6434d9
PM
354void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
355{
356 BUG();
357 return NULL;
358}
359EXPORT_SYMBOL(vm_map_ram);
360
361void vm_unmap_ram(const void *mem, unsigned int count)
362{
363 BUG();
364}
365EXPORT_SYMBOL(vm_unmap_ram);
366
367void vm_unmap_aliases(void)
368{
369}
370EXPORT_SYMBOL_GPL(vm_unmap_aliases);
371
1eeb66a1
CH
372/*
373 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
374 * have one.
375 */
3b32123d 376void __weak vmalloc_sync_all(void)
1eeb66a1
CH
377{
378}
379
cd12909c 380struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
29c185e5
PM
381{
382 BUG();
383 return NULL;
384}
385EXPORT_SYMBOL_GPL(alloc_vm_area);
386
387void free_vm_area(struct vm_struct *area)
388{
389 BUG();
390}
391EXPORT_SYMBOL_GPL(free_vm_area);
392
b5073173
PM
393int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
394 struct page *page)
395{
396 return -EINVAL;
397}
398EXPORT_SYMBOL(vm_insert_page);
399
a667d745
SJ
400int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
401 unsigned long num)
402{
403 return -EINVAL;
404}
405EXPORT_SYMBOL(vm_map_pages);
406
407int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
408 unsigned long num)
409{
410 return -EINVAL;
411}
412EXPORT_SYMBOL(vm_map_pages_zero);
413
1da177e4
LT
414/*
415 * sys_brk() for the most part doesn't need the global kernel
416 * lock, except when an application is doing something nasty
417 * like trying to un-brk an area that has already been mapped
418 * to a regular file. in this case, the unmapping will need
419 * to invoke file system routines that need the global lock.
420 */
6a6160a7 421SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4
LT
422{
423 struct mm_struct *mm = current->mm;
424
425 if (brk < mm->start_brk || brk > mm->context.end_brk)
426 return mm->brk;
427
428 if (mm->brk == brk)
429 return mm->brk;
430
431 /*
432 * Always allow shrinking brk
433 */
434 if (brk <= mm->brk) {
435 mm->brk = brk;
436 return brk;
437 }
438
439 /*
440 * Ok, looks good - let it rip.
441 */
cfe79c00 442 flush_icache_range(mm->brk, brk);
1da177e4
LT
443 return mm->brk = brk;
444}
445
8feae131 446/*
3edf41d8 447 * initialise the percpu counter for VM and region record slabs
8feae131
DH
448 */
449void __init mmap_init(void)
1da177e4 450{
00a62ce9
KM
451 int ret;
452
908c7f19 453 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
00a62ce9 454 VM_BUG_ON(ret);
5d097056 455 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
1da177e4 456}
1da177e4 457
3034097a 458/*
8feae131
DH
459 * validate the region tree
460 * - the caller must hold the region lock
3034097a 461 */
8feae131
DH
462#ifdef CONFIG_DEBUG_NOMMU_REGIONS
463static noinline void validate_nommu_regions(void)
3034097a 464{
8feae131
DH
465 struct vm_region *region, *last;
466 struct rb_node *p, *lastp;
3034097a 467
8feae131
DH
468 lastp = rb_first(&nommu_region_tree);
469 if (!lastp)
470 return;
471
472 last = rb_entry(lastp, struct vm_region, vm_rb);
c9427bc0
GT
473 BUG_ON(last->vm_end <= last->vm_start);
474 BUG_ON(last->vm_top < last->vm_end);
8feae131
DH
475
476 while ((p = rb_next(lastp))) {
477 region = rb_entry(p, struct vm_region, vm_rb);
478 last = rb_entry(lastp, struct vm_region, vm_rb);
479
c9427bc0
GT
480 BUG_ON(region->vm_end <= region->vm_start);
481 BUG_ON(region->vm_top < region->vm_end);
482 BUG_ON(region->vm_start < last->vm_top);
3034097a 483
8feae131
DH
484 lastp = p;
485 }
3034097a 486}
8feae131 487#else
33e5d769
DH
488static void validate_nommu_regions(void)
489{
490}
8feae131 491#endif
3034097a
DH
492
493/*
8feae131 494 * add a region into the global tree
3034097a 495 */
8feae131 496static void add_nommu_region(struct vm_region *region)
3034097a 497{
8feae131
DH
498 struct vm_region *pregion;
499 struct rb_node **p, *parent;
3034097a 500
8feae131
DH
501 validate_nommu_regions();
502
8feae131
DH
503 parent = NULL;
504 p = &nommu_region_tree.rb_node;
505 while (*p) {
506 parent = *p;
507 pregion = rb_entry(parent, struct vm_region, vm_rb);
508 if (region->vm_start < pregion->vm_start)
509 p = &(*p)->rb_left;
510 else if (region->vm_start > pregion->vm_start)
511 p = &(*p)->rb_right;
512 else if (pregion == region)
513 return;
514 else
515 BUG();
3034097a
DH
516 }
517
8feae131
DH
518 rb_link_node(&region->vm_rb, parent, p);
519 rb_insert_color(&region->vm_rb, &nommu_region_tree);
3034097a 520
8feae131 521 validate_nommu_regions();
3034097a 522}
3034097a 523
930e652a 524/*
8feae131 525 * delete a region from the global tree
930e652a 526 */
8feae131 527static void delete_nommu_region(struct vm_region *region)
930e652a 528{
8feae131 529 BUG_ON(!nommu_region_tree.rb_node);
930e652a 530
8feae131
DH
531 validate_nommu_regions();
532 rb_erase(&region->vm_rb, &nommu_region_tree);
533 validate_nommu_regions();
57c8f63e
GU
534}
535
6fa5f80b 536/*
8feae131 537 * free a contiguous series of pages
6fa5f80b 538 */
8feae131 539static void free_page_series(unsigned long from, unsigned long to)
6fa5f80b 540{
8feae131
DH
541 for (; from < to; from += PAGE_SIZE) {
542 struct page *page = virt_to_page(from);
543
33e5d769 544 atomic_long_dec(&mmap_pages_allocated);
8feae131 545 put_page(page);
6fa5f80b 546 }
6fa5f80b
DH
547}
548
3034097a 549/*
8feae131 550 * release a reference to a region
33e5d769 551 * - the caller must hold the region semaphore for writing, which this releases
dd8632a1 552 * - the region may not have been added to the tree yet, in which case vm_top
8feae131 553 * will equal vm_start
3034097a 554 */
8feae131
DH
555static void __put_nommu_region(struct vm_region *region)
556 __releases(nommu_region_sem)
1da177e4 557{
8feae131 558 BUG_ON(!nommu_region_tree.rb_node);
1da177e4 559
1e2ae599 560 if (--region->vm_usage == 0) {
dd8632a1 561 if (region->vm_top > region->vm_start)
8feae131
DH
562 delete_nommu_region(region);
563 up_write(&nommu_region_sem);
564
565 if (region->vm_file)
566 fput(region->vm_file);
567
568 /* IO memory and memory shared directly out of the pagecache
569 * from ramfs/tmpfs mustn't be released here */
22cc877b 570 if (region->vm_flags & VM_MAPPED_COPY)
dd8632a1 571 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
572 kmem_cache_free(vm_region_jar, region);
573 } else {
574 up_write(&nommu_region_sem);
1da177e4 575 }
8feae131 576}
1da177e4 577
8feae131
DH
578/*
579 * release a reference to a region
580 */
581static void put_nommu_region(struct vm_region *region)
582{
583 down_write(&nommu_region_sem);
584 __put_nommu_region(region);
1da177e4
LT
585}
586
3034097a 587/*
8feae131
DH
588 * add a VMA into a process's mm_struct in the appropriate place in the list
589 * and tree and add to the address space's page tree also if not an anonymous
590 * page
591 * - should be called with mm->mmap_sem held writelocked
3034097a 592 */
8feae131 593static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 594{
6038def0 595 struct vm_area_struct *pvma, *prev;
1da177e4 596 struct address_space *mapping;
6038def0 597 struct rb_node **p, *parent, *rb_prev;
8feae131 598
8feae131
DH
599 BUG_ON(!vma->vm_region);
600
601 mm->map_count++;
602 vma->vm_mm = mm;
1da177e4
LT
603
604 /* add the VMA to the mapping */
605 if (vma->vm_file) {
606 mapping = vma->vm_file->f_mapping;
607
83cde9e8 608 i_mmap_lock_write(mapping);
1da177e4 609 flush_dcache_mmap_lock(mapping);
6b2dbba8 610 vma_interval_tree_insert(vma, &mapping->i_mmap);
1da177e4 611 flush_dcache_mmap_unlock(mapping);
83cde9e8 612 i_mmap_unlock_write(mapping);
1da177e4
LT
613 }
614
8feae131 615 /* add the VMA to the tree */
6038def0 616 parent = rb_prev = NULL;
8feae131 617 p = &mm->mm_rb.rb_node;
1da177e4
LT
618 while (*p) {
619 parent = *p;
620 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
621
8feae131
DH
622 /* sort by: start addr, end addr, VMA struct addr in that order
623 * (the latter is necessary as we may get identical VMAs) */
624 if (vma->vm_start < pvma->vm_start)
1da177e4 625 p = &(*p)->rb_left;
6038def0
NK
626 else if (vma->vm_start > pvma->vm_start) {
627 rb_prev = parent;
1da177e4 628 p = &(*p)->rb_right;
6038def0 629 } else if (vma->vm_end < pvma->vm_end)
8feae131 630 p = &(*p)->rb_left;
6038def0
NK
631 else if (vma->vm_end > pvma->vm_end) {
632 rb_prev = parent;
8feae131 633 p = &(*p)->rb_right;
6038def0 634 } else if (vma < pvma)
8feae131 635 p = &(*p)->rb_left;
6038def0
NK
636 else if (vma > pvma) {
637 rb_prev = parent;
8feae131 638 p = &(*p)->rb_right;
6038def0 639 } else
8feae131 640 BUG();
1da177e4
LT
641 }
642
643 rb_link_node(&vma->vm_rb, parent, p);
8feae131
DH
644 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
645
646 /* add VMA to the VMA list also */
6038def0
NK
647 prev = NULL;
648 if (rb_prev)
649 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
8feae131 650
aba6dfb7 651 __vma_link_list(mm, vma, prev);
1da177e4
LT
652}
653
3034097a 654/*
8feae131 655 * delete a VMA from its owning mm_struct and address space
3034097a 656 */
8feae131 657static void delete_vma_from_mm(struct vm_area_struct *vma)
1da177e4 658{
615d6e87 659 int i;
1da177e4 660 struct address_space *mapping;
8feae131 661 struct mm_struct *mm = vma->vm_mm;
615d6e87 662 struct task_struct *curr = current;
8feae131 663
8feae131 664 mm->map_count--;
615d6e87
DB
665 for (i = 0; i < VMACACHE_SIZE; i++) {
666 /* if the vma is cached, invalidate the entire cache */
314ff785 667 if (curr->vmacache.vmas[i] == vma) {
e020d5bd 668 vmacache_invalidate(mm);
615d6e87
DB
669 break;
670 }
671 }
1da177e4
LT
672
673 /* remove the VMA from the mapping */
674 if (vma->vm_file) {
675 mapping = vma->vm_file->f_mapping;
676
83cde9e8 677 i_mmap_lock_write(mapping);
1da177e4 678 flush_dcache_mmap_lock(mapping);
6b2dbba8 679 vma_interval_tree_remove(vma, &mapping->i_mmap);
1da177e4 680 flush_dcache_mmap_unlock(mapping);
83cde9e8 681 i_mmap_unlock_write(mapping);
1da177e4
LT
682 }
683
8feae131
DH
684 /* remove from the MM's tree and list */
685 rb_erase(&vma->vm_rb, &mm->mm_rb);
b951bf2c 686
1b9fc5b2 687 __vma_unlink_list(mm, vma);
8feae131
DH
688}
689
690/*
691 * destroy a VMA record
692 */
693static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
694{
8feae131
DH
695 if (vma->vm_ops && vma->vm_ops->close)
696 vma->vm_ops->close(vma);
e9714acf 697 if (vma->vm_file)
8feae131 698 fput(vma->vm_file);
8feae131 699 put_nommu_region(vma->vm_region);
3928d4f5 700 vm_area_free(vma);
8feae131
DH
701}
702
703/*
704 * look up the first VMA in which addr resides, NULL if none
705 * - should be called with mm->mmap_sem at least held readlocked
706 */
707struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
708{
709 struct vm_area_struct *vma;
8feae131
DH
710
711 /* check the cache first */
615d6e87
DB
712 vma = vmacache_find(mm, addr);
713 if (likely(vma))
8feae131
DH
714 return vma;
715
e922c4c5 716 /* trawl the list (there may be multiple mappings in which addr
8feae131 717 * resides) */
e922c4c5 718 for (vma = mm->mmap; vma; vma = vma->vm_next) {
8feae131
DH
719 if (vma->vm_start > addr)
720 return NULL;
721 if (vma->vm_end > addr) {
615d6e87 722 vmacache_update(addr, vma);
8feae131
DH
723 return vma;
724 }
725 }
726
727 return NULL;
728}
729EXPORT_SYMBOL(find_vma);
730
731/*
732 * find a VMA
733 * - we don't extend stack VMAs under NOMMU conditions
734 */
735struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
736{
7561e8ca 737 return find_vma(mm, addr);
8feae131
DH
738}
739
740/*
741 * expand a stack to a given address
742 * - not supported under NOMMU conditions
743 */
744int expand_stack(struct vm_area_struct *vma, unsigned long address)
745{
746 return -ENOMEM;
747}
748
749/*
750 * look up the first VMA exactly that exactly matches addr
751 * - should be called with mm->mmap_sem at least held readlocked
752 */
753static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
754 unsigned long addr,
755 unsigned long len)
756{
757 struct vm_area_struct *vma;
8feae131
DH
758 unsigned long end = addr + len;
759
760 /* check the cache first */
615d6e87
DB
761 vma = vmacache_find_exact(mm, addr, end);
762 if (vma)
8feae131
DH
763 return vma;
764
e922c4c5 765 /* trawl the list (there may be multiple mappings in which addr
8feae131 766 * resides) */
e922c4c5 767 for (vma = mm->mmap; vma; vma = vma->vm_next) {
8feae131
DH
768 if (vma->vm_start < addr)
769 continue;
770 if (vma->vm_start > addr)
771 return NULL;
772 if (vma->vm_end == end) {
615d6e87 773 vmacache_update(addr, vma);
8feae131
DH
774 return vma;
775 }
776 }
777
778 return NULL;
1da177e4
LT
779}
780
781/*
782 * determine whether a mapping should be permitted and, if so, what sort of
783 * mapping we're capable of supporting
784 */
785static int validate_mmap_request(struct file *file,
786 unsigned long addr,
787 unsigned long len,
788 unsigned long prot,
789 unsigned long flags,
790 unsigned long pgoff,
791 unsigned long *_capabilities)
792{
8feae131 793 unsigned long capabilities, rlen;
1da177e4
LT
794 int ret;
795
796 /* do the simple checks first */
22cc877b 797 if (flags & MAP_FIXED)
1da177e4 798 return -EINVAL;
1da177e4
LT
799
800 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
801 (flags & MAP_TYPE) != MAP_SHARED)
802 return -EINVAL;
803
f81cff0d 804 if (!len)
1da177e4
LT
805 return -EINVAL;
806
f81cff0d 807 /* Careful about overflows.. */
8feae131
DH
808 rlen = PAGE_ALIGN(len);
809 if (!rlen || rlen > TASK_SIZE)
f81cff0d
MF
810 return -ENOMEM;
811
1da177e4 812 /* offset overflow? */
8feae131 813 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
f81cff0d 814 return -EOVERFLOW;
1da177e4
LT
815
816 if (file) {
1da177e4 817 /* files must support mmap */
72c2d531 818 if (!file->f_op->mmap)
1da177e4
LT
819 return -ENODEV;
820
821 /* work out if what we've got could possibly be shared
822 * - we support chardevs that provide their own "memory"
823 * - we support files/blockdevs that are memory backed
824 */
b4caecd4
CH
825 if (file->f_op->mmap_capabilities) {
826 capabilities = file->f_op->mmap_capabilities(file);
827 } else {
1da177e4
LT
828 /* no explicit capabilities set, so assume some
829 * defaults */
496ad9aa 830 switch (file_inode(file)->i_mode & S_IFMT) {
1da177e4
LT
831 case S_IFREG:
832 case S_IFBLK:
b4caecd4 833 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
834 break;
835
836 case S_IFCHR:
837 capabilities =
b4caecd4
CH
838 NOMMU_MAP_DIRECT |
839 NOMMU_MAP_READ |
840 NOMMU_MAP_WRITE;
1da177e4
LT
841 break;
842
843 default:
844 return -EINVAL;
845 }
846 }
847
848 /* eliminate any capabilities that we can't support on this
849 * device */
850 if (!file->f_op->get_unmapped_area)
b4caecd4 851 capabilities &= ~NOMMU_MAP_DIRECT;
6e242a1c 852 if (!(file->f_mode & FMODE_CAN_READ))
b4caecd4 853 capabilities &= ~NOMMU_MAP_COPY;
1da177e4 854
28d7a6ae
GY
855 /* The file shall have been opened with read permission. */
856 if (!(file->f_mode & FMODE_READ))
857 return -EACCES;
858
1da177e4
LT
859 if (flags & MAP_SHARED) {
860 /* do checks for writing, appending and locking */
861 if ((prot & PROT_WRITE) &&
862 !(file->f_mode & FMODE_WRITE))
863 return -EACCES;
864
496ad9aa 865 if (IS_APPEND(file_inode(file)) &&
1da177e4
LT
866 (file->f_mode & FMODE_WRITE))
867 return -EACCES;
868
d7a06983 869 if (locks_verify_locked(file))
1da177e4
LT
870 return -EAGAIN;
871
b4caecd4 872 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
873 return -ENODEV;
874
1da177e4 875 /* we mustn't privatise shared mappings */
b4caecd4 876 capabilities &= ~NOMMU_MAP_COPY;
ac714904 877 } else {
1da177e4
LT
878 /* we're going to read the file into private memory we
879 * allocate */
b4caecd4 880 if (!(capabilities & NOMMU_MAP_COPY))
1da177e4
LT
881 return -ENODEV;
882
883 /* we don't permit a private writable mapping to be
884 * shared with the backing device */
885 if (prot & PROT_WRITE)
b4caecd4 886 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4
LT
887 }
888
b4caecd4
CH
889 if (capabilities & NOMMU_MAP_DIRECT) {
890 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
891 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
892 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
3c7b2045 893 ) {
b4caecd4 894 capabilities &= ~NOMMU_MAP_DIRECT;
3c7b2045 895 if (flags & MAP_SHARED) {
22cc877b 896 pr_warn("MAP_SHARED not completely supported on !MMU\n");
3c7b2045
BS
897 return -EINVAL;
898 }
899 }
900 }
901
1da177e4
LT
902 /* handle executable mappings and implied executable
903 * mappings */
90f8572b 904 if (path_noexec(&file->f_path)) {
1da177e4
LT
905 if (prot & PROT_EXEC)
906 return -EPERM;
ac714904 907 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1da177e4
LT
908 /* handle implication of PROT_EXEC by PROT_READ */
909 if (current->personality & READ_IMPLIES_EXEC) {
b4caecd4 910 if (capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
911 prot |= PROT_EXEC;
912 }
ac714904 913 } else if ((prot & PROT_READ) &&
1da177e4 914 (prot & PROT_EXEC) &&
b4caecd4 915 !(capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
916 ) {
917 /* backing file is not executable, try to copy */
b4caecd4 918 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4 919 }
ac714904 920 } else {
1da177e4
LT
921 /* anonymous mappings are always memory backed and can be
922 * privately mapped
923 */
b4caecd4 924 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
925
926 /* handle PROT_EXEC implication by PROT_READ */
927 if ((prot & PROT_READ) &&
928 (current->personality & READ_IMPLIES_EXEC))
929 prot |= PROT_EXEC;
930 }
931
932 /* allow the security API to have its say */
e5467859 933 ret = security_mmap_addr(addr);
1da177e4
LT
934 if (ret < 0)
935 return ret;
936
937 /* looks okay */
938 *_capabilities = capabilities;
939 return 0;
940}
941
942/*
943 * we've determined that we can make the mapping, now translate what we
944 * now know into VMA flags
945 */
946static unsigned long determine_vm_flags(struct file *file,
947 unsigned long prot,
948 unsigned long flags,
949 unsigned long capabilities)
950{
951 unsigned long vm_flags;
952
e6bfb709 953 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1da177e4
LT
954 /* vm_flags |= mm->def_flags; */
955
b4caecd4 956 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1da177e4 957 /* attempt to share read-only copies of mapped file chunks */
3c7b2045 958 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1da177e4
LT
959 if (file && !(prot & PROT_WRITE))
960 vm_flags |= VM_MAYSHARE;
3c7b2045 961 } else {
1da177e4
LT
962 /* overlay a shareable mapping on the backing device or inode
963 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
964 * romfs/cramfs */
b4caecd4 965 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1da177e4 966 if (flags & MAP_SHARED)
3c7b2045 967 vm_flags |= VM_SHARED;
1da177e4
LT
968 }
969
970 /* refuse to let anyone share private mappings with this process if
971 * it's being traced - otherwise breakpoints set in it may interfere
972 * with another untraced process
973 */
a288eecc 974 if ((flags & MAP_PRIVATE) && current->ptrace)
1da177e4
LT
975 vm_flags &= ~VM_MAYSHARE;
976
977 return vm_flags;
978}
979
980/*
8feae131
DH
981 * set up a shared mapping on a file (the driver or filesystem provides and
982 * pins the storage)
1da177e4 983 */
8feae131 984static int do_mmap_shared_file(struct vm_area_struct *vma)
1da177e4
LT
985{
986 int ret;
987
f74ac015 988 ret = call_mmap(vma->vm_file, vma);
dd8632a1
PM
989 if (ret == 0) {
990 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 991 return 0;
dd8632a1 992 }
1da177e4
LT
993 if (ret != -ENOSYS)
994 return ret;
995
3fa30460
DH
996 /* getting -ENOSYS indicates that direct mmap isn't possible (as
997 * opposed to tried but failed) so we can only give a suitable error as
998 * it's not possible to make a private copy if MAP_SHARED was given */
1da177e4
LT
999 return -ENODEV;
1000}
1001
1002/*
1003 * set up a private mapping or an anonymous shared mapping
1004 */
8feae131
DH
1005static int do_mmap_private(struct vm_area_struct *vma,
1006 struct vm_region *region,
645d83c5
DH
1007 unsigned long len,
1008 unsigned long capabilities)
1da177e4 1009{
dbc8358c 1010 unsigned long total, point;
1da177e4 1011 void *base;
8feae131 1012 int ret, order;
1da177e4
LT
1013
1014 /* invoke the file's mapping function so that it can keep track of
1015 * shared mappings on devices or memory
1016 * - VM_MAYSHARE will be set if it may attempt to share
1017 */
b4caecd4 1018 if (capabilities & NOMMU_MAP_DIRECT) {
f74ac015 1019 ret = call_mmap(vma->vm_file, vma);
dd8632a1 1020 if (ret == 0) {
1da177e4 1021 /* shouldn't return success if we're not sharing */
dd8632a1
PM
1022 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1023 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 1024 return 0;
1da177e4 1025 }
dd8632a1
PM
1026 if (ret != -ENOSYS)
1027 return ret;
1da177e4
LT
1028
1029 /* getting an ENOSYS error indicates that direct mmap isn't
1030 * possible (as opposed to tried but failed) so we'll try to
1031 * make a private copy of the data and map that instead */
1032 }
1033
8feae131 1034
1da177e4
LT
1035 /* allocate some memory to hold the mapping
1036 * - note that this may not return a page-aligned address if the object
1037 * we're allocating is smaller than a page
1038 */
f67d9b15 1039 order = get_order(len);
8feae131 1040 total = 1 << order;
f67d9b15 1041 point = len >> PAGE_SHIFT;
dd8632a1 1042
dbc8358c 1043 /* we don't want to allocate a power-of-2 sized page set */
22cc877b 1044 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
dbc8358c 1045 total = point;
8feae131 1046
da616534 1047 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
dbc8358c
JK
1048 if (!base)
1049 goto enomem;
1050
1051 atomic_long_add(total, &mmap_pages_allocated);
1da177e4 1052
8feae131
DH
1053 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1054 region->vm_start = (unsigned long) base;
f67d9b15 1055 region->vm_end = region->vm_start + len;
dd8632a1 1056 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
8feae131
DH
1057
1058 vma->vm_start = region->vm_start;
1059 vma->vm_end = region->vm_start + len;
1da177e4
LT
1060
1061 if (vma->vm_file) {
1062 /* read the contents of a file into the copy */
1da177e4
LT
1063 loff_t fpos;
1064
1065 fpos = vma->vm_pgoff;
1066 fpos <<= PAGE_SHIFT;
1067
b4bf802a 1068 ret = kernel_read(vma->vm_file, base, len, &fpos);
1da177e4
LT
1069 if (ret < 0)
1070 goto error_free;
1071
1072 /* clear the last little bit */
f67d9b15
BL
1073 if (ret < len)
1074 memset(base + ret, 0, len - ret);
1da177e4 1075
bfd40eaf
KS
1076 } else {
1077 vma_set_anonymous(vma);
1da177e4
LT
1078 }
1079
1080 return 0;
1081
1082error_free:
7223bb4a 1083 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
1084 region->vm_start = vma->vm_start = 0;
1085 region->vm_end = vma->vm_end = 0;
dd8632a1 1086 region->vm_top = 0;
1da177e4
LT
1087 return ret;
1088
1089enomem:
b1de0d13 1090 pr_err("Allocation of length %lu from process %d (%s) failed\n",
05ae6fa3 1091 len, current->pid, current->comm);
9af744d7 1092 show_free_areas(0, NULL);
1da177e4
LT
1093 return -ENOMEM;
1094}
1095
1096/*
1097 * handle mapping creation for uClinux
1098 */
1fcfd8db
ON
1099unsigned long do_mmap(struct file *file,
1100 unsigned long addr,
1101 unsigned long len,
1102 unsigned long prot,
1103 unsigned long flags,
1104 vm_flags_t vm_flags,
1105 unsigned long pgoff,
897ab3e0
MR
1106 unsigned long *populate,
1107 struct list_head *uf)
1da177e4 1108{
8feae131
DH
1109 struct vm_area_struct *vma;
1110 struct vm_region *region;
1da177e4 1111 struct rb_node *rb;
1fcfd8db 1112 unsigned long capabilities, result;
1da177e4
LT
1113 int ret;
1114
41badc15 1115 *populate = 0;
bebeb3d6 1116
1da177e4
LT
1117 /* decide whether we should attempt the mapping, and if so what sort of
1118 * mapping */
1119 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1120 &capabilities);
22cc877b 1121 if (ret < 0)
1da177e4
LT
1122 return ret;
1123
06aab5a3
DH
1124 /* we ignore the address hint */
1125 addr = 0;
f67d9b15 1126 len = PAGE_ALIGN(len);
06aab5a3 1127
1da177e4
LT
1128 /* we've determined that we can make the mapping, now translate what we
1129 * now know into VMA flags */
1fcfd8db 1130 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1da177e4 1131
8feae131
DH
1132 /* we're going to need to record the mapping */
1133 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1134 if (!region)
1135 goto error_getting_region;
1136
490fc053 1137 vma = vm_area_alloc(current->mm);
8feae131
DH
1138 if (!vma)
1139 goto error_getting_vma;
1da177e4 1140
1e2ae599 1141 region->vm_usage = 1;
8feae131
DH
1142 region->vm_flags = vm_flags;
1143 region->vm_pgoff = pgoff;
1144
8feae131
DH
1145 vma->vm_flags = vm_flags;
1146 vma->vm_pgoff = pgoff;
1da177e4 1147
8feae131 1148 if (file) {
cb0942b8
AV
1149 region->vm_file = get_file(file);
1150 vma->vm_file = get_file(file);
8feae131
DH
1151 }
1152
1153 down_write(&nommu_region_sem);
1154
1155 /* if we want to share, we need to check for regions created by other
1da177e4 1156 * mmap() calls that overlap with our proposed mapping
8feae131 1157 * - we can only share with a superset match on most regular files
1da177e4
LT
1158 * - shared mappings on character devices and memory backed files are
1159 * permitted to overlap inexactly as far as we are concerned for in
1160 * these cases, sharing is handled in the driver or filesystem rather
1161 * than here
1162 */
1163 if (vm_flags & VM_MAYSHARE) {
8feae131
DH
1164 struct vm_region *pregion;
1165 unsigned long pglen, rpglen, pgend, rpgend, start;
1da177e4 1166
8feae131
DH
1167 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1168 pgend = pgoff + pglen;
165b2392 1169
8feae131
DH
1170 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1171 pregion = rb_entry(rb, struct vm_region, vm_rb);
1da177e4 1172
8feae131 1173 if (!(pregion->vm_flags & VM_MAYSHARE))
1da177e4
LT
1174 continue;
1175
1176 /* search for overlapping mappings on the same file */
496ad9aa
AV
1177 if (file_inode(pregion->vm_file) !=
1178 file_inode(file))
1da177e4
LT
1179 continue;
1180
8feae131 1181 if (pregion->vm_pgoff >= pgend)
1da177e4
LT
1182 continue;
1183
8feae131
DH
1184 rpglen = pregion->vm_end - pregion->vm_start;
1185 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1186 rpgend = pregion->vm_pgoff + rpglen;
1187 if (pgoff >= rpgend)
1da177e4
LT
1188 continue;
1189
8feae131
DH
1190 /* handle inexactly overlapping matches between
1191 * mappings */
1192 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1193 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1194 /* new mapping is not a subset of the region */
b4caecd4 1195 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
1196 goto sharing_violation;
1197 continue;
1198 }
1199
8feae131 1200 /* we've found a region we can share */
1e2ae599 1201 pregion->vm_usage++;
8feae131
DH
1202 vma->vm_region = pregion;
1203 start = pregion->vm_start;
1204 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1205 vma->vm_start = start;
1206 vma->vm_end = start + len;
1207
22cc877b 1208 if (pregion->vm_flags & VM_MAPPED_COPY)
8feae131 1209 vma->vm_flags |= VM_MAPPED_COPY;
22cc877b 1210 else {
8feae131
DH
1211 ret = do_mmap_shared_file(vma);
1212 if (ret < 0) {
1213 vma->vm_region = NULL;
1214 vma->vm_start = 0;
1215 vma->vm_end = 0;
1e2ae599 1216 pregion->vm_usage--;
8feae131
DH
1217 pregion = NULL;
1218 goto error_just_free;
1219 }
1220 }
1221 fput(region->vm_file);
1222 kmem_cache_free(vm_region_jar, region);
1223 region = pregion;
1224 result = start;
1225 goto share;
1da177e4
LT
1226 }
1227
1da177e4
LT
1228 /* obtain the address at which to make a shared mapping
1229 * - this is the hook for quasi-memory character devices to
1230 * tell us the location of a shared mapping
1231 */
b4caecd4 1232 if (capabilities & NOMMU_MAP_DIRECT) {
1da177e4
LT
1233 addr = file->f_op->get_unmapped_area(file, addr, len,
1234 pgoff, flags);
bb005a59 1235 if (IS_ERR_VALUE(addr)) {
1da177e4 1236 ret = addr;
bb005a59 1237 if (ret != -ENOSYS)
8feae131 1238 goto error_just_free;
1da177e4
LT
1239
1240 /* the driver refused to tell us where to site
1241 * the mapping so we'll have to attempt to copy
1242 * it */
bb005a59 1243 ret = -ENODEV;
b4caecd4 1244 if (!(capabilities & NOMMU_MAP_COPY))
8feae131 1245 goto error_just_free;
1da177e4 1246
b4caecd4 1247 capabilities &= ~NOMMU_MAP_DIRECT;
8feae131
DH
1248 } else {
1249 vma->vm_start = region->vm_start = addr;
1250 vma->vm_end = region->vm_end = addr + len;
1da177e4
LT
1251 }
1252 }
1253 }
1254
8feae131 1255 vma->vm_region = region;
1da177e4 1256
645d83c5 1257 /* set up the mapping
b4caecd4 1258 * - the region is filled in if NOMMU_MAP_DIRECT is still set
645d83c5 1259 */
1da177e4 1260 if (file && vma->vm_flags & VM_SHARED)
8feae131 1261 ret = do_mmap_shared_file(vma);
1da177e4 1262 else
645d83c5 1263 ret = do_mmap_private(vma, region, len, capabilities);
1da177e4 1264 if (ret < 0)
645d83c5
DH
1265 goto error_just_free;
1266 add_nommu_region(region);
8feae131 1267
ea637639 1268 /* clear anonymous mappings that don't ask for uninitialized data */
0bf5f949
CH
1269 if (!vma->vm_file &&
1270 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) ||
1271 !(flags & MAP_UNINITIALIZED)))
ea637639
JZ
1272 memset((void *)region->vm_start, 0,
1273 region->vm_end - region->vm_start);
1274
1da177e4 1275 /* okay... we have a mapping; now we have to register it */
8feae131 1276 result = vma->vm_start;
1da177e4 1277
1da177e4
LT
1278 current->mm->total_vm += len >> PAGE_SHIFT;
1279
8feae131
DH
1280share:
1281 add_vma_to_mm(current->mm, vma);
1da177e4 1282
cfe79c00
MF
1283 /* we flush the region from the icache only when the first executable
1284 * mapping of it is made */
1285 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1286 flush_icache_range(region->vm_start, region->vm_end);
1287 region->vm_icache_flushed = true;
1288 }
1da177e4 1289
cfe79c00 1290 up_write(&nommu_region_sem);
1da177e4 1291
8feae131 1292 return result;
1da177e4 1293
8feae131
DH
1294error_just_free:
1295 up_write(&nommu_region_sem);
1296error:
89a86402
DH
1297 if (region->vm_file)
1298 fput(region->vm_file);
8feae131 1299 kmem_cache_free(vm_region_jar, region);
89a86402
DH
1300 if (vma->vm_file)
1301 fput(vma->vm_file);
3928d4f5 1302 vm_area_free(vma);
8feae131
DH
1303 return ret;
1304
1305sharing_violation:
1306 up_write(&nommu_region_sem);
22cc877b 1307 pr_warn("Attempt to share mismatched mappings\n");
8feae131
DH
1308 ret = -EINVAL;
1309 goto error;
1da177e4 1310
8feae131
DH
1311error_getting_vma:
1312 kmem_cache_free(vm_region_jar, region);
22cc877b
LR
1313 pr_warn("Allocation of vma 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
8feae131 1318error_getting_region:
22cc877b
LR
1319 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1320 len, current->pid);
9af744d7 1321 show_free_areas(0, NULL);
1da177e4
LT
1322 return -ENOMEM;
1323}
6be5ceb0 1324
a90f590a
DB
1325unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1326 unsigned long prot, unsigned long flags,
1327 unsigned long fd, unsigned long pgoff)
66f0dc48
HD
1328{
1329 struct file *file = NULL;
1330 unsigned long retval = -EBADF;
1331
120a795d 1332 audit_mmap_fd(fd, flags);
66f0dc48
HD
1333 if (!(flags & MAP_ANONYMOUS)) {
1334 file = fget(fd);
1335 if (!file)
1336 goto out;
1337 }
1338
1339 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1340
ad1ed293 1341 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
66f0dc48
HD
1342
1343 if (file)
1344 fput(file);
1345out:
1346 return retval;
1347}
1348
a90f590a
DB
1349SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1350 unsigned long, prot, unsigned long, flags,
1351 unsigned long, fd, unsigned long, pgoff)
1352{
1353 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1354}
1355
a4679373
CH
1356#ifdef __ARCH_WANT_SYS_OLD_MMAP
1357struct mmap_arg_struct {
1358 unsigned long addr;
1359 unsigned long len;
1360 unsigned long prot;
1361 unsigned long flags;
1362 unsigned long fd;
1363 unsigned long offset;
1364};
1365
1366SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1367{
1368 struct mmap_arg_struct a;
1369
1370 if (copy_from_user(&a, arg, sizeof(a)))
1371 return -EFAULT;
1824cb75 1372 if (offset_in_page(a.offset))
a4679373
CH
1373 return -EINVAL;
1374
a90f590a
DB
1375 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1376 a.offset >> PAGE_SHIFT);
a4679373
CH
1377}
1378#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1379
1da177e4 1380/*
8feae131
DH
1381 * split a vma into two pieces at address 'addr', a new vma is allocated either
1382 * for the first part or the tail.
1da177e4 1383 */
8feae131
DH
1384int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1385 unsigned long addr, int new_below)
1da177e4 1386{
8feae131
DH
1387 struct vm_area_struct *new;
1388 struct vm_region *region;
1389 unsigned long npages;
1da177e4 1390
779c1023
DH
1391 /* we're only permitted to split anonymous regions (these should have
1392 * only a single usage on the region) */
1393 if (vma->vm_file)
8feae131 1394 return -ENOMEM;
1da177e4 1395
8feae131
DH
1396 if (mm->map_count >= sysctl_max_map_count)
1397 return -ENOMEM;
1da177e4 1398
8feae131
DH
1399 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1400 if (!region)
1401 return -ENOMEM;
1da177e4 1402
3928d4f5 1403 new = vm_area_dup(vma);
8feae131
DH
1404 if (!new) {
1405 kmem_cache_free(vm_region_jar, region);
1406 return -ENOMEM;
1407 }
1408
1409 /* most fields are the same, copy all, and then fixup */
8feae131
DH
1410 *region = *vma->vm_region;
1411 new->vm_region = region;
1412
1413 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1414
1415 if (new_below) {
dd8632a1 1416 region->vm_top = region->vm_end = new->vm_end = addr;
8feae131
DH
1417 } else {
1418 region->vm_start = new->vm_start = addr;
1419 region->vm_pgoff = new->vm_pgoff += npages;
1da177e4 1420 }
8feae131
DH
1421
1422 if (new->vm_ops && new->vm_ops->open)
1423 new->vm_ops->open(new);
1424
1425 delete_vma_from_mm(vma);
1426 down_write(&nommu_region_sem);
1427 delete_nommu_region(vma->vm_region);
1428 if (new_below) {
1429 vma->vm_region->vm_start = vma->vm_start = addr;
1430 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1431 } else {
1432 vma->vm_region->vm_end = vma->vm_end = addr;
dd8632a1 1433 vma->vm_region->vm_top = addr;
8feae131
DH
1434 }
1435 add_nommu_region(vma->vm_region);
1436 add_nommu_region(new->vm_region);
1437 up_write(&nommu_region_sem);
1438 add_vma_to_mm(mm, vma);
1439 add_vma_to_mm(mm, new);
1440 return 0;
1da177e4
LT
1441}
1442
3034097a 1443/*
8feae131
DH
1444 * shrink a VMA by removing the specified chunk from either the beginning or
1445 * the end
3034097a 1446 */
8feae131
DH
1447static int shrink_vma(struct mm_struct *mm,
1448 struct vm_area_struct *vma,
1449 unsigned long from, unsigned long to)
1da177e4 1450{
8feae131 1451 struct vm_region *region;
1da177e4 1452
8feae131
DH
1453 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1454 * and list */
1455 delete_vma_from_mm(vma);
1456 if (from > vma->vm_start)
1457 vma->vm_end = from;
1458 else
1459 vma->vm_start = to;
1460 add_vma_to_mm(mm, vma);
1da177e4 1461
8feae131
DH
1462 /* cut the backing region down to size */
1463 region = vma->vm_region;
1e2ae599 1464 BUG_ON(region->vm_usage != 1);
8feae131
DH
1465
1466 down_write(&nommu_region_sem);
1467 delete_nommu_region(region);
dd8632a1
PM
1468 if (from > region->vm_start) {
1469 to = region->vm_top;
1470 region->vm_top = region->vm_end = from;
1471 } else {
8feae131 1472 region->vm_start = to;
dd8632a1 1473 }
8feae131
DH
1474 add_nommu_region(region);
1475 up_write(&nommu_region_sem);
1476
1477 free_page_series(from, to);
1478 return 0;
1479}
1da177e4 1480
8feae131
DH
1481/*
1482 * release a mapping
1483 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1484 * VMA, though it need not cover the whole VMA
1485 */
897ab3e0 1486int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf)
8feae131
DH
1487{
1488 struct vm_area_struct *vma;
f67d9b15 1489 unsigned long end;
8feae131 1490 int ret;
1da177e4 1491
f67d9b15 1492 len = PAGE_ALIGN(len);
8feae131
DH
1493 if (len == 0)
1494 return -EINVAL;
365e9c87 1495
f67d9b15
BL
1496 end = start + len;
1497
8feae131
DH
1498 /* find the first potentially overlapping VMA */
1499 vma = find_vma(mm, start);
1500 if (!vma) {
ac714904 1501 static int limit;
33e5d769 1502 if (limit < 5) {
22cc877b
LR
1503 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1504 current->pid, current->comm,
1505 start, start + len - 1);
33e5d769
DH
1506 limit++;
1507 }
8feae131
DH
1508 return -EINVAL;
1509 }
1da177e4 1510
8feae131
DH
1511 /* we're allowed to split an anonymous VMA but not a file-backed one */
1512 if (vma->vm_file) {
1513 do {
22cc877b 1514 if (start > vma->vm_start)
8feae131 1515 return -EINVAL;
8feae131
DH
1516 if (end == vma->vm_end)
1517 goto erase_whole_vma;
d75a310c
NK
1518 vma = vma->vm_next;
1519 } while (vma);
8feae131
DH
1520 return -EINVAL;
1521 } else {
1522 /* the chunk must be a subset of the VMA found */
1523 if (start == vma->vm_start && end == vma->vm_end)
1524 goto erase_whole_vma;
22cc877b 1525 if (start < vma->vm_start || end > vma->vm_end)
8feae131 1526 return -EINVAL;
1824cb75 1527 if (offset_in_page(start))
8feae131 1528 return -EINVAL;
1824cb75 1529 if (end != vma->vm_end && offset_in_page(end))
8feae131 1530 return -EINVAL;
8feae131
DH
1531 if (start != vma->vm_start && end != vma->vm_end) {
1532 ret = split_vma(mm, vma, start, 1);
22cc877b 1533 if (ret < 0)
8feae131 1534 return ret;
8feae131
DH
1535 }
1536 return shrink_vma(mm, vma, start, end);
1537 }
1da177e4 1538
8feae131
DH
1539erase_whole_vma:
1540 delete_vma_from_mm(vma);
1541 delete_vma(mm, vma);
1da177e4
LT
1542 return 0;
1543}
b5073173 1544EXPORT_SYMBOL(do_munmap);
1da177e4 1545
bfce281c 1546int vm_munmap(unsigned long addr, size_t len)
3034097a 1547{
bfce281c 1548 struct mm_struct *mm = current->mm;
3034097a 1549 int ret;
3034097a
DH
1550
1551 down_write(&mm->mmap_sem);
897ab3e0 1552 ret = do_munmap(mm, addr, len, NULL);
3034097a
DH
1553 up_write(&mm->mmap_sem);
1554 return ret;
1555}
a46ef99d
LT
1556EXPORT_SYMBOL(vm_munmap);
1557
1558SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1559{
bfce281c 1560 return vm_munmap(addr, len);
a46ef99d 1561}
3034097a
DH
1562
1563/*
8feae131 1564 * release all the mappings made in a process's VM space
3034097a 1565 */
8feae131 1566void exit_mmap(struct mm_struct *mm)
1da177e4 1567{
8feae131 1568 struct vm_area_struct *vma;
1da177e4 1569
8feae131
DH
1570 if (!mm)
1571 return;
1da177e4 1572
8feae131 1573 mm->total_vm = 0;
1da177e4 1574
8feae131
DH
1575 while ((vma = mm->mmap)) {
1576 mm->mmap = vma->vm_next;
1577 delete_vma_from_mm(vma);
1578 delete_vma(mm, vma);
04c34961 1579 cond_resched();
1da177e4
LT
1580 }
1581}
1582
5d22fc25 1583int vm_brk(unsigned long addr, unsigned long len)
1da177e4
LT
1584{
1585 return -ENOMEM;
1586}
1587
1588/*
6fa5f80b
DH
1589 * expand (or shrink) an existing mapping, potentially moving it at the same
1590 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1da177e4 1591 *
6fa5f80b 1592 * under NOMMU conditions, we only permit changing a mapping's size, and only
8feae131
DH
1593 * as long as it stays within the region allocated by do_mmap_private() and the
1594 * block is not shareable
1da177e4 1595 *
6fa5f80b 1596 * MREMAP_FIXED is not supported under NOMMU conditions
1da177e4 1597 */
4b377bab 1598static unsigned long do_mremap(unsigned long addr,
1da177e4
LT
1599 unsigned long old_len, unsigned long new_len,
1600 unsigned long flags, unsigned long new_addr)
1601{
6fa5f80b 1602 struct vm_area_struct *vma;
1da177e4
LT
1603
1604 /* insanity checks first */
f67d9b15
BL
1605 old_len = PAGE_ALIGN(old_len);
1606 new_len = PAGE_ALIGN(new_len);
8feae131 1607 if (old_len == 0 || new_len == 0)
1da177e4
LT
1608 return (unsigned long) -EINVAL;
1609
1824cb75 1610 if (offset_in_page(addr))
8feae131
DH
1611 return -EINVAL;
1612
1da177e4
LT
1613 if (flags & MREMAP_FIXED && new_addr != addr)
1614 return (unsigned long) -EINVAL;
1615
8feae131 1616 vma = find_vma_exact(current->mm, addr, old_len);
6fa5f80b
DH
1617 if (!vma)
1618 return (unsigned long) -EINVAL;
1da177e4 1619
6fa5f80b 1620 if (vma->vm_end != vma->vm_start + old_len)
1da177e4
LT
1621 return (unsigned long) -EFAULT;
1622
6fa5f80b 1623 if (vma->vm_flags & VM_MAYSHARE)
1da177e4
LT
1624 return (unsigned long) -EPERM;
1625
8feae131 1626 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1da177e4
LT
1627 return (unsigned long) -ENOMEM;
1628
1629 /* all checks complete - do it */
6fa5f80b 1630 vma->vm_end = vma->vm_start + new_len;
6fa5f80b
DH
1631 return vma->vm_start;
1632}
1633
6a6160a7
HC
1634SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1635 unsigned long, new_len, unsigned long, flags,
1636 unsigned long, new_addr)
6fa5f80b
DH
1637{
1638 unsigned long ret;
1639
1640 down_write(&current->mm->mmap_sem);
1641 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1642 up_write(&current->mm->mmap_sem);
1643 return ret;
1da177e4
LT
1644}
1645
df06b37f
KB
1646struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1647 unsigned int foll_flags)
1da177e4
LT
1648{
1649 return NULL;
1650}
1651
8f3b1327
BL
1652int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1653 unsigned long pfn, unsigned long size, pgprot_t prot)
1da177e4 1654{
8f3b1327
BL
1655 if (addr != (pfn << PAGE_SHIFT))
1656 return -EINVAL;
1657
314e51b9 1658 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
66aa2b4b 1659 return 0;
1da177e4 1660}
22c4af40 1661EXPORT_SYMBOL(remap_pfn_range);
1da177e4 1662
3c0b9de6
LT
1663int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1664{
1665 unsigned long pfn = start >> PAGE_SHIFT;
1666 unsigned long vm_len = vma->vm_end - vma->vm_start;
1667
1668 pfn += vma->vm_pgoff;
1669 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1670}
1671EXPORT_SYMBOL(vm_iomap_memory);
1672
f905bc44
PM
1673int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1674 unsigned long pgoff)
1675{
1676 unsigned int size = vma->vm_end - vma->vm_start;
1677
1678 if (!(vma->vm_flags & VM_USERMAP))
1679 return -EINVAL;
1680
1681 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1682 vma->vm_end = vma->vm_start + size;
1683
1684 return 0;
1685}
1686EXPORT_SYMBOL(remap_vmalloc_range);
1687
1da177e4
LT
1688unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1689 unsigned long len, unsigned long pgoff, unsigned long flags)
1690{
1691 return -ENOMEM;
1692}
1693
2bcd6454 1694vm_fault_t filemap_fault(struct vm_fault *vmf)
b0e15190
DH
1695{
1696 BUG();
d0217ac0 1697 return 0;
b0e15190 1698}
b5073173 1699EXPORT_SYMBOL(filemap_fault);
0ec76a11 1700
82b0f8c3 1701void filemap_map_pages(struct vm_fault *vmf,
bae473a4 1702 pgoff_t start_pgoff, pgoff_t end_pgoff)
f1820361
KS
1703{
1704 BUG();
1705}
1706EXPORT_SYMBOL(filemap_map_pages);
1707
84d77d3f 1708int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
442486ec 1709 unsigned long addr, void *buf, int len, unsigned int gup_flags)
0ec76a11 1710{
0ec76a11 1711 struct vm_area_struct *vma;
442486ec 1712 int write = gup_flags & FOLL_WRITE;
0ec76a11 1713
1e426fe2
KK
1714 if (down_read_killable(&mm->mmap_sem))
1715 return 0;
0ec76a11
DH
1716
1717 /* the access must start within one of the target process's mappings */
0159b141
DH
1718 vma = find_vma(mm, addr);
1719 if (vma) {
0ec76a11
DH
1720 /* don't overrun this mapping */
1721 if (addr + len >= vma->vm_end)
1722 len = vma->vm_end - addr;
1723
1724 /* only read or write mappings where it is permitted */
d00c7b99 1725 if (write && vma->vm_flags & VM_MAYWRITE)
7959722b
JZ
1726 copy_to_user_page(vma, NULL, addr,
1727 (void *) addr, buf, len);
d00c7b99 1728 else if (!write && vma->vm_flags & VM_MAYREAD)
7959722b
JZ
1729 copy_from_user_page(vma, NULL, addr,
1730 buf, (void *) addr, len);
0ec76a11
DH
1731 else
1732 len = 0;
1733 } else {
1734 len = 0;
1735 }
1736
1737 up_read(&mm->mmap_sem);
f55f199b
MF
1738
1739 return len;
1740}
1741
1742/**
b7701a5f 1743 * access_remote_vm - access another process' address space
f55f199b
MF
1744 * @mm: the mm_struct of the target address space
1745 * @addr: start address to access
1746 * @buf: source or destination buffer
1747 * @len: number of bytes to transfer
6347e8d5 1748 * @gup_flags: flags modifying lookup behaviour
f55f199b
MF
1749 *
1750 * The caller must hold a reference on @mm.
1751 */
1752int access_remote_vm(struct mm_struct *mm, unsigned long addr,
6347e8d5 1753 void *buf, int len, unsigned int gup_flags)
f55f199b 1754{
6347e8d5 1755 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
f55f199b
MF
1756}
1757
1758/*
1759 * Access another process' address space.
1760 * - source/target buffer must be kernel space
1761 */
f307ab6d
LS
1762int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1763 unsigned int gup_flags)
f55f199b
MF
1764{
1765 struct mm_struct *mm;
1766
1767 if (addr + len < addr)
1768 return 0;
1769
1770 mm = get_task_mm(tsk);
1771 if (!mm)
1772 return 0;
1773
f307ab6d 1774 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
f55f199b 1775
0ec76a11
DH
1776 mmput(mm);
1777 return len;
1778}
fcd35857 1779EXPORT_SYMBOL_GPL(access_process_vm);
7e660872
DH
1780
1781/**
1782 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1783 * @inode: The inode to check
1784 * @size: The current filesize of the inode
1785 * @newsize: The proposed filesize of the inode
1786 *
1787 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1788 * make sure that that any outstanding VMAs aren't broken and then shrink the
1789 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1790 * automatically grant mappings that are too large.
1791 */
1792int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1793 size_t newsize)
1794{
1795 struct vm_area_struct *vma;
7e660872
DH
1796 struct vm_region *region;
1797 pgoff_t low, high;
1798 size_t r_size, r_top;
1799
1800 low = newsize >> PAGE_SHIFT;
1801 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1802
1803 down_write(&nommu_region_sem);
1acf2e04 1804 i_mmap_lock_read(inode->i_mapping);
7e660872
DH
1805
1806 /* search for VMAs that fall within the dead zone */
6b2dbba8 1807 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
7e660872
DH
1808 /* found one - only interested if it's shared out of the page
1809 * cache */
1810 if (vma->vm_flags & VM_SHARED) {
1acf2e04 1811 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1812 up_write(&nommu_region_sem);
1813 return -ETXTBSY; /* not quite true, but near enough */
1814 }
1815 }
1816
1817 /* reduce any regions that overlap the dead zone - if in existence,
1818 * these will be pointed to by VMAs that don't overlap the dead zone
1819 *
1820 * we don't check for any regions that start beyond the EOF as there
1821 * shouldn't be any
1822 */
1acf2e04 1823 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
7e660872
DH
1824 if (!(vma->vm_flags & VM_SHARED))
1825 continue;
1826
1827 region = vma->vm_region;
1828 r_size = region->vm_top - region->vm_start;
1829 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1830
1831 if (r_top > newsize) {
1832 region->vm_top -= r_top - newsize;
1833 if (region->vm_end > region->vm_top)
1834 region->vm_end = region->vm_top;
1835 }
1836 }
1837
1acf2e04 1838 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1839 up_write(&nommu_region_sem);
1840 return 0;
1841}
c9b1d098
AS
1842
1843/*
1844 * Initialise sysctl_user_reserve_kbytes.
1845 *
1846 * This is intended to prevent a user from starting a single memory hogging
1847 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1848 * mode.
1849 *
1850 * The default value is min(3% of free memory, 128MB)
1851 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1852 */
1853static int __meminit init_user_reserve(void)
1854{
1855 unsigned long free_kbytes;
1856
c41f012a 1857 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
c9b1d098
AS
1858
1859 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1860 return 0;
1861}
a4bc6fc7 1862subsys_initcall(init_user_reserve);
4eeab4f5
AS
1863
1864/*
1865 * Initialise sysctl_admin_reserve_kbytes.
1866 *
1867 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1868 * to log in and kill a memory hogging process.
1869 *
1870 * Systems with more than 256MB will reserve 8MB, enough to recover
1871 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1872 * only reserve 3% of free pages by default.
1873 */
1874static int __meminit init_admin_reserve(void)
1875{
1876 unsigned long free_kbytes;
1877
c41f012a 1878 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
4eeab4f5
AS
1879
1880 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1881 return 0;
1882}
a4bc6fc7 1883subsys_initcall(init_admin_reserve);