]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - mm/nommu.c
Merge tag 'pinctrl-v4.10-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
[mirror_ubuntu-bionic-kernel.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
b1de0d13
MH
16#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17
b95f1b31 18#include <linux/export.h>
1da177e4 19#include <linux/mm.h>
615d6e87 20#include <linux/vmacache.h>
1da177e4
LT
21#include <linux/mman.h>
22#include <linux/swap.h>
23#include <linux/file.h>
24#include <linux/highmem.h>
25#include <linux/pagemap.h>
26#include <linux/slab.h>
27#include <linux/vmalloc.h>
1da177e4
LT
28#include <linux/blkdev.h>
29#include <linux/backing-dev.h>
3b32123d 30#include <linux/compiler.h>
1da177e4
LT
31#include <linux/mount.h>
32#include <linux/personality.h>
33#include <linux/security.h>
34#include <linux/syscalls.h>
120a795d 35#include <linux/audit.h>
b1de0d13 36#include <linux/printk.h>
1da177e4 37
7c0f6ba6 38#include <linux/uaccess.h>
1da177e4
LT
39#include <asm/tlb.h>
40#include <asm/tlbflush.h>
eb8cdec4 41#include <asm/mmu_context.h>
8feae131
DH
42#include "internal.h"
43
1da177e4 44void *high_memory;
944b6874 45EXPORT_SYMBOL(high_memory);
1da177e4
LT
46struct page *mem_map;
47unsigned long max_mapnr;
5b8bf307 48EXPORT_SYMBOL(max_mapnr);
4266c97a 49unsigned long highest_memmap_pfn;
fc4d5c29 50int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
1da177e4
LT
51int heap_stack_gap = 0;
52
33e5d769 53atomic_long_t mmap_pages_allocated;
8feae131 54
1da177e4 55EXPORT_SYMBOL(mem_map);
1da177e4 56
8feae131
DH
57/* list of mapped, potentially shareable regions */
58static struct kmem_cache *vm_region_jar;
59struct rb_root nommu_region_tree = RB_ROOT;
60DECLARE_RWSEM(nommu_region_sem);
1da177e4 61
f0f37e2f 62const struct vm_operations_struct generic_file_vm_ops = {
1da177e4
LT
63};
64
1da177e4
LT
65/*
66 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
68 *
69 * Doesn't have to be accurate, i.e. may have races.
70 */
71unsigned int kobjsize(const void *objp)
72{
73 struct page *page;
74
4016a139
MH
75 /*
76 * If the object we have should not have ksize performed on it,
77 * return size of 0
78 */
5a1603be 79 if (!objp || !virt_addr_valid(objp))
6cfd53fc
PM
80 return 0;
81
82 page = virt_to_head_page(objp);
6cfd53fc
PM
83
84 /*
85 * If the allocator sets PageSlab, we know the pointer came from
86 * kmalloc().
87 */
1da177e4
LT
88 if (PageSlab(page))
89 return ksize(objp);
90
ab2e83ea
PM
91 /*
92 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
96 */
97 if (!PageCompound(page)) {
98 struct vm_area_struct *vma;
99
100 vma = find_vma(current->mm, (unsigned long)objp);
101 if (vma)
102 return vma->vm_end - vma->vm_start;
103 }
104
6cfd53fc
PM
105 /*
106 * The ksize() function is only guaranteed to work for pointers
5a1603be 107 * returned by kmalloc(). So handle arbitrary pointers here.
6cfd53fc 108 */
5a1603be 109 return PAGE_SIZE << compound_order(page);
1da177e4
LT
110}
111
0d731759 112static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
28a35716
ML
113 unsigned long start, unsigned long nr_pages,
114 unsigned int foll_flags, struct page **pages,
115 struct vm_area_struct **vmas, int *nonblocking)
1da177e4 116{
910e46da 117 struct vm_area_struct *vma;
7b4d5b8b
DH
118 unsigned long vm_flags;
119 int i;
120
121 /* calculate required read or write permissions.
58fa879e 122 * If FOLL_FORCE is set, we only require the "MAY" flags.
7b4d5b8b 123 */
58fa879e
HD
124 vm_flags = (foll_flags & FOLL_WRITE) ?
125 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
126 vm_flags &= (foll_flags & FOLL_FORCE) ?
127 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1da177e4 128
9d73777e 129 for (i = 0; i < nr_pages; i++) {
7561e8ca 130 vma = find_vma(mm, start);
7b4d5b8b
DH
131 if (!vma)
132 goto finish_or_fault;
133
134 /* protect what we can, including chardevs */
1c3aff1c
HD
135 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
136 !(vm_flags & vma->vm_flags))
7b4d5b8b 137 goto finish_or_fault;
910e46da 138
1da177e4
LT
139 if (pages) {
140 pages[i] = virt_to_page(start);
141 if (pages[i])
09cbfeaf 142 get_page(pages[i]);
1da177e4
LT
143 }
144 if (vmas)
910e46da 145 vmas[i] = vma;
e1ee65d8 146 start = (start + PAGE_SIZE) & PAGE_MASK;
1da177e4 147 }
7b4d5b8b
DH
148
149 return i;
150
151finish_or_fault:
152 return i ? : -EFAULT;
1da177e4 153}
b291f000 154
b291f000
NP
155/*
156 * get a list of pages in an address range belonging to the specified process
157 * and indicate the VMA that covers each page
158 * - this is potentially dodgy as we may end incrementing the page count of a
159 * slab page or a secondary page from a compound page
160 * - don't permit access to VMAs that don't support it, such as I/O mappings
161 */
c12d2da5 162long get_user_pages(unsigned long start, unsigned long nr_pages,
768ae309 163 unsigned int gup_flags, struct page **pages,
28a35716 164 struct vm_area_struct **vmas)
b291f000 165{
768ae309
LS
166 return __get_user_pages(current, current->mm, start, nr_pages,
167 gup_flags, pages, vmas, NULL);
b291f000 168}
c12d2da5 169EXPORT_SYMBOL(get_user_pages);
66aa2b4b 170
c12d2da5 171long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
3b913179 172 unsigned int gup_flags, struct page **pages,
cde70140 173 int *locked)
f0818f47 174{
768ae309 175 return get_user_pages(start, nr_pages, gup_flags, pages, NULL);
f0818f47 176}
c12d2da5 177EXPORT_SYMBOL(get_user_pages_locked);
f0818f47 178
8b7457ef
LS
179static long __get_user_pages_unlocked(struct task_struct *tsk,
180 struct mm_struct *mm, unsigned long start,
181 unsigned long nr_pages, struct page **pages,
182 unsigned int gup_flags)
f0818f47
AA
183{
184 long ret;
185 down_read(&mm->mmap_sem);
cde70140
DH
186 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
187 NULL, NULL);
f0818f47
AA
188 up_read(&mm->mmap_sem);
189 return ret;
190}
0fd71a56 191
c12d2da5 192long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
c164154f 193 struct page **pages, unsigned int gup_flags)
0fd71a56 194{
cde70140 195 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
c164154f 196 pages, gup_flags);
0fd71a56 197}
c12d2da5 198EXPORT_SYMBOL(get_user_pages_unlocked);
f0818f47 199
dfc2f91a
PM
200/**
201 * follow_pfn - look up PFN at a user virtual address
202 * @vma: memory mapping
203 * @address: user virtual address
204 * @pfn: location to store found PFN
205 *
206 * Only IO mappings and raw PFN mappings are allowed.
207 *
208 * Returns zero and the pfn at @pfn on success, -ve otherwise.
209 */
210int follow_pfn(struct vm_area_struct *vma, unsigned long address,
211 unsigned long *pfn)
212{
213 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
214 return -EINVAL;
215
216 *pfn = address >> PAGE_SHIFT;
217 return 0;
218}
219EXPORT_SYMBOL(follow_pfn);
220
f1c4069e 221LIST_HEAD(vmap_area_list);
1da177e4 222
b3bdda02 223void vfree(const void *addr)
1da177e4
LT
224{
225 kfree(addr);
226}
b5073173 227EXPORT_SYMBOL(vfree);
1da177e4 228
dd0fc66f 229void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1da177e4
LT
230{
231 /*
8518609d
RD
232 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
233 * returns only a logical address.
1da177e4 234 */
84097518 235 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
1da177e4 236}
b5073173 237EXPORT_SYMBOL(__vmalloc);
1da177e4 238
f905bc44
PM
239void *vmalloc_user(unsigned long size)
240{
241 void *ret;
242
243 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
244 PAGE_KERNEL);
245 if (ret) {
246 struct vm_area_struct *vma;
247
248 down_write(&current->mm->mmap_sem);
249 vma = find_vma(current->mm, (unsigned long)ret);
250 if (vma)
251 vma->vm_flags |= VM_USERMAP;
252 up_write(&current->mm->mmap_sem);
253 }
254
255 return ret;
256}
257EXPORT_SYMBOL(vmalloc_user);
258
b3bdda02 259struct page *vmalloc_to_page(const void *addr)
1da177e4
LT
260{
261 return virt_to_page(addr);
262}
b5073173 263EXPORT_SYMBOL(vmalloc_to_page);
1da177e4 264
b3bdda02 265unsigned long vmalloc_to_pfn(const void *addr)
1da177e4
LT
266{
267 return page_to_pfn(virt_to_page(addr));
268}
b5073173 269EXPORT_SYMBOL(vmalloc_to_pfn);
1da177e4
LT
270
271long vread(char *buf, char *addr, unsigned long count)
272{
9bde916b
CG
273 /* Don't allow overflow */
274 if ((unsigned long) buf + count < count)
275 count = -(unsigned long) buf;
276
1da177e4
LT
277 memcpy(buf, addr, count);
278 return count;
279}
280
281long vwrite(char *buf, char *addr, unsigned long count)
282{
283 /* Don't allow overflow */
284 if ((unsigned long) addr + count < count)
285 count = -(unsigned long) addr;
286
287 memcpy(addr, buf, count);
ac714904 288 return count;
1da177e4
LT
289}
290
291/*
e1c05067 292 * vmalloc - allocate virtually contiguous memory
1da177e4
LT
293 *
294 * @size: allocation size
295 *
296 * Allocate enough pages to cover @size from the page level
e1c05067 297 * allocator and map them into contiguous kernel virtual space.
1da177e4 298 *
c1c8897f 299 * For tight control over page level allocator and protection flags
1da177e4
LT
300 * use __vmalloc() instead.
301 */
302void *vmalloc(unsigned long size)
303{
304 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
305}
f6138882
AM
306EXPORT_SYMBOL(vmalloc);
307
e1ca7788 308/*
e1c05067 309 * vzalloc - allocate virtually contiguous memory with zero fill
e1ca7788
DY
310 *
311 * @size: allocation size
312 *
313 * Allocate enough pages to cover @size from the page level
e1c05067 314 * allocator and map them into contiguous kernel virtual space.
e1ca7788
DY
315 * The memory allocated is set to zero.
316 *
317 * For tight control over page level allocator and protection flags
318 * use __vmalloc() instead.
319 */
320void *vzalloc(unsigned long size)
321{
322 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
323 PAGE_KERNEL);
324}
325EXPORT_SYMBOL(vzalloc);
326
327/**
328 * vmalloc_node - allocate memory on a specific node
329 * @size: allocation size
330 * @node: numa node
331 *
332 * Allocate enough pages to cover @size from the page level
333 * allocator and map them into contiguous kernel virtual space.
334 *
335 * For tight control over page level allocator and protection flags
336 * use __vmalloc() instead.
337 */
f6138882
AM
338void *vmalloc_node(unsigned long size, int node)
339{
340 return vmalloc(size);
341}
9a14f653 342EXPORT_SYMBOL(vmalloc_node);
e1ca7788
DY
343
344/**
345 * vzalloc_node - allocate memory on a specific node with zero fill
346 * @size: allocation size
347 * @node: numa node
348 *
349 * Allocate enough pages to cover @size from the page level
350 * allocator and map them into contiguous kernel virtual space.
351 * The memory allocated is set to zero.
352 *
353 * For tight control over page level allocator and protection flags
354 * use __vmalloc() instead.
355 */
356void *vzalloc_node(unsigned long size, int node)
357{
358 return vzalloc(size);
359}
360EXPORT_SYMBOL(vzalloc_node);
1da177e4 361
1af446ed
PM
362#ifndef PAGE_KERNEL_EXEC
363# define PAGE_KERNEL_EXEC PAGE_KERNEL
364#endif
365
366/**
367 * vmalloc_exec - allocate virtually contiguous, executable memory
368 * @size: allocation size
369 *
370 * Kernel-internal function to allocate enough pages to cover @size
371 * the page level allocator and map them into contiguous and
372 * executable kernel virtual space.
373 *
374 * For tight control over page level allocator and protection flags
375 * use __vmalloc() instead.
376 */
377
378void *vmalloc_exec(unsigned long size)
379{
380 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
381}
382
b5073173
PM
383/**
384 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1da177e4
LT
385 * @size: allocation size
386 *
387 * Allocate enough 32bit PA addressable pages to cover @size from the
e1c05067 388 * page level allocator and map them into contiguous kernel virtual space.
1da177e4
LT
389 */
390void *vmalloc_32(unsigned long size)
391{
392 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
393}
b5073173
PM
394EXPORT_SYMBOL(vmalloc_32);
395
396/**
397 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
398 * @size: allocation size
399 *
400 * The resulting memory area is 32bit addressable and zeroed so it can be
401 * mapped to userspace without leaking data.
f905bc44
PM
402 *
403 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
404 * remap_vmalloc_range() are permissible.
b5073173
PM
405 */
406void *vmalloc_32_user(unsigned long size)
407{
f905bc44
PM
408 /*
409 * We'll have to sort out the ZONE_DMA bits for 64-bit,
410 * but for now this can simply use vmalloc_user() directly.
411 */
412 return vmalloc_user(size);
b5073173
PM
413}
414EXPORT_SYMBOL(vmalloc_32_user);
1da177e4
LT
415
416void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
417{
418 BUG();
419 return NULL;
420}
b5073173 421EXPORT_SYMBOL(vmap);
1da177e4 422
b3bdda02 423void vunmap(const void *addr)
1da177e4
LT
424{
425 BUG();
426}
b5073173 427EXPORT_SYMBOL(vunmap);
1da177e4 428
eb6434d9
PM
429void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
430{
431 BUG();
432 return NULL;
433}
434EXPORT_SYMBOL(vm_map_ram);
435
436void vm_unmap_ram(const void *mem, unsigned int count)
437{
438 BUG();
439}
440EXPORT_SYMBOL(vm_unmap_ram);
441
442void vm_unmap_aliases(void)
443{
444}
445EXPORT_SYMBOL_GPL(vm_unmap_aliases);
446
1eeb66a1
CH
447/*
448 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
449 * have one.
450 */
3b32123d 451void __weak vmalloc_sync_all(void)
1eeb66a1
CH
452{
453}
454
29c185e5
PM
455/**
456 * alloc_vm_area - allocate a range of kernel address space
457 * @size: size of the area
458 *
459 * Returns: NULL on failure, vm_struct on success
460 *
461 * This function reserves a range of kernel address space, and
462 * allocates pagetables to map that range. No actual mappings
463 * are created. If the kernel address space is not shared
464 * between processes, it syncs the pagetable across all
465 * processes.
466 */
cd12909c 467struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
29c185e5
PM
468{
469 BUG();
470 return NULL;
471}
472EXPORT_SYMBOL_GPL(alloc_vm_area);
473
474void free_vm_area(struct vm_struct *area)
475{
476 BUG();
477}
478EXPORT_SYMBOL_GPL(free_vm_area);
479
b5073173
PM
480int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
481 struct page *page)
482{
483 return -EINVAL;
484}
485EXPORT_SYMBOL(vm_insert_page);
486
1da177e4
LT
487/*
488 * sys_brk() for the most part doesn't need the global kernel
489 * lock, except when an application is doing something nasty
490 * like trying to un-brk an area that has already been mapped
491 * to a regular file. in this case, the unmapping will need
492 * to invoke file system routines that need the global lock.
493 */
6a6160a7 494SYSCALL_DEFINE1(brk, unsigned long, brk)
1da177e4
LT
495{
496 struct mm_struct *mm = current->mm;
497
498 if (brk < mm->start_brk || brk > mm->context.end_brk)
499 return mm->brk;
500
501 if (mm->brk == brk)
502 return mm->brk;
503
504 /*
505 * Always allow shrinking brk
506 */
507 if (brk <= mm->brk) {
508 mm->brk = brk;
509 return brk;
510 }
511
512 /*
513 * Ok, looks good - let it rip.
514 */
cfe79c00 515 flush_icache_range(mm->brk, brk);
1da177e4
LT
516 return mm->brk = brk;
517}
518
8feae131
DH
519/*
520 * initialise the VMA and region record slabs
521 */
522void __init mmap_init(void)
1da177e4 523{
00a62ce9
KM
524 int ret;
525
908c7f19 526 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
00a62ce9 527 VM_BUG_ON(ret);
5d097056 528 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT);
1da177e4 529}
1da177e4 530
3034097a 531/*
8feae131
DH
532 * validate the region tree
533 * - the caller must hold the region lock
3034097a 534 */
8feae131
DH
535#ifdef CONFIG_DEBUG_NOMMU_REGIONS
536static noinline void validate_nommu_regions(void)
3034097a 537{
8feae131
DH
538 struct vm_region *region, *last;
539 struct rb_node *p, *lastp;
3034097a 540
8feae131
DH
541 lastp = rb_first(&nommu_region_tree);
542 if (!lastp)
543 return;
544
545 last = rb_entry(lastp, struct vm_region, vm_rb);
c9427bc0
GT
546 BUG_ON(last->vm_end <= last->vm_start);
547 BUG_ON(last->vm_top < last->vm_end);
8feae131
DH
548
549 while ((p = rb_next(lastp))) {
550 region = rb_entry(p, struct vm_region, vm_rb);
551 last = rb_entry(lastp, struct vm_region, vm_rb);
552
c9427bc0
GT
553 BUG_ON(region->vm_end <= region->vm_start);
554 BUG_ON(region->vm_top < region->vm_end);
555 BUG_ON(region->vm_start < last->vm_top);
3034097a 556
8feae131
DH
557 lastp = p;
558 }
3034097a 559}
8feae131 560#else
33e5d769
DH
561static void validate_nommu_regions(void)
562{
563}
8feae131 564#endif
3034097a
DH
565
566/*
8feae131 567 * add a region into the global tree
3034097a 568 */
8feae131 569static void add_nommu_region(struct vm_region *region)
3034097a 570{
8feae131
DH
571 struct vm_region *pregion;
572 struct rb_node **p, *parent;
3034097a 573
8feae131
DH
574 validate_nommu_regions();
575
8feae131
DH
576 parent = NULL;
577 p = &nommu_region_tree.rb_node;
578 while (*p) {
579 parent = *p;
580 pregion = rb_entry(parent, struct vm_region, vm_rb);
581 if (region->vm_start < pregion->vm_start)
582 p = &(*p)->rb_left;
583 else if (region->vm_start > pregion->vm_start)
584 p = &(*p)->rb_right;
585 else if (pregion == region)
586 return;
587 else
588 BUG();
3034097a
DH
589 }
590
8feae131
DH
591 rb_link_node(&region->vm_rb, parent, p);
592 rb_insert_color(&region->vm_rb, &nommu_region_tree);
3034097a 593
8feae131 594 validate_nommu_regions();
3034097a 595}
3034097a 596
930e652a 597/*
8feae131 598 * delete a region from the global tree
930e652a 599 */
8feae131 600static void delete_nommu_region(struct vm_region *region)
930e652a 601{
8feae131 602 BUG_ON(!nommu_region_tree.rb_node);
930e652a 603
8feae131
DH
604 validate_nommu_regions();
605 rb_erase(&region->vm_rb, &nommu_region_tree);
606 validate_nommu_regions();
57c8f63e
GU
607}
608
6fa5f80b 609/*
8feae131 610 * free a contiguous series of pages
6fa5f80b 611 */
8feae131 612static void free_page_series(unsigned long from, unsigned long to)
6fa5f80b 613{
8feae131
DH
614 for (; from < to; from += PAGE_SIZE) {
615 struct page *page = virt_to_page(from);
616
33e5d769 617 atomic_long_dec(&mmap_pages_allocated);
8feae131 618 put_page(page);
6fa5f80b 619 }
6fa5f80b
DH
620}
621
3034097a 622/*
8feae131 623 * release a reference to a region
33e5d769 624 * - the caller must hold the region semaphore for writing, which this releases
dd8632a1 625 * - the region may not have been added to the tree yet, in which case vm_top
8feae131 626 * will equal vm_start
3034097a 627 */
8feae131
DH
628static void __put_nommu_region(struct vm_region *region)
629 __releases(nommu_region_sem)
1da177e4 630{
8feae131 631 BUG_ON(!nommu_region_tree.rb_node);
1da177e4 632
1e2ae599 633 if (--region->vm_usage == 0) {
dd8632a1 634 if (region->vm_top > region->vm_start)
8feae131
DH
635 delete_nommu_region(region);
636 up_write(&nommu_region_sem);
637
638 if (region->vm_file)
639 fput(region->vm_file);
640
641 /* IO memory and memory shared directly out of the pagecache
642 * from ramfs/tmpfs mustn't be released here */
22cc877b 643 if (region->vm_flags & VM_MAPPED_COPY)
dd8632a1 644 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
645 kmem_cache_free(vm_region_jar, region);
646 } else {
647 up_write(&nommu_region_sem);
1da177e4 648 }
8feae131 649}
1da177e4 650
8feae131
DH
651/*
652 * release a reference to a region
653 */
654static void put_nommu_region(struct vm_region *region)
655{
656 down_write(&nommu_region_sem);
657 __put_nommu_region(region);
1da177e4
LT
658}
659
eb8cdec4
BS
660/*
661 * update protection on a vma
662 */
663static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
664{
665#ifdef CONFIG_MPU
666 struct mm_struct *mm = vma->vm_mm;
667 long start = vma->vm_start & PAGE_MASK;
668 while (start < vma->vm_end) {
669 protect_page(mm, start, flags);
670 start += PAGE_SIZE;
671 }
672 update_protections(mm);
673#endif
674}
675
3034097a 676/*
8feae131
DH
677 * add a VMA into a process's mm_struct in the appropriate place in the list
678 * and tree and add to the address space's page tree also if not an anonymous
679 * page
680 * - should be called with mm->mmap_sem held writelocked
3034097a 681 */
8feae131 682static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
1da177e4 683{
6038def0 684 struct vm_area_struct *pvma, *prev;
1da177e4 685 struct address_space *mapping;
6038def0 686 struct rb_node **p, *parent, *rb_prev;
8feae131 687
8feae131
DH
688 BUG_ON(!vma->vm_region);
689
690 mm->map_count++;
691 vma->vm_mm = mm;
1da177e4 692
eb8cdec4
BS
693 protect_vma(vma, vma->vm_flags);
694
1da177e4
LT
695 /* add the VMA to the mapping */
696 if (vma->vm_file) {
697 mapping = vma->vm_file->f_mapping;
698
83cde9e8 699 i_mmap_lock_write(mapping);
1da177e4 700 flush_dcache_mmap_lock(mapping);
6b2dbba8 701 vma_interval_tree_insert(vma, &mapping->i_mmap);
1da177e4 702 flush_dcache_mmap_unlock(mapping);
83cde9e8 703 i_mmap_unlock_write(mapping);
1da177e4
LT
704 }
705
8feae131 706 /* add the VMA to the tree */
6038def0 707 parent = rb_prev = NULL;
8feae131 708 p = &mm->mm_rb.rb_node;
1da177e4
LT
709 while (*p) {
710 parent = *p;
711 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
712
8feae131
DH
713 /* sort by: start addr, end addr, VMA struct addr in that order
714 * (the latter is necessary as we may get identical VMAs) */
715 if (vma->vm_start < pvma->vm_start)
1da177e4 716 p = &(*p)->rb_left;
6038def0
NK
717 else if (vma->vm_start > pvma->vm_start) {
718 rb_prev = parent;
1da177e4 719 p = &(*p)->rb_right;
6038def0 720 } else if (vma->vm_end < pvma->vm_end)
8feae131 721 p = &(*p)->rb_left;
6038def0
NK
722 else if (vma->vm_end > pvma->vm_end) {
723 rb_prev = parent;
8feae131 724 p = &(*p)->rb_right;
6038def0 725 } else if (vma < pvma)
8feae131 726 p = &(*p)->rb_left;
6038def0
NK
727 else if (vma > pvma) {
728 rb_prev = parent;
8feae131 729 p = &(*p)->rb_right;
6038def0 730 } else
8feae131 731 BUG();
1da177e4
LT
732 }
733
734 rb_link_node(&vma->vm_rb, parent, p);
8feae131
DH
735 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
736
737 /* add VMA to the VMA list also */
6038def0
NK
738 prev = NULL;
739 if (rb_prev)
740 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
8feae131 741
6038def0 742 __vma_link_list(mm, vma, prev, parent);
1da177e4
LT
743}
744
3034097a 745/*
8feae131 746 * delete a VMA from its owning mm_struct and address space
3034097a 747 */
8feae131 748static void delete_vma_from_mm(struct vm_area_struct *vma)
1da177e4 749{
615d6e87 750 int i;
1da177e4 751 struct address_space *mapping;
8feae131 752 struct mm_struct *mm = vma->vm_mm;
615d6e87 753 struct task_struct *curr = current;
8feae131 754
eb8cdec4
BS
755 protect_vma(vma, 0);
756
8feae131 757 mm->map_count--;
615d6e87
DB
758 for (i = 0; i < VMACACHE_SIZE; i++) {
759 /* if the vma is cached, invalidate the entire cache */
760 if (curr->vmacache[i] == vma) {
e020d5bd 761 vmacache_invalidate(mm);
615d6e87
DB
762 break;
763 }
764 }
1da177e4
LT
765
766 /* remove the VMA from the mapping */
767 if (vma->vm_file) {
768 mapping = vma->vm_file->f_mapping;
769
83cde9e8 770 i_mmap_lock_write(mapping);
1da177e4 771 flush_dcache_mmap_lock(mapping);
6b2dbba8 772 vma_interval_tree_remove(vma, &mapping->i_mmap);
1da177e4 773 flush_dcache_mmap_unlock(mapping);
83cde9e8 774 i_mmap_unlock_write(mapping);
1da177e4
LT
775 }
776
8feae131
DH
777 /* remove from the MM's tree and list */
778 rb_erase(&vma->vm_rb, &mm->mm_rb);
b951bf2c
NK
779
780 if (vma->vm_prev)
781 vma->vm_prev->vm_next = vma->vm_next;
782 else
783 mm->mmap = vma->vm_next;
784
785 if (vma->vm_next)
786 vma->vm_next->vm_prev = vma->vm_prev;
8feae131
DH
787}
788
789/*
790 * destroy a VMA record
791 */
792static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
793{
8feae131
DH
794 if (vma->vm_ops && vma->vm_ops->close)
795 vma->vm_ops->close(vma);
e9714acf 796 if (vma->vm_file)
8feae131 797 fput(vma->vm_file);
8feae131
DH
798 put_nommu_region(vma->vm_region);
799 kmem_cache_free(vm_area_cachep, vma);
800}
801
802/*
803 * look up the first VMA in which addr resides, NULL if none
804 * - should be called with mm->mmap_sem at least held readlocked
805 */
806struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
807{
808 struct vm_area_struct *vma;
8feae131
DH
809
810 /* check the cache first */
615d6e87
DB
811 vma = vmacache_find(mm, addr);
812 if (likely(vma))
8feae131
DH
813 return vma;
814
e922c4c5 815 /* trawl the list (there may be multiple mappings in which addr
8feae131 816 * resides) */
e922c4c5 817 for (vma = mm->mmap; vma; vma = vma->vm_next) {
8feae131
DH
818 if (vma->vm_start > addr)
819 return NULL;
820 if (vma->vm_end > addr) {
615d6e87 821 vmacache_update(addr, vma);
8feae131
DH
822 return vma;
823 }
824 }
825
826 return NULL;
827}
828EXPORT_SYMBOL(find_vma);
829
830/*
831 * find a VMA
832 * - we don't extend stack VMAs under NOMMU conditions
833 */
834struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
835{
7561e8ca 836 return find_vma(mm, addr);
8feae131
DH
837}
838
839/*
840 * expand a stack to a given address
841 * - not supported under NOMMU conditions
842 */
843int expand_stack(struct vm_area_struct *vma, unsigned long address)
844{
845 return -ENOMEM;
846}
847
848/*
849 * look up the first VMA exactly that exactly matches addr
850 * - should be called with mm->mmap_sem at least held readlocked
851 */
852static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
853 unsigned long addr,
854 unsigned long len)
855{
856 struct vm_area_struct *vma;
8feae131
DH
857 unsigned long end = addr + len;
858
859 /* check the cache first */
615d6e87
DB
860 vma = vmacache_find_exact(mm, addr, end);
861 if (vma)
8feae131
DH
862 return vma;
863
e922c4c5 864 /* trawl the list (there may be multiple mappings in which addr
8feae131 865 * resides) */
e922c4c5 866 for (vma = mm->mmap; vma; vma = vma->vm_next) {
8feae131
DH
867 if (vma->vm_start < addr)
868 continue;
869 if (vma->vm_start > addr)
870 return NULL;
871 if (vma->vm_end == end) {
615d6e87 872 vmacache_update(addr, vma);
8feae131
DH
873 return vma;
874 }
875 }
876
877 return NULL;
1da177e4
LT
878}
879
880/*
881 * determine whether a mapping should be permitted and, if so, what sort of
882 * mapping we're capable of supporting
883 */
884static int validate_mmap_request(struct file *file,
885 unsigned long addr,
886 unsigned long len,
887 unsigned long prot,
888 unsigned long flags,
889 unsigned long pgoff,
890 unsigned long *_capabilities)
891{
8feae131 892 unsigned long capabilities, rlen;
1da177e4
LT
893 int ret;
894
895 /* do the simple checks first */
22cc877b 896 if (flags & MAP_FIXED)
1da177e4 897 return -EINVAL;
1da177e4
LT
898
899 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
900 (flags & MAP_TYPE) != MAP_SHARED)
901 return -EINVAL;
902
f81cff0d 903 if (!len)
1da177e4
LT
904 return -EINVAL;
905
f81cff0d 906 /* Careful about overflows.. */
8feae131
DH
907 rlen = PAGE_ALIGN(len);
908 if (!rlen || rlen > TASK_SIZE)
f81cff0d
MF
909 return -ENOMEM;
910
1da177e4 911 /* offset overflow? */
8feae131 912 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
f81cff0d 913 return -EOVERFLOW;
1da177e4
LT
914
915 if (file) {
1da177e4 916 /* files must support mmap */
72c2d531 917 if (!file->f_op->mmap)
1da177e4
LT
918 return -ENODEV;
919
920 /* work out if what we've got could possibly be shared
921 * - we support chardevs that provide their own "memory"
922 * - we support files/blockdevs that are memory backed
923 */
b4caecd4
CH
924 if (file->f_op->mmap_capabilities) {
925 capabilities = file->f_op->mmap_capabilities(file);
926 } else {
1da177e4
LT
927 /* no explicit capabilities set, so assume some
928 * defaults */
496ad9aa 929 switch (file_inode(file)->i_mode & S_IFMT) {
1da177e4
LT
930 case S_IFREG:
931 case S_IFBLK:
b4caecd4 932 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
933 break;
934
935 case S_IFCHR:
936 capabilities =
b4caecd4
CH
937 NOMMU_MAP_DIRECT |
938 NOMMU_MAP_READ |
939 NOMMU_MAP_WRITE;
1da177e4
LT
940 break;
941
942 default:
943 return -EINVAL;
944 }
945 }
946
947 /* eliminate any capabilities that we can't support on this
948 * device */
949 if (!file->f_op->get_unmapped_area)
b4caecd4 950 capabilities &= ~NOMMU_MAP_DIRECT;
6e242a1c 951 if (!(file->f_mode & FMODE_CAN_READ))
b4caecd4 952 capabilities &= ~NOMMU_MAP_COPY;
1da177e4 953
28d7a6ae
GY
954 /* The file shall have been opened with read permission. */
955 if (!(file->f_mode & FMODE_READ))
956 return -EACCES;
957
1da177e4
LT
958 if (flags & MAP_SHARED) {
959 /* do checks for writing, appending and locking */
960 if ((prot & PROT_WRITE) &&
961 !(file->f_mode & FMODE_WRITE))
962 return -EACCES;
963
496ad9aa 964 if (IS_APPEND(file_inode(file)) &&
1da177e4
LT
965 (file->f_mode & FMODE_WRITE))
966 return -EACCES;
967
d7a06983 968 if (locks_verify_locked(file))
1da177e4
LT
969 return -EAGAIN;
970
b4caecd4 971 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
972 return -ENODEV;
973
1da177e4 974 /* we mustn't privatise shared mappings */
b4caecd4 975 capabilities &= ~NOMMU_MAP_COPY;
ac714904 976 } else {
1da177e4
LT
977 /* we're going to read the file into private memory we
978 * allocate */
b4caecd4 979 if (!(capabilities & NOMMU_MAP_COPY))
1da177e4
LT
980 return -ENODEV;
981
982 /* we don't permit a private writable mapping to be
983 * shared with the backing device */
984 if (prot & PROT_WRITE)
b4caecd4 985 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4
LT
986 }
987
b4caecd4
CH
988 if (capabilities & NOMMU_MAP_DIRECT) {
989 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
990 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
991 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
3c7b2045 992 ) {
b4caecd4 993 capabilities &= ~NOMMU_MAP_DIRECT;
3c7b2045 994 if (flags & MAP_SHARED) {
22cc877b 995 pr_warn("MAP_SHARED not completely supported on !MMU\n");
3c7b2045
BS
996 return -EINVAL;
997 }
998 }
999 }
1000
1da177e4
LT
1001 /* handle executable mappings and implied executable
1002 * mappings */
90f8572b 1003 if (path_noexec(&file->f_path)) {
1da177e4
LT
1004 if (prot & PROT_EXEC)
1005 return -EPERM;
ac714904 1006 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1da177e4
LT
1007 /* handle implication of PROT_EXEC by PROT_READ */
1008 if (current->personality & READ_IMPLIES_EXEC) {
b4caecd4 1009 if (capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
1010 prot |= PROT_EXEC;
1011 }
ac714904 1012 } else if ((prot & PROT_READ) &&
1da177e4 1013 (prot & PROT_EXEC) &&
b4caecd4 1014 !(capabilities & NOMMU_MAP_EXEC)
1da177e4
LT
1015 ) {
1016 /* backing file is not executable, try to copy */
b4caecd4 1017 capabilities &= ~NOMMU_MAP_DIRECT;
1da177e4 1018 }
ac714904 1019 } else {
1da177e4
LT
1020 /* anonymous mappings are always memory backed and can be
1021 * privately mapped
1022 */
b4caecd4 1023 capabilities = NOMMU_MAP_COPY;
1da177e4
LT
1024
1025 /* handle PROT_EXEC implication by PROT_READ */
1026 if ((prot & PROT_READ) &&
1027 (current->personality & READ_IMPLIES_EXEC))
1028 prot |= PROT_EXEC;
1029 }
1030
1031 /* allow the security API to have its say */
e5467859 1032 ret = security_mmap_addr(addr);
1da177e4
LT
1033 if (ret < 0)
1034 return ret;
1035
1036 /* looks okay */
1037 *_capabilities = capabilities;
1038 return 0;
1039}
1040
1041/*
1042 * we've determined that we can make the mapping, now translate what we
1043 * now know into VMA flags
1044 */
1045static unsigned long determine_vm_flags(struct file *file,
1046 unsigned long prot,
1047 unsigned long flags,
1048 unsigned long capabilities)
1049{
1050 unsigned long vm_flags;
1051
e6bfb709 1052 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags);
1da177e4
LT
1053 /* vm_flags |= mm->def_flags; */
1054
b4caecd4 1055 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1da177e4 1056 /* attempt to share read-only copies of mapped file chunks */
3c7b2045 1057 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1da177e4
LT
1058 if (file && !(prot & PROT_WRITE))
1059 vm_flags |= VM_MAYSHARE;
3c7b2045 1060 } else {
1da177e4
LT
1061 /* overlay a shareable mapping on the backing device or inode
1062 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1063 * romfs/cramfs */
b4caecd4 1064 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1da177e4 1065 if (flags & MAP_SHARED)
3c7b2045 1066 vm_flags |= VM_SHARED;
1da177e4
LT
1067 }
1068
1069 /* refuse to let anyone share private mappings with this process if
1070 * it's being traced - otherwise breakpoints set in it may interfere
1071 * with another untraced process
1072 */
a288eecc 1073 if ((flags & MAP_PRIVATE) && current->ptrace)
1da177e4
LT
1074 vm_flags &= ~VM_MAYSHARE;
1075
1076 return vm_flags;
1077}
1078
1079/*
8feae131
DH
1080 * set up a shared mapping on a file (the driver or filesystem provides and
1081 * pins the storage)
1da177e4 1082 */
8feae131 1083static int do_mmap_shared_file(struct vm_area_struct *vma)
1da177e4
LT
1084{
1085 int ret;
1086
1087 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
dd8632a1
PM
1088 if (ret == 0) {
1089 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 1090 return 0;
dd8632a1 1091 }
1da177e4
LT
1092 if (ret != -ENOSYS)
1093 return ret;
1094
3fa30460
DH
1095 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1096 * opposed to tried but failed) so we can only give a suitable error as
1097 * it's not possible to make a private copy if MAP_SHARED was given */
1da177e4
LT
1098 return -ENODEV;
1099}
1100
1101/*
1102 * set up a private mapping or an anonymous shared mapping
1103 */
8feae131
DH
1104static int do_mmap_private(struct vm_area_struct *vma,
1105 struct vm_region *region,
645d83c5
DH
1106 unsigned long len,
1107 unsigned long capabilities)
1da177e4 1108{
dbc8358c 1109 unsigned long total, point;
1da177e4 1110 void *base;
8feae131 1111 int ret, order;
1da177e4
LT
1112
1113 /* invoke the file's mapping function so that it can keep track of
1114 * shared mappings on devices or memory
1115 * - VM_MAYSHARE will be set if it may attempt to share
1116 */
b4caecd4 1117 if (capabilities & NOMMU_MAP_DIRECT) {
1da177e4 1118 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
dd8632a1 1119 if (ret == 0) {
1da177e4 1120 /* shouldn't return success if we're not sharing */
dd8632a1
PM
1121 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1122 vma->vm_region->vm_top = vma->vm_region->vm_end;
645d83c5 1123 return 0;
1da177e4 1124 }
dd8632a1
PM
1125 if (ret != -ENOSYS)
1126 return ret;
1da177e4
LT
1127
1128 /* getting an ENOSYS error indicates that direct mmap isn't
1129 * possible (as opposed to tried but failed) so we'll try to
1130 * make a private copy of the data and map that instead */
1131 }
1132
8feae131 1133
1da177e4
LT
1134 /* allocate some memory to hold the mapping
1135 * - note that this may not return a page-aligned address if the object
1136 * we're allocating is smaller than a page
1137 */
f67d9b15 1138 order = get_order(len);
8feae131 1139 total = 1 << order;
f67d9b15 1140 point = len >> PAGE_SHIFT;
dd8632a1 1141
dbc8358c 1142 /* we don't want to allocate a power-of-2 sized page set */
22cc877b 1143 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
dbc8358c 1144 total = point;
8feae131 1145
da616534 1146 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
dbc8358c
JK
1147 if (!base)
1148 goto enomem;
1149
1150 atomic_long_add(total, &mmap_pages_allocated);
1da177e4 1151
8feae131
DH
1152 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1153 region->vm_start = (unsigned long) base;
f67d9b15 1154 region->vm_end = region->vm_start + len;
dd8632a1 1155 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
8feae131
DH
1156
1157 vma->vm_start = region->vm_start;
1158 vma->vm_end = region->vm_start + len;
1da177e4
LT
1159
1160 if (vma->vm_file) {
1161 /* read the contents of a file into the copy */
1162 mm_segment_t old_fs;
1163 loff_t fpos;
1164
1165 fpos = vma->vm_pgoff;
1166 fpos <<= PAGE_SHIFT;
1167
1168 old_fs = get_fs();
1169 set_fs(KERNEL_DS);
6e242a1c 1170 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1da177e4
LT
1171 set_fs(old_fs);
1172
1173 if (ret < 0)
1174 goto error_free;
1175
1176 /* clear the last little bit */
f67d9b15
BL
1177 if (ret < len)
1178 memset(base + ret, 0, len - ret);
1da177e4 1179
1da177e4
LT
1180 }
1181
1182 return 0;
1183
1184error_free:
7223bb4a 1185 free_page_series(region->vm_start, region->vm_top);
8feae131
DH
1186 region->vm_start = vma->vm_start = 0;
1187 region->vm_end = vma->vm_end = 0;
dd8632a1 1188 region->vm_top = 0;
1da177e4
LT
1189 return ret;
1190
1191enomem:
b1de0d13 1192 pr_err("Allocation of length %lu from process %d (%s) failed\n",
05ae6fa3 1193 len, current->pid, current->comm);
7bf02ea2 1194 show_free_areas(0);
1da177e4
LT
1195 return -ENOMEM;
1196}
1197
1198/*
1199 * handle mapping creation for uClinux
1200 */
1fcfd8db
ON
1201unsigned long do_mmap(struct file *file,
1202 unsigned long addr,
1203 unsigned long len,
1204 unsigned long prot,
1205 unsigned long flags,
1206 vm_flags_t vm_flags,
1207 unsigned long pgoff,
1208 unsigned long *populate)
1da177e4 1209{
8feae131
DH
1210 struct vm_area_struct *vma;
1211 struct vm_region *region;
1da177e4 1212 struct rb_node *rb;
1fcfd8db 1213 unsigned long capabilities, result;
1da177e4
LT
1214 int ret;
1215
41badc15 1216 *populate = 0;
bebeb3d6 1217
1da177e4
LT
1218 /* decide whether we should attempt the mapping, and if so what sort of
1219 * mapping */
1220 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1221 &capabilities);
22cc877b 1222 if (ret < 0)
1da177e4
LT
1223 return ret;
1224
06aab5a3
DH
1225 /* we ignore the address hint */
1226 addr = 0;
f67d9b15 1227 len = PAGE_ALIGN(len);
06aab5a3 1228
1da177e4
LT
1229 /* we've determined that we can make the mapping, now translate what we
1230 * now know into VMA flags */
1fcfd8db 1231 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1da177e4 1232
8feae131
DH
1233 /* we're going to need to record the mapping */
1234 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1235 if (!region)
1236 goto error_getting_region;
1237
1238 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1239 if (!vma)
1240 goto error_getting_vma;
1da177e4 1241
1e2ae599 1242 region->vm_usage = 1;
8feae131
DH
1243 region->vm_flags = vm_flags;
1244 region->vm_pgoff = pgoff;
1245
5beb4930 1246 INIT_LIST_HEAD(&vma->anon_vma_chain);
8feae131
DH
1247 vma->vm_flags = vm_flags;
1248 vma->vm_pgoff = pgoff;
1da177e4 1249
8feae131 1250 if (file) {
cb0942b8
AV
1251 region->vm_file = get_file(file);
1252 vma->vm_file = get_file(file);
8feae131
DH
1253 }
1254
1255 down_write(&nommu_region_sem);
1256
1257 /* if we want to share, we need to check for regions created by other
1da177e4 1258 * mmap() calls that overlap with our proposed mapping
8feae131 1259 * - we can only share with a superset match on most regular files
1da177e4
LT
1260 * - shared mappings on character devices and memory backed files are
1261 * permitted to overlap inexactly as far as we are concerned for in
1262 * these cases, sharing is handled in the driver or filesystem rather
1263 * than here
1264 */
1265 if (vm_flags & VM_MAYSHARE) {
8feae131
DH
1266 struct vm_region *pregion;
1267 unsigned long pglen, rpglen, pgend, rpgend, start;
1da177e4 1268
8feae131
DH
1269 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1270 pgend = pgoff + pglen;
165b2392 1271
8feae131
DH
1272 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1273 pregion = rb_entry(rb, struct vm_region, vm_rb);
1da177e4 1274
8feae131 1275 if (!(pregion->vm_flags & VM_MAYSHARE))
1da177e4
LT
1276 continue;
1277
1278 /* search for overlapping mappings on the same file */
496ad9aa
AV
1279 if (file_inode(pregion->vm_file) !=
1280 file_inode(file))
1da177e4
LT
1281 continue;
1282
8feae131 1283 if (pregion->vm_pgoff >= pgend)
1da177e4
LT
1284 continue;
1285
8feae131
DH
1286 rpglen = pregion->vm_end - pregion->vm_start;
1287 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1288 rpgend = pregion->vm_pgoff + rpglen;
1289 if (pgoff >= rpgend)
1da177e4
LT
1290 continue;
1291
8feae131
DH
1292 /* handle inexactly overlapping matches between
1293 * mappings */
1294 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1295 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1296 /* new mapping is not a subset of the region */
b4caecd4 1297 if (!(capabilities & NOMMU_MAP_DIRECT))
1da177e4
LT
1298 goto sharing_violation;
1299 continue;
1300 }
1301
8feae131 1302 /* we've found a region we can share */
1e2ae599 1303 pregion->vm_usage++;
8feae131
DH
1304 vma->vm_region = pregion;
1305 start = pregion->vm_start;
1306 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1307 vma->vm_start = start;
1308 vma->vm_end = start + len;
1309
22cc877b 1310 if (pregion->vm_flags & VM_MAPPED_COPY)
8feae131 1311 vma->vm_flags |= VM_MAPPED_COPY;
22cc877b 1312 else {
8feae131
DH
1313 ret = do_mmap_shared_file(vma);
1314 if (ret < 0) {
1315 vma->vm_region = NULL;
1316 vma->vm_start = 0;
1317 vma->vm_end = 0;
1e2ae599 1318 pregion->vm_usage--;
8feae131
DH
1319 pregion = NULL;
1320 goto error_just_free;
1321 }
1322 }
1323 fput(region->vm_file);
1324 kmem_cache_free(vm_region_jar, region);
1325 region = pregion;
1326 result = start;
1327 goto share;
1da177e4
LT
1328 }
1329
1da177e4
LT
1330 /* obtain the address at which to make a shared mapping
1331 * - this is the hook for quasi-memory character devices to
1332 * tell us the location of a shared mapping
1333 */
b4caecd4 1334 if (capabilities & NOMMU_MAP_DIRECT) {
1da177e4
LT
1335 addr = file->f_op->get_unmapped_area(file, addr, len,
1336 pgoff, flags);
bb005a59 1337 if (IS_ERR_VALUE(addr)) {
1da177e4 1338 ret = addr;
bb005a59 1339 if (ret != -ENOSYS)
8feae131 1340 goto error_just_free;
1da177e4
LT
1341
1342 /* the driver refused to tell us where to site
1343 * the mapping so we'll have to attempt to copy
1344 * it */
bb005a59 1345 ret = -ENODEV;
b4caecd4 1346 if (!(capabilities & NOMMU_MAP_COPY))
8feae131 1347 goto error_just_free;
1da177e4 1348
b4caecd4 1349 capabilities &= ~NOMMU_MAP_DIRECT;
8feae131
DH
1350 } else {
1351 vma->vm_start = region->vm_start = addr;
1352 vma->vm_end = region->vm_end = addr + len;
1da177e4
LT
1353 }
1354 }
1355 }
1356
8feae131 1357 vma->vm_region = region;
1da177e4 1358
645d83c5 1359 /* set up the mapping
b4caecd4 1360 * - the region is filled in if NOMMU_MAP_DIRECT is still set
645d83c5 1361 */
1da177e4 1362 if (file && vma->vm_flags & VM_SHARED)
8feae131 1363 ret = do_mmap_shared_file(vma);
1da177e4 1364 else
645d83c5 1365 ret = do_mmap_private(vma, region, len, capabilities);
1da177e4 1366 if (ret < 0)
645d83c5
DH
1367 goto error_just_free;
1368 add_nommu_region(region);
8feae131 1369
ea637639
JZ
1370 /* clear anonymous mappings that don't ask for uninitialized data */
1371 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1372 memset((void *)region->vm_start, 0,
1373 region->vm_end - region->vm_start);
1374
1da177e4 1375 /* okay... we have a mapping; now we have to register it */
8feae131 1376 result = vma->vm_start;
1da177e4 1377
1da177e4
LT
1378 current->mm->total_vm += len >> PAGE_SHIFT;
1379
8feae131
DH
1380share:
1381 add_vma_to_mm(current->mm, vma);
1da177e4 1382
cfe79c00
MF
1383 /* we flush the region from the icache only when the first executable
1384 * mapping of it is made */
1385 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1386 flush_icache_range(region->vm_start, region->vm_end);
1387 region->vm_icache_flushed = true;
1388 }
1da177e4 1389
cfe79c00 1390 up_write(&nommu_region_sem);
1da177e4 1391
8feae131 1392 return result;
1da177e4 1393
8feae131
DH
1394error_just_free:
1395 up_write(&nommu_region_sem);
1396error:
89a86402
DH
1397 if (region->vm_file)
1398 fput(region->vm_file);
8feae131 1399 kmem_cache_free(vm_region_jar, region);
89a86402
DH
1400 if (vma->vm_file)
1401 fput(vma->vm_file);
8feae131 1402 kmem_cache_free(vm_area_cachep, vma);
8feae131
DH
1403 return ret;
1404
1405sharing_violation:
1406 up_write(&nommu_region_sem);
22cc877b 1407 pr_warn("Attempt to share mismatched mappings\n");
8feae131
DH
1408 ret = -EINVAL;
1409 goto error;
1da177e4 1410
8feae131
DH
1411error_getting_vma:
1412 kmem_cache_free(vm_region_jar, region);
22cc877b
LR
1413 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1414 len, current->pid);
7bf02ea2 1415 show_free_areas(0);
1da177e4
LT
1416 return -ENOMEM;
1417
8feae131 1418error_getting_region:
22cc877b
LR
1419 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1420 len, current->pid);
7bf02ea2 1421 show_free_areas(0);
1da177e4
LT
1422 return -ENOMEM;
1423}
6be5ceb0 1424
66f0dc48
HD
1425SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1426 unsigned long, prot, unsigned long, flags,
1427 unsigned long, fd, unsigned long, pgoff)
1428{
1429 struct file *file = NULL;
1430 unsigned long retval = -EBADF;
1431
120a795d 1432 audit_mmap_fd(fd, flags);
66f0dc48
HD
1433 if (!(flags & MAP_ANONYMOUS)) {
1434 file = fget(fd);
1435 if (!file)
1436 goto out;
1437 }
1438
1439 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1440
ad1ed293 1441 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
66f0dc48
HD
1442
1443 if (file)
1444 fput(file);
1445out:
1446 return retval;
1447}
1448
a4679373
CH
1449#ifdef __ARCH_WANT_SYS_OLD_MMAP
1450struct mmap_arg_struct {
1451 unsigned long addr;
1452 unsigned long len;
1453 unsigned long prot;
1454 unsigned long flags;
1455 unsigned long fd;
1456 unsigned long offset;
1457};
1458
1459SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1460{
1461 struct mmap_arg_struct a;
1462
1463 if (copy_from_user(&a, arg, sizeof(a)))
1464 return -EFAULT;
1824cb75 1465 if (offset_in_page(a.offset))
a4679373
CH
1466 return -EINVAL;
1467
1468 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1469 a.offset >> PAGE_SHIFT);
1470}
1471#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1472
1da177e4 1473/*
8feae131
DH
1474 * split a vma into two pieces at address 'addr', a new vma is allocated either
1475 * for the first part or the tail.
1da177e4 1476 */
8feae131
DH
1477int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1478 unsigned long addr, int new_below)
1da177e4 1479{
8feae131
DH
1480 struct vm_area_struct *new;
1481 struct vm_region *region;
1482 unsigned long npages;
1da177e4 1483
779c1023
DH
1484 /* we're only permitted to split anonymous regions (these should have
1485 * only a single usage on the region) */
1486 if (vma->vm_file)
8feae131 1487 return -ENOMEM;
1da177e4 1488
8feae131
DH
1489 if (mm->map_count >= sysctl_max_map_count)
1490 return -ENOMEM;
1da177e4 1491
8feae131
DH
1492 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1493 if (!region)
1494 return -ENOMEM;
1da177e4 1495
8feae131
DH
1496 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1497 if (!new) {
1498 kmem_cache_free(vm_region_jar, region);
1499 return -ENOMEM;
1500 }
1501
1502 /* most fields are the same, copy all, and then fixup */
1503 *new = *vma;
1504 *region = *vma->vm_region;
1505 new->vm_region = region;
1506
1507 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1508
1509 if (new_below) {
dd8632a1 1510 region->vm_top = region->vm_end = new->vm_end = addr;
8feae131
DH
1511 } else {
1512 region->vm_start = new->vm_start = addr;
1513 region->vm_pgoff = new->vm_pgoff += npages;
1da177e4 1514 }
8feae131
DH
1515
1516 if (new->vm_ops && new->vm_ops->open)
1517 new->vm_ops->open(new);
1518
1519 delete_vma_from_mm(vma);
1520 down_write(&nommu_region_sem);
1521 delete_nommu_region(vma->vm_region);
1522 if (new_below) {
1523 vma->vm_region->vm_start = vma->vm_start = addr;
1524 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1525 } else {
1526 vma->vm_region->vm_end = vma->vm_end = addr;
dd8632a1 1527 vma->vm_region->vm_top = addr;
8feae131
DH
1528 }
1529 add_nommu_region(vma->vm_region);
1530 add_nommu_region(new->vm_region);
1531 up_write(&nommu_region_sem);
1532 add_vma_to_mm(mm, vma);
1533 add_vma_to_mm(mm, new);
1534 return 0;
1da177e4
LT
1535}
1536
3034097a 1537/*
8feae131
DH
1538 * shrink a VMA by removing the specified chunk from either the beginning or
1539 * the end
3034097a 1540 */
8feae131
DH
1541static int shrink_vma(struct mm_struct *mm,
1542 struct vm_area_struct *vma,
1543 unsigned long from, unsigned long to)
1da177e4 1544{
8feae131 1545 struct vm_region *region;
1da177e4 1546
8feae131
DH
1547 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1548 * and list */
1549 delete_vma_from_mm(vma);
1550 if (from > vma->vm_start)
1551 vma->vm_end = from;
1552 else
1553 vma->vm_start = to;
1554 add_vma_to_mm(mm, vma);
1da177e4 1555
8feae131
DH
1556 /* cut the backing region down to size */
1557 region = vma->vm_region;
1e2ae599 1558 BUG_ON(region->vm_usage != 1);
8feae131
DH
1559
1560 down_write(&nommu_region_sem);
1561 delete_nommu_region(region);
dd8632a1
PM
1562 if (from > region->vm_start) {
1563 to = region->vm_top;
1564 region->vm_top = region->vm_end = from;
1565 } else {
8feae131 1566 region->vm_start = to;
dd8632a1 1567 }
8feae131
DH
1568 add_nommu_region(region);
1569 up_write(&nommu_region_sem);
1570
1571 free_page_series(from, to);
1572 return 0;
1573}
1da177e4 1574
8feae131
DH
1575/*
1576 * release a mapping
1577 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1578 * VMA, though it need not cover the whole VMA
1579 */
1580int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1581{
1582 struct vm_area_struct *vma;
f67d9b15 1583 unsigned long end;
8feae131 1584 int ret;
1da177e4 1585
f67d9b15 1586 len = PAGE_ALIGN(len);
8feae131
DH
1587 if (len == 0)
1588 return -EINVAL;
365e9c87 1589
f67d9b15
BL
1590 end = start + len;
1591
8feae131
DH
1592 /* find the first potentially overlapping VMA */
1593 vma = find_vma(mm, start);
1594 if (!vma) {
ac714904 1595 static int limit;
33e5d769 1596 if (limit < 5) {
22cc877b
LR
1597 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1598 current->pid, current->comm,
1599 start, start + len - 1);
33e5d769
DH
1600 limit++;
1601 }
8feae131
DH
1602 return -EINVAL;
1603 }
1da177e4 1604
8feae131
DH
1605 /* we're allowed to split an anonymous VMA but not a file-backed one */
1606 if (vma->vm_file) {
1607 do {
22cc877b 1608 if (start > vma->vm_start)
8feae131 1609 return -EINVAL;
8feae131
DH
1610 if (end == vma->vm_end)
1611 goto erase_whole_vma;
d75a310c
NK
1612 vma = vma->vm_next;
1613 } while (vma);
8feae131
DH
1614 return -EINVAL;
1615 } else {
1616 /* the chunk must be a subset of the VMA found */
1617 if (start == vma->vm_start && end == vma->vm_end)
1618 goto erase_whole_vma;
22cc877b 1619 if (start < vma->vm_start || end > vma->vm_end)
8feae131 1620 return -EINVAL;
1824cb75 1621 if (offset_in_page(start))
8feae131 1622 return -EINVAL;
1824cb75 1623 if (end != vma->vm_end && offset_in_page(end))
8feae131 1624 return -EINVAL;
8feae131
DH
1625 if (start != vma->vm_start && end != vma->vm_end) {
1626 ret = split_vma(mm, vma, start, 1);
22cc877b 1627 if (ret < 0)
8feae131 1628 return ret;
8feae131
DH
1629 }
1630 return shrink_vma(mm, vma, start, end);
1631 }
1da177e4 1632
8feae131
DH
1633erase_whole_vma:
1634 delete_vma_from_mm(vma);
1635 delete_vma(mm, vma);
1da177e4
LT
1636 return 0;
1637}
b5073173 1638EXPORT_SYMBOL(do_munmap);
1da177e4 1639
bfce281c 1640int vm_munmap(unsigned long addr, size_t len)
3034097a 1641{
bfce281c 1642 struct mm_struct *mm = current->mm;
3034097a 1643 int ret;
3034097a
DH
1644
1645 down_write(&mm->mmap_sem);
1646 ret = do_munmap(mm, addr, len);
1647 up_write(&mm->mmap_sem);
1648 return ret;
1649}
a46ef99d
LT
1650EXPORT_SYMBOL(vm_munmap);
1651
1652SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1653{
bfce281c 1654 return vm_munmap(addr, len);
a46ef99d 1655}
3034097a
DH
1656
1657/*
8feae131 1658 * release all the mappings made in a process's VM space
3034097a 1659 */
8feae131 1660void exit_mmap(struct mm_struct *mm)
1da177e4 1661{
8feae131 1662 struct vm_area_struct *vma;
1da177e4 1663
8feae131
DH
1664 if (!mm)
1665 return;
1da177e4 1666
8feae131 1667 mm->total_vm = 0;
1da177e4 1668
8feae131
DH
1669 while ((vma = mm->mmap)) {
1670 mm->mmap = vma->vm_next;
1671 delete_vma_from_mm(vma);
1672 delete_vma(mm, vma);
04c34961 1673 cond_resched();
1da177e4
LT
1674 }
1675}
1676
5d22fc25 1677int vm_brk(unsigned long addr, unsigned long len)
1da177e4
LT
1678{
1679 return -ENOMEM;
1680}
1681
1682/*
6fa5f80b
DH
1683 * expand (or shrink) an existing mapping, potentially moving it at the same
1684 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1da177e4 1685 *
6fa5f80b 1686 * under NOMMU conditions, we only permit changing a mapping's size, and only
8feae131
DH
1687 * as long as it stays within the region allocated by do_mmap_private() and the
1688 * block is not shareable
1da177e4 1689 *
6fa5f80b 1690 * MREMAP_FIXED is not supported under NOMMU conditions
1da177e4 1691 */
4b377bab 1692static unsigned long do_mremap(unsigned long addr,
1da177e4
LT
1693 unsigned long old_len, unsigned long new_len,
1694 unsigned long flags, unsigned long new_addr)
1695{
6fa5f80b 1696 struct vm_area_struct *vma;
1da177e4
LT
1697
1698 /* insanity checks first */
f67d9b15
BL
1699 old_len = PAGE_ALIGN(old_len);
1700 new_len = PAGE_ALIGN(new_len);
8feae131 1701 if (old_len == 0 || new_len == 0)
1da177e4
LT
1702 return (unsigned long) -EINVAL;
1703
1824cb75 1704 if (offset_in_page(addr))
8feae131
DH
1705 return -EINVAL;
1706
1da177e4
LT
1707 if (flags & MREMAP_FIXED && new_addr != addr)
1708 return (unsigned long) -EINVAL;
1709
8feae131 1710 vma = find_vma_exact(current->mm, addr, old_len);
6fa5f80b
DH
1711 if (!vma)
1712 return (unsigned long) -EINVAL;
1da177e4 1713
6fa5f80b 1714 if (vma->vm_end != vma->vm_start + old_len)
1da177e4
LT
1715 return (unsigned long) -EFAULT;
1716
6fa5f80b 1717 if (vma->vm_flags & VM_MAYSHARE)
1da177e4
LT
1718 return (unsigned long) -EPERM;
1719
8feae131 1720 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1da177e4
LT
1721 return (unsigned long) -ENOMEM;
1722
1723 /* all checks complete - do it */
6fa5f80b 1724 vma->vm_end = vma->vm_start + new_len;
6fa5f80b
DH
1725 return vma->vm_start;
1726}
1727
6a6160a7
HC
1728SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1729 unsigned long, new_len, unsigned long, flags,
1730 unsigned long, new_addr)
6fa5f80b
DH
1731{
1732 unsigned long ret;
1733
1734 down_write(&current->mm->mmap_sem);
1735 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1736 up_write(&current->mm->mmap_sem);
1737 return ret;
1da177e4
LT
1738}
1739
240aadee
ML
1740struct page *follow_page_mask(struct vm_area_struct *vma,
1741 unsigned long address, unsigned int flags,
1742 unsigned int *page_mask)
1da177e4 1743{
240aadee 1744 *page_mask = 0;
1da177e4
LT
1745 return NULL;
1746}
1747
8f3b1327
BL
1748int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1749 unsigned long pfn, unsigned long size, pgprot_t prot)
1da177e4 1750{
8f3b1327
BL
1751 if (addr != (pfn << PAGE_SHIFT))
1752 return -EINVAL;
1753
314e51b9 1754 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
66aa2b4b 1755 return 0;
1da177e4 1756}
22c4af40 1757EXPORT_SYMBOL(remap_pfn_range);
1da177e4 1758
3c0b9de6
LT
1759int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1760{
1761 unsigned long pfn = start >> PAGE_SHIFT;
1762 unsigned long vm_len = vma->vm_end - vma->vm_start;
1763
1764 pfn += vma->vm_pgoff;
1765 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1766}
1767EXPORT_SYMBOL(vm_iomap_memory);
1768
f905bc44
PM
1769int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1770 unsigned long pgoff)
1771{
1772 unsigned int size = vma->vm_end - vma->vm_start;
1773
1774 if (!(vma->vm_flags & VM_USERMAP))
1775 return -EINVAL;
1776
1777 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1778 vma->vm_end = vma->vm_start + size;
1779
1780 return 0;
1781}
1782EXPORT_SYMBOL(remap_vmalloc_range);
1783
1da177e4
LT
1784unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1785 unsigned long len, unsigned long pgoff, unsigned long flags)
1786{
1787 return -ENOMEM;
1788}
1789
1da177e4
LT
1790void unmap_mapping_range(struct address_space *mapping,
1791 loff_t const holebegin, loff_t const holelen,
1792 int even_cows)
1793{
1794}
22c4af40 1795EXPORT_SYMBOL(unmap_mapping_range);
1da177e4 1796
d0217ac0 1797int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
b0e15190
DH
1798{
1799 BUG();
d0217ac0 1800 return 0;
b0e15190 1801}
b5073173 1802EXPORT_SYMBOL(filemap_fault);
0ec76a11 1803
82b0f8c3 1804void filemap_map_pages(struct vm_fault *vmf,
bae473a4 1805 pgoff_t start_pgoff, pgoff_t end_pgoff)
f1820361
KS
1806{
1807 BUG();
1808}
1809EXPORT_SYMBOL(filemap_map_pages);
1810
84d77d3f 1811int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
442486ec 1812 unsigned long addr, void *buf, int len, unsigned int gup_flags)
0ec76a11 1813{
0ec76a11 1814 struct vm_area_struct *vma;
442486ec 1815 int write = gup_flags & FOLL_WRITE;
0ec76a11
DH
1816
1817 down_read(&mm->mmap_sem);
1818
1819 /* the access must start within one of the target process's mappings */
0159b141
DH
1820 vma = find_vma(mm, addr);
1821 if (vma) {
0ec76a11
DH
1822 /* don't overrun this mapping */
1823 if (addr + len >= vma->vm_end)
1824 len = vma->vm_end - addr;
1825
1826 /* only read or write mappings where it is permitted */
d00c7b99 1827 if (write && vma->vm_flags & VM_MAYWRITE)
7959722b
JZ
1828 copy_to_user_page(vma, NULL, addr,
1829 (void *) addr, buf, len);
d00c7b99 1830 else if (!write && vma->vm_flags & VM_MAYREAD)
7959722b
JZ
1831 copy_from_user_page(vma, NULL, addr,
1832 buf, (void *) addr, len);
0ec76a11
DH
1833 else
1834 len = 0;
1835 } else {
1836 len = 0;
1837 }
1838
1839 up_read(&mm->mmap_sem);
f55f199b
MF
1840
1841 return len;
1842}
1843
1844/**
1845 * @access_remote_vm - access another process' address space
1846 * @mm: the mm_struct of the target address space
1847 * @addr: start address to access
1848 * @buf: source or destination buffer
1849 * @len: number of bytes to transfer
6347e8d5 1850 * @gup_flags: flags modifying lookup behaviour
f55f199b
MF
1851 *
1852 * The caller must hold a reference on @mm.
1853 */
1854int access_remote_vm(struct mm_struct *mm, unsigned long addr,
6347e8d5 1855 void *buf, int len, unsigned int gup_flags)
f55f199b 1856{
6347e8d5 1857 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
f55f199b
MF
1858}
1859
1860/*
1861 * Access another process' address space.
1862 * - source/target buffer must be kernel space
1863 */
f307ab6d
LS
1864int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
1865 unsigned int gup_flags)
f55f199b
MF
1866{
1867 struct mm_struct *mm;
1868
1869 if (addr + len < addr)
1870 return 0;
1871
1872 mm = get_task_mm(tsk);
1873 if (!mm)
1874 return 0;
1875
f307ab6d 1876 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
f55f199b 1877
0ec76a11
DH
1878 mmput(mm);
1879 return len;
1880}
fcd35857 1881EXPORT_SYMBOL_GPL(access_process_vm);
7e660872
DH
1882
1883/**
1884 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1885 * @inode: The inode to check
1886 * @size: The current filesize of the inode
1887 * @newsize: The proposed filesize of the inode
1888 *
1889 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1890 * make sure that that any outstanding VMAs aren't broken and then shrink the
1891 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1892 * automatically grant mappings that are too large.
1893 */
1894int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
1895 size_t newsize)
1896{
1897 struct vm_area_struct *vma;
7e660872
DH
1898 struct vm_region *region;
1899 pgoff_t low, high;
1900 size_t r_size, r_top;
1901
1902 low = newsize >> PAGE_SHIFT;
1903 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1904
1905 down_write(&nommu_region_sem);
1acf2e04 1906 i_mmap_lock_read(inode->i_mapping);
7e660872
DH
1907
1908 /* search for VMAs that fall within the dead zone */
6b2dbba8 1909 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
7e660872
DH
1910 /* found one - only interested if it's shared out of the page
1911 * cache */
1912 if (vma->vm_flags & VM_SHARED) {
1acf2e04 1913 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1914 up_write(&nommu_region_sem);
1915 return -ETXTBSY; /* not quite true, but near enough */
1916 }
1917 }
1918
1919 /* reduce any regions that overlap the dead zone - if in existence,
1920 * these will be pointed to by VMAs that don't overlap the dead zone
1921 *
1922 * we don't check for any regions that start beyond the EOF as there
1923 * shouldn't be any
1924 */
1acf2e04 1925 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
7e660872
DH
1926 if (!(vma->vm_flags & VM_SHARED))
1927 continue;
1928
1929 region = vma->vm_region;
1930 r_size = region->vm_top - region->vm_start;
1931 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
1932
1933 if (r_top > newsize) {
1934 region->vm_top -= r_top - newsize;
1935 if (region->vm_end > region->vm_top)
1936 region->vm_end = region->vm_top;
1937 }
1938 }
1939
1acf2e04 1940 i_mmap_unlock_read(inode->i_mapping);
7e660872
DH
1941 up_write(&nommu_region_sem);
1942 return 0;
1943}
c9b1d098
AS
1944
1945/*
1946 * Initialise sysctl_user_reserve_kbytes.
1947 *
1948 * This is intended to prevent a user from starting a single memory hogging
1949 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1950 * mode.
1951 *
1952 * The default value is min(3% of free memory, 128MB)
1953 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1954 */
1955static int __meminit init_user_reserve(void)
1956{
1957 unsigned long free_kbytes;
1958
1959 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1960
1961 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
1962 return 0;
1963}
a4bc6fc7 1964subsys_initcall(init_user_reserve);
4eeab4f5
AS
1965
1966/*
1967 * Initialise sysctl_admin_reserve_kbytes.
1968 *
1969 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1970 * to log in and kill a memory hogging process.
1971 *
1972 * Systems with more than 256MB will reserve 8MB, enough to recover
1973 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1974 * only reserve 3% of free pages by default.
1975 */
1976static int __meminit init_admin_reserve(void)
1977{
1978 unsigned long free_kbytes;
1979
1980 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
1981
1982 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
1983 return 0;
1984}
a4bc6fc7 1985subsys_initcall(init_admin_reserve);