mm: nommu: sort mm->mmap list properly
[linux-2.6-block.git] / mm / nommu.c
CommitLineData
1da177e4
LT
1/*
2 * linux/mm/nommu.c
3 *
4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
6 *
7 * See Documentation/nommu-mmap.txt
8 *
8feae131 9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
1da177e4
LT
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
29c185e5 13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
1da177e4
LT
14 */
15
f2b8544f 16#include <linux/module.h>
1da177e4
LT
17#include <linux/mm.h>
18#include <linux/mman.h>
19#include <linux/swap.h>
20#include <linux/file.h>
21#include <linux/highmem.h>
22#include <linux/pagemap.h>
23#include <linux/slab.h>
24#include <linux/vmalloc.h>
fa8e26cc 25#include <linux/tracehook.h>
1da177e4
LT
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/mount.h>
29#include <linux/personality.h>
30#include <linux/security.h>
31#include <linux/syscalls.h>
120a795d 32#include <linux/audit.h>
1da177e4
LT
33
34#include <asm/uaccess.h>
35#include <asm/tlb.h>
36#include <asm/tlbflush.h>
eb8cdec4 37#include <asm/mmu_context.h>
8feae131
DH
38#include "internal.h"
39
8feae131
DH
40#if 0
41#define kenter(FMT, ...) \
42 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
43#define kleave(FMT, ...) \
44 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
45#define kdebug(FMT, ...) \
46 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
47#else
48#define kenter(FMT, ...) \
49 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
50#define kleave(FMT, ...) \
51 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
52#define kdebug(FMT, ...) \
53 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
54#endif
1da177e4
LT
55
56void *high_memory;
57struct page *mem_map;
58unsigned long max_mapnr;
59unsigned long num_physpages;
4266c97a 60unsigned long highest_memmap_pfn;
00a62ce9 61struct percpu_counter vm_committed_as;
1da177e4
LT
62int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
63int sysctl_overcommit_ratio = 50; /* default is 50% */
64int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
fc4d5c29 65int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
1da177e4
LT
66int heap_stack_gap = 0;
67
33e5d769 68atomic_long_t mmap_pages_allocated;
8feae131 69
1da177e4 70EXPORT_SYMBOL(mem_map);
6a04de6d 71EXPORT_SYMBOL(num_physpages);
1da177e4 72
8feae131
DH
73/* list of mapped, potentially shareable regions */
74static struct kmem_cache *vm_region_jar;
75struct rb_root nommu_region_tree = RB_ROOT;
76DECLARE_RWSEM(nommu_region_sem);
1da177e4 77
f0f37e2f 78const struct vm_operations_struct generic_file_vm_ops = {
1da177e4
LT
79};
80
1da177e4
LT
81/*
82 * Return the total memory allocated for this pointer, not
83 * just what the caller asked for.
84 *
85 * Doesn't have to be accurate, i.e. may have races.
86 */
87unsigned int kobjsize(const void *objp)
88{
89 struct page *page;
90
4016a139
MH
91 /*
92 * If the object we have should not have ksize performed on it,
93 * return size of 0
94 */
5a1603be 95 if (!objp || !virt_addr_valid(objp))
6cfd53fc
PM
96 return 0;
97
98 page = virt_to_head_page(objp);
6cfd53fc
PM
99
100 /*
101 * If the allocator sets PageSlab, we know the pointer came from
102 * kmalloc().
103 */
1da177e4
LT
104 if (PageSlab(page))
105 return ksize(objp);
106
ab2e83ea
PM
107 /*
108 * If it's not a compound page, see if we have a matching VMA
109 * region. This test is intentionally done in reverse order,
110 * so if there's no VMA, we still fall through and hand back
111 * PAGE_SIZE for 0-order pages.
112 */
113 if (!PageCompound(page)) {
114 struct vm_area_struct *vma;
115
116 vma = find_vma(current->mm, (unsigned long)objp);
117 if (vma)
118 return vma->vm_end - vma->vm_start;
119 }
120
6cfd53fc
PM
121 /*
122 * The ksize() function is only guaranteed to work for pointers
5a1603be 123 * returned by kmalloc(). So handle arbitrary pointers here.
6cfd53fc 124 */
5a1603be 125 return PAGE_SIZE << compound_order(page);
1da177e4
LT
126}
127
b291f000 128int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
4266c97a 129 unsigned long start, int nr_pages, unsigned int foll_flags,
53a7706d
ML
130 struct page **pages, struct vm_area_struct **vmas,
131 int *retry)
1da177e4 132{
910e46da 133 struct vm_area_struct *vma;
7b4d5b8b
DH
134 unsigned long vm_flags;
135 int i;
136
137 /* calculate required read or write permissions.
58fa879e 138 * If FOLL_FORCE is set, we only require the "MAY" flags.
7b4d5b8b 139 */
58fa879e
HD
140 vm_flags = (foll_flags & FOLL_WRITE) ?
141 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
142 vm_flags &= (foll_flags & FOLL_FORCE) ?
143 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1da177e4 144
9d73777e 145 for (i = 0; i < nr_pages; i++) {
7561e8ca 146 vma = find_vma(mm, start);
7b4d5b8b
DH
147 if (!vma)
148 goto finish_or_fault;
149
150 /* protect what we can, including chardevs */
1c3aff1c
HD
151 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
152 !(vm_flags & vma->vm_flags))
7b4d5b8b 153 goto finish_or_fault;
910e46da 154
1da177e4
LT
155 if (pages) {
156 pages[i] = virt_to_page(start);
157 if (pages[i])
158 page_cache_get(pages[i]);
159 }
160 if (vmas)
910e46da 161 vmas[i] = vma;
e1ee65d8 162 start = (start + PAGE_SIZE) & PAGE_MASK;
1da177e4 163 }
7b4d5b8b
DH
164
165 return i;
166
167finish_or_fault:
168 return i ? : -EFAULT;
1da177e4 169}
b291f000 170
b291f000
NP
171/*
172 * get a list of pages in an address range belonging to the specified process
173 * and indicate the VMA that covers each page
174 * - this is potentially dodgy as we may end incrementing the page count of a
175 * slab page or a secondary page from a compound page
176 * - don't permit access to VMAs that don't support it, such as I/O mappings
177 */
178int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
9d73777e 179 unsigned long start, int nr_pages, int write, int force,
b291f000
NP
180 struct page **pages, struct vm_area_struct **vmas)
181{
182 int flags = 0;
183
184 if (write)
58fa879e 185 flags |= FOLL_WRITE;
b291f000 186 if (force)
58fa879e 187 flags |= FOLL_FORCE;
b291f000 188
53a7706d
ML
189 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
190 NULL);
b291f000 191}
66aa2b4b
GU
192EXPORT_SYMBOL(get_user_pages);
193
dfc2f91a
PM
194/**
195 * follow_pfn - look up PFN at a user virtual address
196 * @vma: memory mapping
197 * @address: user virtual address
198 * @pfn: location to store found PFN
199 *
200 * Only IO mappings and raw PFN mappings are allowed.
201 *
202 * Returns zero and the pfn at @pfn on success, -ve otherwise.
203 */
204int follow_pfn(struct vm_area_struct *vma, unsigned long address,
205 unsigned long *pfn)
206{
207 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
208 return -EINVAL;
209
210 *pfn = address >> PAGE_SHIFT;
211 return 0;
212}
213EXPORT_SYMBOL(follow_pfn);
214
1da177e4
LT
215DEFINE_RWLOCK(vmlist_lock);
216struct vm_struct *vmlist;
217
b3bdda02 218void vfree(const void *addr)
1da177e4
LT
219{
220 kfree(addr);
221}
b5073173 222EXPORT_SYMBOL(vfree);
1da177e4 223
dd0fc66f 224void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1da177e4
LT
225{
226 /*
8518609d
RD
227 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
228 * returns only a logical address.
1da177e4 229 */
84097518 230 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
1da177e4 231}
b5073173 232EXPORT_SYMBOL(__vmalloc);
1da177e4 233
f905bc44
PM
234void *vmalloc_user(unsigned long size)
235{
236 void *ret;
237
238 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
239 PAGE_KERNEL);
240 if (ret) {
241 struct vm_area_struct *vma;
242
243 down_write(&current->mm->mmap_sem);
244 vma = find_vma(current->mm, (unsigned long)ret);
245 if (vma)
246 vma->vm_flags |= VM_USERMAP;
247 up_write(&current->mm->mmap_sem);
248 }
249
250 return ret;
251}
252EXPORT_SYMBOL(vmalloc_user);
253
b3bdda02 254struct page *vmalloc_to_page(const void *addr)
1da177e4
LT
255{
256 return virt_to_page(addr);
257}
b5073173 258EXPORT_SYMBOL(vmalloc_to_page);
1da177e4 259
b3bdda02 260unsigned long vmalloc_to_pfn(const void *addr)
1da177e4
LT
261{
262 return page_to_pfn(virt_to_page(addr));
263}
b5073173 264EXPORT_SYMBOL(vmalloc_to_pfn);
1da177e4
LT
265
266long vread(char *buf, char *addr, unsigned long count)
267{
268 memcpy(buf, addr, count);
269 return count;
270}
271
272long vwrite(char *buf, char *addr, unsigned long count)
273{
274 /* Don't allow overflow */
275 if ((unsigned long) addr + count < count)
276 count = -(unsigned long) addr;
277
278 memcpy(addr, buf, count);
279 return(count);
280}
281
282/*
283 * vmalloc - allocate virtually continguos memory
284 *
285 * @size: allocation size
286 *
287 * Allocate enough pages to cover @size from the page level
288 * allocator and map them into continguos kernel virtual space.
289 *
c1c8897f 290 * For tight control over page level allocator and protection flags
1da177e4
LT
291 * use __vmalloc() instead.
292 */
293void *vmalloc(unsigned long size)
294{
295 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
296}
f6138882
AM
297EXPORT_SYMBOL(vmalloc);
298
e1ca7788
DY
299/*
300 * vzalloc - allocate virtually continguos memory with zero fill
301 *
302 * @size: allocation size
303 *
304 * Allocate enough pages to cover @size from the page level
305 * allocator and map them into continguos kernel virtual space.
306 * The memory allocated is set to zero.
307 *
308 * For tight control over page level allocator and protection flags
309 * use __vmalloc() instead.
310 */
311void *vzalloc(unsigned long size)
312{
313 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
314 PAGE_KERNEL);
315}
316EXPORT_SYMBOL(vzalloc);
317
318/**
319 * vmalloc_node - allocate memory on a specific node
320 * @size: allocation size
321 * @node: numa node
322 *
323 * Allocate enough pages to cover @size from the page level
324 * allocator and map them into contiguous kernel virtual space.
325 *
326 * For tight control over page level allocator and protection flags
327 * use __vmalloc() instead.
328 */
f6138882
AM
329void *vmalloc_node(unsigned long size, int node)
330{
331 return vmalloc(size);
332}
9a14f653 333EXPORT_SYMBOL(vmalloc_node);
e1ca7788
DY
334
335/**
336 * vzalloc_node - allocate memory on a specific node with zero fill
337 * @size: allocation size
338 * @node: numa node
339 *
340 * Allocate enough pages to cover @size from the page level
341 * allocator and map them into contiguous kernel virtual space.
342 * The memory allocated is set to zero.
343 *
344 * For tight control over page level allocator and protection flags
345 * use __vmalloc() instead.
346 */
347void *vzalloc_node(unsigned long size, int node)
348{
349 return vzalloc(size);
350}
351EXPORT_SYMBOL(vzalloc_node);
1da177e4 352
1af446ed
PM
353#ifndef PAGE_KERNEL_EXEC
354# define PAGE_KERNEL_EXEC PAGE_KERNEL
355#endif
356
357/**
358 * vmalloc_exec - allocate virtually contiguous, executable memory
359 * @size: allocation size
360 *
361 * Kernel-internal function to allocate enough pages to cover @size
362 * the page level allocator and map them into contiguous and
363 * executable kernel virtual space.
364 *
365 * For tight control over page level allocator and protection flags
366 * use __vmalloc() instead.
367 */
368
369void *vmalloc_exec(unsigned long size)
370{
371 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
372}
373
b5073173
PM
374/**
375 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1da177e4
LT
376 * @size: allocation size
377 *
378 * Allocate enough 32bit PA addressable pages to cover @size from the
379 * page level allocator and map them into continguos kernel virtual space.
380 */
381void *vmalloc_32(unsigned long size)
382{
383 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
384}
b5073173
PM
385EXPORT_SYMBOL(vmalloc_32);
386
387/**
388 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
389 * @size: allocation size
390 *
391 * The resulting memory area is 32bit addressable and zeroed so it can be
392 * mapped to userspace without leaking data.
f905bc44
PM
393 *
394 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
395 * remap_vmalloc_range() are permissible.
b5073173
PM
396 */
397void *vmalloc_32_user(unsigned long size)
398{
f905bc44
PM
399 /*
400 * We'll have to sort out the ZONE_DMA bits for 64-bit,
401 * but for now this can simply use vmalloc_user() directly.
402 */
403 return vmalloc_user(size);
b5073173
PM
404}
405EXPORT_SYMBOL(vmalloc_32_user);
1da177e4
LT
406
407void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
408{
409 BUG();
410 return NULL;
411}
b5073173 412EXPORT_SYMBOL(vmap);
1da177e4 413
b3bdda02 414void vunmap(const void *addr)
1da177e4
LT
415{
416 BUG();
417}
b5073173 418EXPORT_SYMBOL(vunmap);
1da177e4 419
eb6434d9
PM
420void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
421{
422 BUG();
423 return NULL;
424}
425EXPORT_SYMBOL(vm_map_ram);
426
427void vm_unmap_ram(const void *mem, unsigned int count)
428{
429 BUG();
430}
431EXPORT_SYMBOL(vm_unmap_ram);
432
433void vm_unmap_aliases(void)
434{
435}
436EXPORT_SYMBOL_GPL(vm_unmap_aliases);
437
1eeb66a1
CH
438/*
439 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
440 * have one.
441 */
442void __attribute__((weak)) vmalloc_sync_all(void)
443{
444}
445
29c185e5
PM
446/**
447 * alloc_vm_area - allocate a range of kernel address space
448 * @size: size of the area
449 *
450 * Returns: NULL on failure, vm_struct on success
451 *
452 * This function reserves a range of kernel address space, and
453 * allocates pagetables to map that range. No actual mappings
454 * are created. If the kernel address space is not shared
455 * between processes, it syncs the pagetable across all
456 * processes.
457 */
458struct vm_struct *alloc_vm_area(size_t size)
459{
460 BUG();
461 return NULL;
462}
463EXPORT_SYMBOL_GPL(alloc_vm_area);
464
465void free_vm_area(struct vm_struct *area)
466{
467 BUG();
468}
469EXPORT_SYMBOL_GPL(free_vm_area);
470
b5073173
PM
471int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
472 struct page *page)
473{
474 return -EINVAL;
475}
476EXPORT_SYMBOL(vm_insert_page);
477
1da177e4
LT
478/*
479 * sys_brk() for the most part doesn't need the global kernel
480 * lock, except when an application is doing something nasty
481 * like trying to un-brk an area that has already been mapped
482 * to a regular file. in this case, the unmapping will need
483 * to invoke file system routines that need the global lock.
484 */
6a6160a7 485SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4
LT
486{
487 struct mm_struct *mm = current->mm;
488
489 if (brk < mm->start_brk || brk > mm->context.end_brk)
490 return mm->brk;
491
492 if (mm->brk == brk)
493 return mm->brk;
494
495 /*
496 * Always allow shrinking brk
497 */
498 if (brk <= mm->brk) {
499 mm->brk = brk;
500 return brk;
501 }
502
503 /*
504 * Ok, looks good - let it rip.
505 */
cfe79c00 506 flush_icache_range(mm->brk, brk);
1da177e4
LT
507 return mm->brk = brk;
508}
509
8feae131
DH
510/*
511 * initialise the VMA and region record slabs
512 */
513void __init mmap_init(void)
1da177e4 514{
00a62ce9
KM
515 int ret;
516
517 ret = percpu_counter_init(&vm_committed_as, 0);
518 VM_BUG_ON(ret);
33e5d769 519 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
1da177e4 520}
1da177e4 521
3034097a 522/*
8feae131
DH
523 * validate the region tree
524 * - the caller must hold the region lock
3034097a 525 */
8feae131
DH
526#ifdef CONFIG_DEBUG_NOMMU_REGIONS
527static noinline void validate_nommu_regions(void)
3034097a 528{
8feae131
DH
529 struct vm_region *region, *last;
530 struct rb_node *p, *lastp;
3034097a 531
8feae131
DH
532 lastp = rb_first(&nommu_region_tree);
533 if (!lastp)
534 return;
535
536 last = rb_entry(lastp, struct vm_region, vm_rb);
33e5d769
DH
537 BUG_ON(unlikely(last->vm_end <= last->vm_start));
538 BUG_ON(unlikely(last->vm_top < last->vm_end));
8feae131
DH
539
540 while ((p = rb_next(lastp))) {
541 region = rb_entry(p, struct vm_region, vm_rb);
542 last = rb_entry(lastp, struct vm_region, vm_rb);
543
33e5d769
DH
544 BUG_ON(unlikely(region->vm_end <= region->vm_start));
545 BUG_ON(unlikely(region->vm_top < region->vm_end));
546 BUG_ON(unlikely(region->vm_start < last->vm_top));
3034097a 547
8feae131
DH
548 lastp = p;
549 }
3034097a 550}
8feae131 551#else
33e5d769
DH
552static void validate_nommu_regions(void)
553{
554}
8feae131 555#endif
3034097a
DH
556
557/*
8feae131 558 * add a region into the global tree
3034097a 559 */
8feae131 560static void add_nommu_region(struct vm_region *region)
3034097a 561{
8feae131
DH
562 struct vm_region *pregion;
563 struct rb_node **p, *parent;
3034097a 564
8feae131
DH
565 validate_nommu_regions();
566
8feae131
DH
567 parent = NULL;
568 p = &nommu_region_tree.rb_node;
569 while (*p) {
570 parent = *p;
571 pregion = rb_entry(parent, struct vm_region, vm_rb);
572 if (region->vm_start < pregion->vm_start)
573 p = &(*p)->rb_left;
574 else if (region->vm_start > pregion->vm_start)
575 p = &(*p)->rb_right;
576 else if (pregion == region)
577 return;
578 else
579 BUG();
3034097a
DH
580 }
581
8feae131
DH
582 rb_link_node(&region->vm_rb, parent, p);
583 rb_insert_color(&region->vm_rb, &nommu_region_tree);
3034097a 584
8feae131 585 validate_nommu_regions();
3034097a 586}
3034097a 587
930e652a 588/*
8feae131 589 * delete a region from the global tree
930e652a 590 */
8feae131 591static void delete_nommu_region(struct vm_region *region)
930e652a 592{
8feae131 593 BUG_ON(!nommu_region_tree.rb_node);
930e652a 594
8feae131
DH
595 validate_nommu_regions();
596 rb_erase(&region->vm_rb, &nommu_region_tree);
597 validate_nommu_regions();
57c8f63e
GU
598}
599
6fa5f80b 600/*
8feae131 601 * free a contiguous series of pages
6fa5f80b 602 */
8feae131 603static void free_page_series(unsigned long from, unsigned long to)
6fa5f80b 604{
8feae131
DH
605 for (; from < to; from += PAGE_SIZE) {
606 struct page *page = virt_to_page(from);
607
608 kdebug("- free %lx", from);
33e5d769 609 atomic_long_dec(&mmap_pages_allocated);
8feae131 610 if (page_count(page) != 1)
33e5d769
DH
611 kdebug("free page %p: refcount not one: %d",
612 page, page_count(page));
8feae131 613 put_page(page);
6fa5f80b 614 }
6fa5f80b
DH
615}
616
3034097a 617/*
8feae131 618 * release a reference to a region
33e5d769 619 * - the caller must hold the region semaphore for writing, which this releases
dd8632a1 620 * - the region may not have been added to the tree yet, in which case vm_top
8feae131 621 * will equal vm_start
3034097a 622 */
8feae131
DH
623static void __put_nommu_region(struct vm_region *region)
624 __releases(nommu_region_sem)
1da177e4 625{
1e2ae599 626 kenter("%p{%d}", region, region->vm_usage);
1da177e4 627
8feae131 628 BUG_ON(!nommu_region_tree.rb_node);
1da177e4 629
1e2ae599 630 if (--region->vm_usage == 0) {
dd8632a1 631 if (region->vm_top > region->vm_start)
8feae131
DH
632 delete_nommu_region(region);
633 up_write(&nommu_region_sem);
634
635 if (region->vm_file)
636 fput(region->vm_file);
637
638 /* IO memory and memory shared directly out of the pagecache
639 * from ramfs/tmpfs mustn't be released here */
640 if (region->vm_flags & VM_MAPPED_COPY) {
641 kdebug("free series");
dd8632a1 642 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
643 }
644 kmem_cache_free(vm_region_jar, region);
645 } else {
646 up_write(&nommu_region_sem);
1da177e4 647 }
8feae131 648}
1da177e4 649
8feae131
DH
650/*
651 * release a reference to a region
652 */
653static void put_nommu_region(struct vm_region *region)
654{
655 down_write(&nommu_region_sem);
656 __put_nommu_region(region);
1da177e4
LT
657}
658
eb8cdec4
BS
659/*
660 * update protection on a vma
661 */
662static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
663{
664#ifdef CONFIG_MPU
665 struct mm_struct *mm = vma->vm_mm;
666 long start = vma->vm_start & PAGE_MASK;
667 while (start < vma->vm_end) {
668 protect_page(mm, start, flags);
669 start += PAGE_SIZE;
670 }
671 update_protections(mm);
672#endif
673}
674
3034097a 675/*
8feae131
DH
676 * add a VMA into a process's mm_struct in the appropriate place in the list
677 * and tree and add to the address space's page tree also if not an anonymous
678 * page
679 * - should be called with mm->mmap_sem held writelocked
3034097a 680 */
8feae131 681static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 682{
6038def0 683 struct vm_area_struct *pvma, *prev;
1da177e4 684 struct address_space *mapping;
6038def0 685 struct rb_node **p, *parent, *rb_prev;
8feae131
DH
686
687 kenter(",%p", vma);
688
689 BUG_ON(!vma->vm_region);
690
691 mm->map_count++;
692 vma->vm_mm = mm;
1da177e4 693
eb8cdec4
BS
694 protect_vma(vma, vma->vm_flags);
695
1da177e4
LT
696 /* add the VMA to the mapping */
697 if (vma->vm_file) {
698 mapping = vma->vm_file->f_mapping;
699
700 flush_dcache_mmap_lock(mapping);
701 vma_prio_tree_insert(vma, &mapping->i_mmap);
702 flush_dcache_mmap_unlock(mapping);
703 }
704
8feae131 705 /* add the VMA to the tree */
6038def0 706 parent = rb_prev = NULL;
8feae131 707 p = &mm->mm_rb.rb_node;
1da177e4
LT
708 while (*p) {
709 parent = *p;
710 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
711
8feae131
DH
712 /* sort by: start addr, end addr, VMA struct addr in that order
713 * (the latter is necessary as we may get identical VMAs) */
714 if (vma->vm_start < pvma->vm_start)
1da177e4 715 p = &(*p)->rb_left;
6038def0
NK
716 else if (vma->vm_start > pvma->vm_start) {
717 rb_prev = parent;
1da177e4 718 p = &(*p)->rb_right;
6038def0 719 } else if (vma->vm_end < pvma->vm_end)
8feae131 720 p = &(*p)->rb_left;
6038def0
NK
721 else if (vma->vm_end > pvma->vm_end) {
722 rb_prev = parent;
8feae131 723 p = &(*p)->rb_right;
6038def0 724 } else if (vma < pvma)
8feae131 725 p = &(*p)->rb_left;
6038def0
NK
726 else if (vma > pvma) {
727 rb_prev = parent;
8feae131 728 p = &(*p)->rb_right;
6038def0 729 } else
8feae131 730 BUG();
1da177e4
LT
731 }
732
733 rb_link_node(&vma->vm_rb, parent, p);
8feae131
DH
734 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
735
736 /* add VMA to the VMA list also */
6038def0
NK
737 prev = NULL;
738 if (rb_prev)
739 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
8feae131 740
6038def0 741 __vma_link_list(mm, vma, prev, parent);
1da177e4
LT
742}
743
3034097a 744/*
8feae131 745 * delete a VMA from its owning mm_struct and address space
3034097a 746 */
8feae131 747static void delete_vma_from_mm(struct vm_area_struct *vma)
1da177e4 748{
8feae131 749 struct vm_area_struct **pp;
1da177e4 750 struct address_space *mapping;
8feae131
DH
751 struct mm_struct *mm = vma->vm_mm;
752
753 kenter("%p", vma);
754
eb8cdec4
BS
755 protect_vma(vma, 0);
756
8feae131
DH
757 mm->map_count--;
758 if (mm->mmap_cache == vma)
759 mm->mmap_cache = NULL;
1da177e4
LT
760
761 /* remove the VMA from the mapping */
762 if (vma->vm_file) {
763 mapping = vma->vm_file->f_mapping;
764
765 flush_dcache_mmap_lock(mapping);
766 vma_prio_tree_remove(vma, &mapping->i_mmap);
767 flush_dcache_mmap_unlock(mapping);
768 }
769
8feae131
DH
770 /* remove from the MM's tree and list */
771 rb_erase(&vma->vm_rb, &mm->mm_rb);
772 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) {
773 if (*pp == vma) {
774 *pp = vma->vm_next;
775 break;
776 }
777 }
778
779 vma->vm_mm = NULL;
780}
781
782/*
783 * destroy a VMA record
784 */
785static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
786{
787 kenter("%p", vma);
788 if (vma->vm_ops && vma->vm_ops->close)
789 vma->vm_ops->close(vma);
790 if (vma->vm_file) {
791 fput(vma->vm_file);
792 if (vma->vm_flags & VM_EXECUTABLE)
793 removed_exe_file_vma(mm);
794 }
795 put_nommu_region(vma->vm_region);
796 kmem_cache_free(vm_area_cachep, vma);
797}
798
799/*
800 * look up the first VMA in which addr resides, NULL if none
801 * - should be called with mm->mmap_sem at least held readlocked
802 */
803struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
804{
805 struct vm_area_struct *vma;
806 struct rb_node *n = mm->mm_rb.rb_node;
807
808 /* check the cache first */
809 vma = mm->mmap_cache;
810 if (vma && vma->vm_start <= addr && vma->vm_end > addr)
811 return vma;
812
813 /* trawl the tree (there may be multiple mappings in which addr
814 * resides) */
815 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
816 vma = rb_entry(n, struct vm_area_struct, vm_rb);
817 if (vma->vm_start > addr)
818 return NULL;
819 if (vma->vm_end > addr) {
820 mm->mmap_cache = vma;
821 return vma;
822 }
823 }
824
825 return NULL;
826}
827EXPORT_SYMBOL(find_vma);
828
829/*
830 * find a VMA
831 * - we don't extend stack VMAs under NOMMU conditions
832 */
833struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
834{
7561e8ca 835 return find_vma(mm, addr);
8feae131
DH
836}
837
838/*
839 * expand a stack to a given address
840 * - not supported under NOMMU conditions
841 */
842int expand_stack(struct vm_area_struct *vma, unsigned long address)
843{
844 return -ENOMEM;
845}
846
847/*
848 * look up the first VMA exactly that exactly matches addr
849 * - should be called with mm->mmap_sem at least held readlocked
850 */
851static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
852 unsigned long addr,
853 unsigned long len)
854{
855 struct vm_area_struct *vma;
856 struct rb_node *n = mm->mm_rb.rb_node;
857 unsigned long end = addr + len;
858
859 /* check the cache first */
860 vma = mm->mmap_cache;
861 if (vma && vma->vm_start == addr && vma->vm_end == end)
862 return vma;
863
864 /* trawl the tree (there may be multiple mappings in which addr
865 * resides) */
866 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) {
867 vma = rb_entry(n, struct vm_area_struct, vm_rb);
868 if (vma->vm_start < addr)
869 continue;
870 if (vma->vm_start > addr)
871 return NULL;
872 if (vma->vm_end == end) {
873 mm->mmap_cache = vma;
874 return vma;
875 }
876 }
877
878 return NULL;
1da177e4
LT
879}
880
881/*
882 * determine whether a mapping should be permitted and, if so, what sort of
883 * mapping we're capable of supporting
884 */
885static int validate_mmap_request(struct file *file,
886 unsigned long addr,
887 unsigned long len,
888 unsigned long prot,
889 unsigned long flags,
890 unsigned long pgoff,
891 unsigned long *_capabilities)
892{
8feae131 893 unsigned long capabilities, rlen;
1da177e4
LT
894 unsigned long reqprot = prot;
895 int ret;
896
897 /* do the simple checks first */
06aab5a3 898 if (flags & MAP_FIXED) {
1da177e4
LT
899 printk(KERN_DEBUG
900 "%d: Can't do fixed-address/overlay mmap of RAM\n",
901 current->pid);
902 return -EINVAL;
903 }
904
905 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
906 (flags & MAP_TYPE) != MAP_SHARED)
907 return -EINVAL;
908
f81cff0d 909 if (!len)
1da177e4
LT
910 return -EINVAL;
911
f81cff0d 912 /* Careful about overflows.. */
8feae131
DH
913 rlen = PAGE_ALIGN(len);
914 if (!rlen || rlen > TASK_SIZE)
f81cff0d
MF
915 return -ENOMEM;
916
1da177e4 917 /* offset overflow? */
8feae131 918 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
f81cff0d 919 return -EOVERFLOW;
1da177e4
LT
920
921 if (file) {
922 /* validate file mapping requests */
923 struct address_space *mapping;
924
925 /* files must support mmap */
926 if (!file->f_op || !file->f_op->mmap)
927 return -ENODEV;
928
929 /* work out if what we've got could possibly be shared
930 * - we support chardevs that provide their own "memory"
931 * - we support files/blockdevs that are memory backed
932 */
933 mapping = file->f_mapping;
934 if (!mapping)
e9536ae7 935 mapping = file->f_path.dentry->d_inode->i_mapping;
1da177e4
LT
936
937 capabilities = 0;
938 if (mapping && mapping->backing_dev_info)
939 capabilities = mapping->backing_dev_info->capabilities;
940
941 if (!capabilities) {
942 /* no explicit capabilities set, so assume some
943 * defaults */
e9536ae7 944 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) {
1da177e4
LT
945 case S_IFREG:
946 case S_IFBLK:
947 capabilities = BDI_CAP_MAP_COPY;
948 break;
949
950 case S_IFCHR:
951 capabilities =
952 BDI_CAP_MAP_DIRECT |
953 BDI_CAP_READ_MAP |
954 BDI_CAP_WRITE_MAP;
955 break;
956
957 default:
958 return -EINVAL;
959 }
960 }
961
962 /* eliminate any capabilities that we can't support on this
963 * device */
964 if (!file->f_op->get_unmapped_area)
965 capabilities &= ~BDI_CAP_MAP_DIRECT;
966 if (!file->f_op->read)
967 capabilities &= ~BDI_CAP_MAP_COPY;
968
28d7a6ae
GY
969 /* The file shall have been opened with read permission. */
970 if (!(file->f_mode & FMODE_READ))
971 return -EACCES;
972
1da177e4
LT
973 if (flags & MAP_SHARED) {
974 /* do checks for writing, appending and locking */
975 if ((prot & PROT_WRITE) &&
976 !(file->f_mode & FMODE_WRITE))
977 return -EACCES;
978
e9536ae7 979 if (IS_APPEND(file->f_path.dentry->d_inode) &&
1da177e4
LT
980 (file->f_mode & FMODE_WRITE))
981 return -EACCES;
982
e9536ae7 983 if (locks_verify_locked(file->f_path.dentry->d_inode))
1da177e4
LT
984 return -EAGAIN;
985
986 if (!(capabilities & BDI_CAP_MAP_DIRECT))
987 return -ENODEV;
988
1da177e4
LT
989 /* we mustn't privatise shared mappings */
990 capabilities &= ~BDI_CAP_MAP_COPY;
991 }
992 else {
993 /* we're going to read the file into private memory we
994 * allocate */
995 if (!(capabilities & BDI_CAP_MAP_COPY))
996 return -ENODEV;
997
998 /* we don't permit a private writable mapping to be
999 * shared with the backing device */
1000 if (prot & PROT_WRITE)
1001 capabilities &= ~BDI_CAP_MAP_DIRECT;
1002 }
1003
3c7b2045
BS
1004 if (capabilities & BDI_CAP_MAP_DIRECT) {
1005 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1006 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1007 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1008 ) {
1009 capabilities &= ~BDI_CAP_MAP_DIRECT;
1010 if (flags & MAP_SHARED) {
1011 printk(KERN_WARNING
1012 "MAP_SHARED not completely supported on !MMU\n");
1013 return -EINVAL;
1014 }
1015 }
1016 }
1017
1da177e4
LT
1018 /* handle executable mappings and implied executable
1019 * mappings */
e9536ae7 1020 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1da177e4
LT
1021 if (prot & PROT_EXEC)
1022 return -EPERM;
1023 }
1024 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1025 /* handle implication of PROT_EXEC by PROT_READ */
1026 if (current->personality & READ_IMPLIES_EXEC) {
1027 if (capabilities & BDI_CAP_EXEC_MAP)
1028 prot |= PROT_EXEC;
1029 }
1030 }
1031 else if ((prot & PROT_READ) &&
1032 (prot & PROT_EXEC) &&
1033 !(capabilities & BDI_CAP_EXEC_MAP)
1034 ) {
1035 /* backing file is not executable, try to copy */
1036 capabilities &= ~BDI_CAP_MAP_DIRECT;
1037 }
1038 }
1039 else {
1040 /* anonymous mappings are always memory backed and can be
1041 * privately mapped
1042 */
1043 capabilities = BDI_CAP_MAP_COPY;
1044
1045 /* handle PROT_EXEC implication by PROT_READ */
1046 if ((prot & PROT_READ) &&
1047 (current->personality & READ_IMPLIES_EXEC))
1048 prot |= PROT_EXEC;
1049 }
1050
1051 /* allow the security API to have its say */
ed032189 1052 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1da177e4
LT
1053 if (ret < 0)
1054 return ret;
1055
1056 /* looks okay */
1057 *_capabilities = capabilities;
1058 return 0;
1059}
1060
1061/*
1062 * we've determined that we can make the mapping, now translate what we
1063 * now know into VMA flags
1064 */
1065static unsigned long determine_vm_flags(struct file *file,
1066 unsigned long prot,
1067 unsigned long flags,
1068 unsigned long capabilities)
1069{
1070 unsigned long vm_flags;
1071
1072 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1da177e4
LT
1073 /* vm_flags |= mm->def_flags; */
1074
1075 if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1076 /* attempt to share read-only copies of mapped file chunks */
3c7b2045 1077 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1da177e4
LT
1078 if (file && !(prot & PROT_WRITE))
1079 vm_flags |= VM_MAYSHARE;
3c7b2045 1080 } else {
1da177e4
LT
1081 /* overlay a shareable mapping on the backing device or inode
1082 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1083 * romfs/cramfs */
3c7b2045 1084 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1da177e4 1085 if (flags & MAP_SHARED)
3c7b2045 1086 vm_flags |= VM_SHARED;
1da177e4
LT
1087 }
1088
1089 /* refuse to let anyone share private mappings with this process if
1090 * it's being traced - otherwise breakpoints set in it may interfere
1091 * with another untraced process
1092 */
fa8e26cc 1093 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current))
1da177e4
LT
1094 vm_flags &= ~VM_MAYSHARE;
1095
1096 return vm_flags;
1097}
1098
1099/*
8feae131
DH
1100 * set up a shared mapping on a file (the driver or filesystem provides and
1101 * pins the storage)
1da177e4 1102 */
8feae131 1103static int do_mmap_shared_file(struct vm_area_struct *vma)
1da177e4
LT
1104{
1105 int ret;
1106
1107 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
dd8632a1
PM
1108 if (ret == 0) {
1109 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 1110 return 0;
dd8632a1 1111 }
1da177e4
LT
1112 if (ret != -ENOSYS)
1113 return ret;
1114
3fa30460
DH
1115 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1116 * opposed to tried but failed) so we can only give a suitable error as
1117 * it's not possible to make a private copy if MAP_SHARED was given */
1da177e4
LT
1118 return -ENODEV;
1119}
1120
1121/*
1122 * set up a private mapping or an anonymous shared mapping
1123 */
8feae131
DH
1124static int do_mmap_private(struct vm_area_struct *vma,
1125 struct vm_region *region,
645d83c5
DH
1126 unsigned long len,
1127 unsigned long capabilities)
1da177e4 1128{
8feae131
DH
1129 struct page *pages;
1130 unsigned long total, point, n, rlen;
1da177e4 1131 void *base;
8feae131 1132 int ret, order;
1da177e4
LT
1133
1134 /* invoke the file's mapping function so that it can keep track of
1135 * shared mappings on devices or memory
1136 * - VM_MAYSHARE will be set if it may attempt to share
1137 */
645d83c5 1138 if (capabilities & BDI_CAP_MAP_DIRECT) {
1da177e4 1139 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
dd8632a1 1140 if (ret == 0) {
1da177e4 1141 /* shouldn't return success if we're not sharing */
dd8632a1
PM
1142 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1143 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 1144 return 0;
1da177e4 1145 }
dd8632a1
PM
1146 if (ret != -ENOSYS)
1147 return ret;
1da177e4
LT
1148
1149 /* getting an ENOSYS error indicates that direct mmap isn't
1150 * possible (as opposed to tried but failed) so we'll try to
1151 * make a private copy of the data and map that instead */
1152 }
1153
8feae131
DH
1154 rlen = PAGE_ALIGN(len);
1155
1da177e4
LT
1156 /* allocate some memory to hold the mapping
1157 * - note that this may not return a page-aligned address if the object
1158 * we're allocating is smaller than a page
1159 */
8feae131
DH
1160 order = get_order(rlen);
1161 kdebug("alloc order %d for %lx", order, len);
1162
1163 pages = alloc_pages(GFP_KERNEL, order);
1164 if (!pages)
1da177e4
LT
1165 goto enomem;
1166
8feae131 1167 total = 1 << order;
33e5d769 1168 atomic_long_add(total, &mmap_pages_allocated);
8feae131
DH
1169
1170 point = rlen >> PAGE_SHIFT;
dd8632a1
PM
1171
1172 /* we allocated a power-of-2 sized page set, so we may want to trim off
1173 * the excess */
1174 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1175 while (total > point) {
1176 order = ilog2(total - point);
1177 n = 1 << order;
1178 kdebug("shave %lu/%lu @%lu", n, total - point, total);
33e5d769 1179 atomic_long_sub(n, &mmap_pages_allocated);
dd8632a1
PM
1180 total -= n;
1181 set_page_refcounted(pages + total);
1182 __free_pages(pages + total, order);
1183 }
8feae131
DH
1184 }
1185
8feae131
DH
1186 for (point = 1; point < total; point++)
1187 set_page_refcounted(&pages[point]);
1da177e4 1188
8feae131
DH
1189 base = page_address(pages);
1190 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1191 region->vm_start = (unsigned long) base;
1192 region->vm_end = region->vm_start + rlen;
dd8632a1 1193 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
8feae131
DH
1194
1195 vma->vm_start = region->vm_start;
1196 vma->vm_end = region->vm_start + len;
1da177e4
LT
1197
1198 if (vma->vm_file) {
1199 /* read the contents of a file into the copy */
1200 mm_segment_t old_fs;
1201 loff_t fpos;
1202
1203 fpos = vma->vm_pgoff;
1204 fpos <<= PAGE_SHIFT;
1205
1206 old_fs = get_fs();
1207 set_fs(KERNEL_DS);
8feae131 1208 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos);
1da177e4
LT
1209 set_fs(old_fs);
1210
1211 if (ret < 0)
1212 goto error_free;
1213
1214 /* clear the last little bit */
8feae131
DH
1215 if (ret < rlen)
1216 memset(base + ret, 0, rlen - ret);
1da177e4 1217
1da177e4
LT
1218 }
1219
1220 return 0;
1221
1222error_free:
8feae131
DH
1223 free_page_series(region->vm_start, region->vm_end);
1224 region->vm_start = vma->vm_start = 0;
1225 region->vm_end = vma->vm_end = 0;
dd8632a1 1226 region->vm_top = 0;
1da177e4
LT
1227 return ret;
1228
1229enomem:
05ae6fa3
GU
1230 printk("Allocation of length %lu from process %d (%s) failed\n",
1231 len, current->pid, current->comm);
7bf02ea2 1232 show_free_areas(0);
1da177e4
LT
1233 return -ENOMEM;
1234}
1235
1236/*
1237 * handle mapping creation for uClinux
1238 */
1239unsigned long do_mmap_pgoff(struct file *file,
1240 unsigned long addr,
1241 unsigned long len,
1242 unsigned long prot,
1243 unsigned long flags,
1244 unsigned long pgoff)
1245{
8feae131
DH
1246 struct vm_area_struct *vma;
1247 struct vm_region *region;
1da177e4 1248 struct rb_node *rb;
8feae131 1249 unsigned long capabilities, vm_flags, result;
1da177e4
LT
1250 int ret;
1251
8feae131
DH
1252 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1253
1da177e4
LT
1254 /* decide whether we should attempt the mapping, and if so what sort of
1255 * mapping */
1256 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1257 &capabilities);
8feae131
DH
1258 if (ret < 0) {
1259 kleave(" = %d [val]", ret);
1da177e4 1260 return ret;
8feae131 1261 }
1da177e4 1262
06aab5a3
DH
1263 /* we ignore the address hint */
1264 addr = 0;
1265
1da177e4
LT
1266 /* we've determined that we can make the mapping, now translate what we
1267 * now know into VMA flags */
1268 vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1269
8feae131
DH
1270 /* we're going to need to record the mapping */
1271 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1272 if (!region)
1273 goto error_getting_region;
1274
1275 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1276 if (!vma)
1277 goto error_getting_vma;
1da177e4 1278
1e2ae599 1279 region->vm_usage = 1;
8feae131
DH
1280 region->vm_flags = vm_flags;
1281 region->vm_pgoff = pgoff;
1282
5beb4930 1283 INIT_LIST_HEAD(&vma->anon_vma_chain);
8feae131
DH
1284 vma->vm_flags = vm_flags;
1285 vma->vm_pgoff = pgoff;
1da177e4 1286
8feae131
DH
1287 if (file) {
1288 region->vm_file = file;
1289 get_file(file);
1290 vma->vm_file = file;
1291 get_file(file);
1292 if (vm_flags & VM_EXECUTABLE) {
1293 added_exe_file_vma(current->mm);
1294 vma->vm_mm = current->mm;
1295 }
1296 }
1297
1298 down_write(&nommu_region_sem);
1299
1300 /* if we want to share, we need to check for regions created by other
1da177e4 1301 * mmap() calls that overlap with our proposed mapping
8feae131 1302 * - we can only share with a superset match on most regular files
1da177e4
LT
1303 * - shared mappings on character devices and memory backed files are
1304 * permitted to overlap inexactly as far as we are concerned for in
1305 * these cases, sharing is handled in the driver or filesystem rather
1306 * than here
1307 */
1308 if (vm_flags & VM_MAYSHARE) {
8feae131
DH
1309 struct vm_region *pregion;
1310 unsigned long pglen, rpglen, pgend, rpgend, start;
1da177e4 1311
8feae131
DH
1312 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1313 pgend = pgoff + pglen;
165b2392 1314
8feae131
DH
1315 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1316 pregion = rb_entry(rb, struct vm_region, vm_rb);
1da177e4 1317
8feae131 1318 if (!(pregion->vm_flags & VM_MAYSHARE))
1da177e4
LT
1319 continue;
1320
1321 /* search for overlapping mappings on the same file */
8feae131
DH
1322 if (pregion->vm_file->f_path.dentry->d_inode !=
1323 file->f_path.dentry->d_inode)
1da177e4
LT
1324 continue;
1325
8feae131 1326 if (pregion->vm_pgoff >= pgend)
1da177e4
LT
1327 continue;
1328
8feae131
DH
1329 rpglen = pregion->vm_end - pregion->vm_start;
1330 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1331 rpgend = pregion->vm_pgoff + rpglen;
1332 if (pgoff >= rpgend)
1da177e4
LT
1333 continue;
1334
8feae131
DH
1335 /* handle inexactly overlapping matches between
1336 * mappings */
1337 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1338 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1339 /* new mapping is not a subset of the region */
1da177e4
LT
1340 if (!(capabilities & BDI_CAP_MAP_DIRECT))
1341 goto sharing_violation;
1342 continue;
1343 }
1344
8feae131 1345 /* we've found a region we can share */
1e2ae599 1346 pregion->vm_usage++;
8feae131
DH
1347 vma->vm_region = pregion;
1348 start = pregion->vm_start;
1349 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1350 vma->vm_start = start;
1351 vma->vm_end = start + len;
1352
1353 if (pregion->vm_flags & VM_MAPPED_COPY) {
1354 kdebug("share copy");
1355 vma->vm_flags |= VM_MAPPED_COPY;
1356 } else {
1357 kdebug("share mmap");
1358 ret = do_mmap_shared_file(vma);
1359 if (ret < 0) {
1360 vma->vm_region = NULL;
1361 vma->vm_start = 0;
1362 vma->vm_end = 0;
1e2ae599 1363 pregion->vm_usage--;
8feae131
DH
1364 pregion = NULL;
1365 goto error_just_free;
1366 }
1367 }
1368 fput(region->vm_file);
1369 kmem_cache_free(vm_region_jar, region);
1370 region = pregion;
1371 result = start;
1372 goto share;
1da177e4
LT
1373 }
1374
1da177e4
LT
1375 /* obtain the address at which to make a shared mapping
1376 * - this is the hook for quasi-memory character devices to
1377 * tell us the location of a shared mapping
1378 */
645d83c5 1379 if (capabilities & BDI_CAP_MAP_DIRECT) {
1da177e4
LT
1380 addr = file->f_op->get_unmapped_area(file, addr, len,
1381 pgoff, flags);
1382 if (IS_ERR((void *) addr)) {
1383 ret = addr;
1384 if (ret != (unsigned long) -ENOSYS)
8feae131 1385 goto error_just_free;
1da177e4
LT
1386
1387 /* the driver refused to tell us where to site
1388 * the mapping so we'll have to attempt to copy
1389 * it */
1390 ret = (unsigned long) -ENODEV;
1391 if (!(capabilities & BDI_CAP_MAP_COPY))
8feae131 1392 goto error_just_free;
1da177e4
LT
1393
1394 capabilities &= ~BDI_CAP_MAP_DIRECT;
8feae131
DH
1395 } else {
1396 vma->vm_start = region->vm_start = addr;
1397 vma->vm_end = region->vm_end = addr + len;
1da177e4
LT
1398 }
1399 }
1400 }
1401
8feae131 1402 vma->vm_region = region;
1da177e4 1403
645d83c5
DH
1404 /* set up the mapping
1405 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1406 */
1da177e4 1407 if (file && vma->vm_flags & VM_SHARED)
8feae131 1408 ret = do_mmap_shared_file(vma);
1da177e4 1409 else
645d83c5 1410 ret = do_mmap_private(vma, region, len, capabilities);
1da177e4 1411 if (ret < 0)
645d83c5
DH
1412 goto error_just_free;
1413 add_nommu_region(region);
8feae131 1414
ea637639
JZ
1415 /* clear anonymous mappings that don't ask for uninitialized data */
1416 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1417 memset((void *)region->vm_start, 0,
1418 region->vm_end - region->vm_start);
1419
1da177e4 1420 /* okay... we have a mapping; now we have to register it */
8feae131 1421 result = vma->vm_start;
1da177e4 1422
1da177e4
LT
1423 current->mm->total_vm += len >> PAGE_SHIFT;
1424
8feae131
DH
1425share:
1426 add_vma_to_mm(current->mm, vma);
1da177e4 1427
cfe79c00
MF
1428 /* we flush the region from the icache only when the first executable
1429 * mapping of it is made */
1430 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1431 flush_icache_range(region->vm_start, region->vm_end);
1432 region->vm_icache_flushed = true;
1433 }
1da177e4 1434
cfe79c00 1435 up_write(&nommu_region_sem);
1da177e4 1436
8feae131
DH
1437 kleave(" = %lx", result);
1438 return result;
1da177e4 1439
8feae131
DH
1440error_just_free:
1441 up_write(&nommu_region_sem);
1442error:
89a86402
DH
1443 if (region->vm_file)
1444 fput(region->vm_file);
8feae131 1445 kmem_cache_free(vm_region_jar, region);
89a86402
DH
1446 if (vma->vm_file)
1447 fput(vma->vm_file);
8feae131
DH
1448 if (vma->vm_flags & VM_EXECUTABLE)
1449 removed_exe_file_vma(vma->vm_mm);
1450 kmem_cache_free(vm_area_cachep, vma);
1451 kleave(" = %d", ret);
1452 return ret;
1453
1454sharing_violation:
1455 up_write(&nommu_region_sem);
1456 printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1457 ret = -EINVAL;
1458 goto error;
1da177e4 1459
8feae131
DH
1460error_getting_vma:
1461 kmem_cache_free(vm_region_jar, region);
1462 printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1463 " from process %d failed\n",
1da177e4 1464 len, current->pid);
7bf02ea2 1465 show_free_areas(0);
1da177e4
LT
1466 return -ENOMEM;
1467
8feae131
DH
1468error_getting_region:
1469 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1470 " from process %d failed\n",
1da177e4 1471 len, current->pid);
7bf02ea2 1472 show_free_areas(0);
1da177e4
LT
1473 return -ENOMEM;
1474}
b5073173 1475EXPORT_SYMBOL(do_mmap_pgoff);
1da177e4 1476
66f0dc48
HD
1477SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1478 unsigned long, prot, unsigned long, flags,
1479 unsigned long, fd, unsigned long, pgoff)
1480{
1481 struct file *file = NULL;
1482 unsigned long retval = -EBADF;
1483
120a795d 1484 audit_mmap_fd(fd, flags);
66f0dc48
HD
1485 if (!(flags & MAP_ANONYMOUS)) {
1486 file = fget(fd);
1487 if (!file)
1488 goto out;
1489 }
1490
1491 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1492
1493 down_write(&current->mm->mmap_sem);
1494 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1495 up_write(&current->mm->mmap_sem);
1496
1497 if (file)
1498 fput(file);
1499out:
1500 return retval;
1501}
1502
a4679373
CH
1503#ifdef __ARCH_WANT_SYS_OLD_MMAP
1504struct mmap_arg_struct {
1505 unsigned long addr;
1506 unsigned long len;
1507 unsigned long prot;
1508 unsigned long flags;
1509 unsigned long fd;
1510 unsigned long offset;
1511};
1512
1513SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1514{
1515 struct mmap_arg_struct a;
1516
1517 if (copy_from_user(&a, arg, sizeof(a)))
1518 return -EFAULT;
1519 if (a.offset & ~PAGE_MASK)
1520 return -EINVAL;
1521
1522 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1523 a.offset >> PAGE_SHIFT);
1524}
1525#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1526
1da177e4 1527/*
8feae131
DH
1528 * split a vma into two pieces at address 'addr', a new vma is allocated either
1529 * for the first part or the tail.
1da177e4 1530 */
8feae131
DH
1531int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1532 unsigned long addr, int new_below)
1da177e4 1533{
8feae131
DH
1534 struct vm_area_struct *new;
1535 struct vm_region *region;
1536 unsigned long npages;
1da177e4 1537
8feae131 1538 kenter("");
1da177e4 1539
779c1023
DH
1540 /* we're only permitted to split anonymous regions (these should have
1541 * only a single usage on the region) */
1542 if (vma->vm_file)
8feae131 1543 return -ENOMEM;
1da177e4 1544
8feae131
DH
1545 if (mm->map_count >= sysctl_max_map_count)
1546 return -ENOMEM;
1da177e4 1547
8feae131
DH
1548 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1549 if (!region)
1550 return -ENOMEM;
1da177e4 1551
8feae131
DH
1552 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1553 if (!new) {
1554 kmem_cache_free(vm_region_jar, region);
1555 return -ENOMEM;
1556 }
1557
1558 /* most fields are the same, copy all, and then fixup */
1559 *new = *vma;
1560 *region = *vma->vm_region;
1561 new->vm_region = region;
1562
1563 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1564
1565 if (new_below) {
dd8632a1 1566 region->vm_top = region->vm_end = new->vm_end = addr;
8feae131
DH
1567 } else {
1568 region->vm_start = new->vm_start = addr;
1569 region->vm_pgoff = new->vm_pgoff += npages;
1da177e4 1570 }
8feae131
DH
1571
1572 if (new->vm_ops && new->vm_ops->open)
1573 new->vm_ops->open(new);
1574
1575 delete_vma_from_mm(vma);
1576 down_write(&nommu_region_sem);
1577 delete_nommu_region(vma->vm_region);
1578 if (new_below) {
1579 vma->vm_region->vm_start = vma->vm_start = addr;
1580 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1581 } else {
1582 vma->vm_region->vm_end = vma->vm_end = addr;
dd8632a1 1583 vma->vm_region->vm_top = addr;
8feae131
DH
1584 }
1585 add_nommu_region(vma->vm_region);
1586 add_nommu_region(new->vm_region);
1587 up_write(&nommu_region_sem);
1588 add_vma_to_mm(mm, vma);
1589 add_vma_to_mm(mm, new);
1590 return 0;
1da177e4
LT
1591}
1592
3034097a 1593/*
8feae131
DH
1594 * shrink a VMA by removing the specified chunk from either the beginning or
1595 * the end
3034097a 1596 */
8feae131
DH
1597static int shrink_vma(struct mm_struct *mm,
1598 struct vm_area_struct *vma,
1599 unsigned long from, unsigned long to)
1da177e4 1600{
8feae131 1601 struct vm_region *region;
1da177e4 1602
8feae131 1603 kenter("");
1da177e4 1604
8feae131
DH
1605 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1606 * and list */
1607 delete_vma_from_mm(vma);
1608 if (from > vma->vm_start)
1609 vma->vm_end = from;
1610 else
1611 vma->vm_start = to;
1612 add_vma_to_mm(mm, vma);
1da177e4 1613
8feae131
DH
1614 /* cut the backing region down to size */
1615 region = vma->vm_region;
1e2ae599 1616 BUG_ON(region->vm_usage != 1);
8feae131
DH
1617
1618 down_write(&nommu_region_sem);
1619 delete_nommu_region(region);
dd8632a1
PM
1620 if (from > region->vm_start) {
1621 to = region->vm_top;
1622 region->vm_top = region->vm_end = from;
1623 } else {
8feae131 1624 region->vm_start = to;
dd8632a1 1625 }
8feae131
DH
1626 add_nommu_region(region);
1627 up_write(&nommu_region_sem);
1628
1629 free_page_series(from, to);
1630 return 0;
1631}
1da177e4 1632
8feae131
DH
1633/*
1634 * release a mapping
1635 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1636 * VMA, though it need not cover the whole VMA
1637 */
1638int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1639{
1640 struct vm_area_struct *vma;
1641 struct rb_node *rb;
1642 unsigned long end = start + len;
1643 int ret;
1da177e4 1644
8feae131 1645 kenter(",%lx,%zx", start, len);
1da177e4 1646
8feae131
DH
1647 if (len == 0)
1648 return -EINVAL;
365e9c87 1649
8feae131
DH
1650 /* find the first potentially overlapping VMA */
1651 vma = find_vma(mm, start);
1652 if (!vma) {
33e5d769
DH
1653 static int limit = 0;
1654 if (limit < 5) {
1655 printk(KERN_WARNING
1656 "munmap of memory not mmapped by process %d"
1657 " (%s): 0x%lx-0x%lx\n",
1658 current->pid, current->comm,
1659 start, start + len - 1);
1660 limit++;
1661 }
8feae131
DH
1662 return -EINVAL;
1663 }
1da177e4 1664
8feae131
DH
1665 /* we're allowed to split an anonymous VMA but not a file-backed one */
1666 if (vma->vm_file) {
1667 do {
1668 if (start > vma->vm_start) {
1669 kleave(" = -EINVAL [miss]");
1670 return -EINVAL;
1671 }
1672 if (end == vma->vm_end)
1673 goto erase_whole_vma;
1674 rb = rb_next(&vma->vm_rb);
1675 vma = rb_entry(rb, struct vm_area_struct, vm_rb);
1676 } while (rb);
1677 kleave(" = -EINVAL [split file]");
1678 return -EINVAL;
1679 } else {
1680 /* the chunk must be a subset of the VMA found */
1681 if (start == vma->vm_start && end == vma->vm_end)
1682 goto erase_whole_vma;
1683 if (start < vma->vm_start || end > vma->vm_end) {
1684 kleave(" = -EINVAL [superset]");
1685 return -EINVAL;
1686 }
1687 if (start & ~PAGE_MASK) {
1688 kleave(" = -EINVAL [unaligned start]");
1689 return -EINVAL;
1690 }
1691 if (end != vma->vm_end && end & ~PAGE_MASK) {
1692 kleave(" = -EINVAL [unaligned split]");
1693 return -EINVAL;
1694 }
1695 if (start != vma->vm_start && end != vma->vm_end) {
1696 ret = split_vma(mm, vma, start, 1);
1697 if (ret < 0) {
1698 kleave(" = %d [split]", ret);
1699 return ret;
1700 }
1701 }
1702 return shrink_vma(mm, vma, start, end);
1703 }
1da177e4 1704
8feae131
DH
1705erase_whole_vma:
1706 delete_vma_from_mm(vma);
1707 delete_vma(mm, vma);
1708 kleave(" = 0");
1da177e4
LT
1709 return 0;
1710}
b5073173 1711EXPORT_SYMBOL(do_munmap);
1da177e4 1712
6a6160a7 1713SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
3034097a
DH
1714{
1715 int ret;
1716 struct mm_struct *mm = current->mm;
1717
1718 down_write(&mm->mmap_sem);
1719 ret = do_munmap(mm, addr, len);
1720 up_write(&mm->mmap_sem);
1721 return ret;
1722}
1723
1724/*
8feae131 1725 * release all the mappings made in a process's VM space
3034097a 1726 */
8feae131 1727void exit_mmap(struct mm_struct *mm)
1da177e4 1728{
8feae131 1729 struct vm_area_struct *vma;
1da177e4 1730
8feae131
DH
1731 if (!mm)
1732 return;
1da177e4 1733
8feae131 1734 kenter("");
1da177e4 1735
8feae131 1736 mm->total_vm = 0;
1da177e4 1737
8feae131
DH
1738 while ((vma = mm->mmap)) {
1739 mm->mmap = vma->vm_next;
1740 delete_vma_from_mm(vma);
1741 delete_vma(mm, vma);
04c34961 1742 cond_resched();
1da177e4 1743 }
8feae131
DH
1744
1745 kleave("");
1da177e4
LT
1746}
1747
1da177e4
LT
1748unsigned long do_brk(unsigned long addr, unsigned long len)
1749{
1750 return -ENOMEM;
1751}
1752
1753/*
6fa5f80b
DH
1754 * expand (or shrink) an existing mapping, potentially moving it at the same
1755 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1da177e4 1756 *
6fa5f80b 1757 * under NOMMU conditions, we only permit changing a mapping's size, and only
8feae131
DH
1758 * as long as it stays within the region allocated by do_mmap_private() and the
1759 * block is not shareable
1da177e4 1760 *
6fa5f80b 1761 * MREMAP_FIXED is not supported under NOMMU conditions
1da177e4
LT
1762 */
1763unsigned long do_mremap(unsigned long addr,
1764 unsigned long old_len, unsigned long new_len,
1765 unsigned long flags, unsigned long new_addr)
1766{
6fa5f80b 1767 struct vm_area_struct *vma;
1da177e4
LT
1768
1769 /* insanity checks first */
8feae131 1770 if (old_len == 0 || new_len == 0)
1da177e4
LT
1771 return (unsigned long) -EINVAL;
1772
8feae131
DH
1773 if (addr & ~PAGE_MASK)
1774 return -EINVAL;
1775
1da177e4
LT
1776 if (flags & MREMAP_FIXED && new_addr != addr)
1777 return (unsigned long) -EINVAL;
1778
8feae131 1779 vma = find_vma_exact(current->mm, addr, old_len);
6fa5f80b
DH
1780 if (!vma)
1781 return (unsigned long) -EINVAL;
1da177e4 1782
6fa5f80b 1783 if (vma->vm_end != vma->vm_start + old_len)
1da177e4
LT
1784 return (unsigned long) -EFAULT;
1785
6fa5f80b 1786 if (vma->vm_flags & VM_MAYSHARE)
1da177e4
LT
1787 return (unsigned long) -EPERM;
1788
8feae131 1789 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1da177e4
LT
1790 return (unsigned long) -ENOMEM;
1791
1792 /* all checks complete - do it */
6fa5f80b 1793 vma->vm_end = vma->vm_start + new_len;
6fa5f80b
DH
1794 return vma->vm_start;
1795}
b5073173 1796EXPORT_SYMBOL(do_mremap);
6fa5f80b 1797
6a6160a7
HC
1798SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1799 unsigned long, new_len, unsigned long, flags,
1800 unsigned long, new_addr)
6fa5f80b
DH
1801{
1802 unsigned long ret;
1803
1804 down_write(&current->mm->mmap_sem);
1805 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1806 up_write(&current->mm->mmap_sem);
1807 return ret;
1da177e4
LT
1808}
1809
6aab341e 1810struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
deceb6cd 1811 unsigned int foll_flags)
1da177e4
LT
1812{
1813 return NULL;
1814}
1815
1da177e4
LT
1816int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
1817 unsigned long to, unsigned long size, pgprot_t prot)
1818{
66aa2b4b
GU
1819 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT;
1820 return 0;
1da177e4 1821}
22c4af40 1822EXPORT_SYMBOL(remap_pfn_range);
1da177e4 1823
f905bc44
PM
1824int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1825 unsigned long pgoff)
1826{
1827 unsigned int size = vma->vm_end - vma->vm_start;
1828
1829 if (!(vma->vm_flags & VM_USERMAP))
1830 return -EINVAL;
1831
1832 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1833 vma->vm_end = vma->vm_start + size;
1834
1835 return 0;
1836}
1837EXPORT_SYMBOL(remap_vmalloc_range);
1838
1da177e4
LT
1839unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1840 unsigned long len, unsigned long pgoff, unsigned long flags)
1841{
1842 return -ENOMEM;
1843}
1844
1363c3cd 1845void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1da177e4
LT
1846{
1847}
1848
1da177e4
LT
1849void unmap_mapping_range(struct address_space *mapping,
1850 loff_t const holebegin, loff_t const holelen,
1851 int even_cows)
1852{
1853}
22c4af40 1854EXPORT_SYMBOL(unmap_mapping_range);
1da177e4
LT
1855
1856/*
1857 * Check that a process has enough memory to allocate a new virtual
1858 * mapping. 0 means there is enough memory for the allocation to
1859 * succeed and -ENOMEM implies there is not.
1860 *
1861 * We currently support three overcommit policies, which are set via the
1862 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1863 *
1864 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1865 * Additional code 2002 Jul 20 by Robert Love.
1866 *
1867 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1868 *
1869 * Note this is a helper function intended to be used by LSMs which
1870 * wish to use this logic.
1871 */
34b4e4aa 1872int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1da177e4
LT
1873{
1874 unsigned long free, allowed;
1875
1876 vm_acct_memory(pages);
1877
1878 /*
1879 * Sometimes we want to use more memory than we have
1880 */
1881 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1882 return 0;
1883
1884 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1885 unsigned long n;
1886
347ce434 1887 free = global_page_state(NR_FILE_PAGES);
1da177e4
LT
1888 free += nr_swap_pages;
1889
1890 /*
1891 * Any slabs which are created with the
1892 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1893 * which are reclaimable, under pressure. The dentry
1894 * cache and most inode caches should fall into this
1895 */
972d1a7b 1896 free += global_page_state(NR_SLAB_RECLAIMABLE);
1da177e4
LT
1897
1898 /*
1899 * Leave the last 3% for root
1900 */
1901 if (!cap_sys_admin)
1902 free -= free / 32;
1903
1904 if (free > pages)
1905 return 0;
1906
1907 /*
1908 * nr_free_pages() is very expensive on large systems,
1909 * only call if we're about to fail.
1910 */
1911 n = nr_free_pages();
d5ddc79b
HA
1912
1913 /*
1914 * Leave reserved pages. The pages are not for anonymous pages.
1915 */
1916 if (n <= totalreserve_pages)
1917 goto error;
1918 else
1919 n -= totalreserve_pages;
1920
1921 /*
1922 * Leave the last 3% for root
1923 */
1da177e4
LT
1924 if (!cap_sys_admin)
1925 n -= n / 32;
1926 free += n;
1927
1928 if (free > pages)
1929 return 0;
d5ddc79b
HA
1930
1931 goto error;
1da177e4
LT
1932 }
1933
1934 allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1935 /*
1936 * Leave the last 3% for root
1937 */
1938 if (!cap_sys_admin)
1939 allowed -= allowed / 32;
1940 allowed += total_swap_pages;
1941
1942 /* Don't let a single process grow too big:
1943 leave 3% of the size of this process for other processes */
731572d3
AC
1944 if (mm)
1945 allowed -= mm->total_vm / 32;
1da177e4 1946
00a62ce9 1947 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1da177e4 1948 return 0;
00a62ce9 1949
d5ddc79b 1950error:
1da177e4
LT
1951 vm_unacct_memory(pages);
1952
1953 return -ENOMEM;
1954}
1955
cae5d390 1956int in_gate_area_no_mm(unsigned long addr)
1da177e4
LT
1957{
1958 return 0;
1959}
b0e15190 1960
d0217ac0 1961int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
b0e15190
DH
1962{
1963 BUG();
d0217ac0 1964 return 0;
b0e15190 1965}
b5073173 1966EXPORT_SYMBOL(filemap_fault);
0ec76a11 1967
f55f199b
MF
1968static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1969 unsigned long addr, void *buf, int len, int write)
0ec76a11 1970{
0ec76a11 1971 struct vm_area_struct *vma;
0ec76a11
DH
1972
1973 down_read(&mm->mmap_sem);
1974
1975 /* the access must start within one of the target process's mappings */
0159b141
DH
1976 vma = find_vma(mm, addr);
1977 if (vma) {
0ec76a11
DH
1978 /* don't overrun this mapping */
1979 if (addr + len >= vma->vm_end)
1980 len = vma->vm_end - addr;
1981
1982 /* only read or write mappings where it is permitted */
d00c7b99 1983 if (write && vma->vm_flags & VM_MAYWRITE)
7959722b
JZ
1984 copy_to_user_page(vma, NULL, addr,
1985 (void *) addr, buf, len);
d00c7b99 1986 else if (!write && vma->vm_flags & VM_MAYREAD)
7959722b
JZ
1987 copy_from_user_page(vma, NULL, addr,
1988 buf, (void *) addr, len);
0ec76a11
DH
1989 else
1990 len = 0;
1991 } else {
1992 len = 0;
1993 }
1994
1995 up_read(&mm->mmap_sem);
f55f199b
MF
1996
1997 return len;
1998}
1999
2000/**
2001 * @access_remote_vm - access another process' address space
2002 * @mm: the mm_struct of the target address space
2003 * @addr: start address to access
2004 * @buf: source or destination buffer
2005 * @len: number of bytes to transfer
2006 * @write: whether the access is a write
2007 *
2008 * The caller must hold a reference on @mm.
2009 */
2010int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2011 void *buf, int len, int write)
2012{
2013 return __access_remote_vm(NULL, mm, addr, buf, len, write);
2014}
2015
2016/*
2017 * Access another process' address space.
2018 * - source/target buffer must be kernel space
2019 */
2020int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2021{
2022 struct mm_struct *mm;
2023
2024 if (addr + len < addr)
2025 return 0;
2026
2027 mm = get_task_mm(tsk);
2028 if (!mm)
2029 return 0;
2030
2031 len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2032
0ec76a11
DH
2033 mmput(mm);
2034 return len;
2035}
7e660872
DH
2036
2037/**
2038 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2039 * @inode: The inode to check
2040 * @size: The current filesize of the inode
2041 * @newsize: The proposed filesize of the inode
2042 *
2043 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2044 * make sure that that any outstanding VMAs aren't broken and then shrink the
2045 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2046 * automatically grant mappings that are too large.
2047 */
2048int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2049 size_t newsize)
2050{
2051 struct vm_area_struct *vma;
2052 struct prio_tree_iter iter;
2053 struct vm_region *region;
2054 pgoff_t low, high;
2055 size_t r_size, r_top;
2056
2057 low = newsize >> PAGE_SHIFT;
2058 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2059
2060 down_write(&nommu_region_sem);
2061
2062 /* search for VMAs that fall within the dead zone */
2063 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2064 low, high) {
2065 /* found one - only interested if it's shared out of the page
2066 * cache */
2067 if (vma->vm_flags & VM_SHARED) {
2068 up_write(&nommu_region_sem);
2069 return -ETXTBSY; /* not quite true, but near enough */
2070 }
2071 }
2072
2073 /* reduce any regions that overlap the dead zone - if in existence,
2074 * these will be pointed to by VMAs that don't overlap the dead zone
2075 *
2076 * we don't check for any regions that start beyond the EOF as there
2077 * shouldn't be any
2078 */
2079 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap,
2080 0, ULONG_MAX) {
2081 if (!(vma->vm_flags & VM_SHARED))
2082 continue;
2083
2084 region = vma->vm_region;
2085 r_size = region->vm_top - region->vm_start;
2086 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2087
2088 if (r_top > newsize) {
2089 region->vm_top -= r_top - newsize;
2090 if (region->vm_end > region->vm_top)
2091 region->vm_end = region->vm_top;
2092 }
2093 }
2094
2095 up_write(&nommu_region_sem);
2096 return 0;
2097}