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 *
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
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>
13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14 */
15
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
17
18 #include <linux/export.h>
19 #include <linux/mm.h>
20 #include <linux/vmacache.h>
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>
28 #include <linux/blkdev.h>
29 #include <linux/backing-dev.h>
30 #include <linux/compiler.h>
31 #include <linux/mount.h>
32 #include <linux/personality.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/printk.h>
38
39 #include <asm/uaccess.h>
40 #include <asm/tlb.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
43 #include "internal.h"
44
45 void *high_memory;
46 EXPORT_SYMBOL(high_memory);
47 struct page *mem_map;
48 unsigned long max_mapnr;
49 EXPORT_SYMBOL(max_mapnr);
50 unsigned long highest_memmap_pfn;
51 struct percpu_counter vm_committed_as;
52 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
53 int sysctl_overcommit_ratio = 50; /* default is 50% */
54 unsigned long sysctl_overcommit_kbytes __read_mostly;
55 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
56 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS;
57 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
58 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
59 int heap_stack_gap = 0;
60
61 atomic_long_t mmap_pages_allocated;
62
63 /*
64 * The global memory commitment made in the system can be a metric
65 * that can be used to drive ballooning decisions when Linux is hosted
66 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
67 * balancing memory across competing virtual machines that are hosted.
68 * Several metrics drive this policy engine including the guest reported
69 * memory commitment.
70 */
vm_memory_committed(void)71 unsigned long vm_memory_committed(void)
72 {
73 return percpu_counter_read_positive(&vm_committed_as);
74 }
75
76 EXPORT_SYMBOL_GPL(vm_memory_committed);
77
78 EXPORT_SYMBOL(mem_map);
79
80 /* list of mapped, potentially shareable regions */
81 static struct kmem_cache *vm_region_jar;
82 struct rb_root nommu_region_tree = RB_ROOT;
83 DECLARE_RWSEM(nommu_region_sem);
84
85 const struct vm_operations_struct generic_file_vm_ops = {
86 };
87
88 /*
89 * Return the total memory allocated for this pointer, not
90 * just what the caller asked for.
91 *
92 * Doesn't have to be accurate, i.e. may have races.
93 */
kobjsize(const void * objp)94 unsigned int kobjsize(const void *objp)
95 {
96 struct page *page;
97
98 /*
99 * If the object we have should not have ksize performed on it,
100 * return size of 0
101 */
102 if (!objp || !virt_addr_valid(objp))
103 return 0;
104
105 page = virt_to_head_page(objp);
106
107 /*
108 * If the allocator sets PageSlab, we know the pointer came from
109 * kmalloc().
110 */
111 if (PageSlab(page))
112 return ksize(objp);
113
114 /*
115 * If it's not a compound page, see if we have a matching VMA
116 * region. This test is intentionally done in reverse order,
117 * so if there's no VMA, we still fall through and hand back
118 * PAGE_SIZE for 0-order pages.
119 */
120 if (!PageCompound(page)) {
121 struct vm_area_struct *vma;
122
123 vma = find_vma(current->mm, (unsigned long)objp);
124 if (vma)
125 return vma->vm_end - vma->vm_start;
126 }
127
128 /*
129 * The ksize() function is only guaranteed to work for pointers
130 * returned by kmalloc(). So handle arbitrary pointers here.
131 */
132 return PAGE_SIZE << compound_order(page);
133 }
134
__get_user_pages(struct task_struct * tsk,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,unsigned int foll_flags,struct page ** pages,struct vm_area_struct ** vmas,int * nonblocking)135 long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
136 unsigned long start, unsigned long nr_pages,
137 unsigned int foll_flags, struct page **pages,
138 struct vm_area_struct **vmas, int *nonblocking)
139 {
140 struct vm_area_struct *vma;
141 unsigned long vm_flags;
142 int i;
143
144 /* calculate required read or write permissions.
145 * If FOLL_FORCE is set, we only require the "MAY" flags.
146 */
147 vm_flags = (foll_flags & FOLL_WRITE) ?
148 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
149 vm_flags &= (foll_flags & FOLL_FORCE) ?
150 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
151
152 for (i = 0; i < nr_pages; i++) {
153 vma = find_vma(mm, start);
154 if (!vma)
155 goto finish_or_fault;
156
157 /* protect what we can, including chardevs */
158 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
159 !(vm_flags & vma->vm_flags))
160 goto finish_or_fault;
161
162 if (pages) {
163 pages[i] = virt_to_page(start);
164 if (pages[i])
165 page_cache_get(pages[i]);
166 }
167 if (vmas)
168 vmas[i] = vma;
169 start = (start + PAGE_SIZE) & PAGE_MASK;
170 }
171
172 return i;
173
174 finish_or_fault:
175 return i ? : -EFAULT;
176 }
177
178 /*
179 * get a list of pages in an address range belonging to the specified process
180 * and indicate the VMA that covers each page
181 * - this is potentially dodgy as we may end incrementing the page count of a
182 * slab page or a secondary page from a compound page
183 * - don't permit access to VMAs that don't support it, such as I/O mappings
184 */
get_user_pages(struct task_struct * tsk,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,unsigned int gup_flags,struct page ** pages,struct vm_area_struct ** vmas)185 long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
186 unsigned long start, unsigned long nr_pages,
187 unsigned int gup_flags, struct page **pages,
188 struct vm_area_struct **vmas)
189 {
190 return __get_user_pages(tsk, mm, start, nr_pages,
191 gup_flags, pages, vmas, NULL);
192 }
193 EXPORT_SYMBOL(get_user_pages);
194
get_user_pages_locked(struct task_struct * tsk,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,unsigned int gup_flags,struct page ** pages,int * locked)195 long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm,
196 unsigned long start, unsigned long nr_pages,
197 unsigned int gup_flags, struct page **pages,
198 int *locked)
199 {
200 return get_user_pages(tsk, mm, start, nr_pages, gup_flags,
201 pages, NULL);
202 }
203 EXPORT_SYMBOL(get_user_pages_locked);
204
__get_user_pages_unlocked(struct task_struct * tsk,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,struct page ** pages,unsigned int gup_flags)205 long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
206 unsigned long start, unsigned long nr_pages,
207 struct page **pages, unsigned int gup_flags)
208 {
209 long ret;
210 down_read(&mm->mmap_sem);
211 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages,
212 NULL, NULL);
213 up_read(&mm->mmap_sem);
214 return ret;
215 }
216 EXPORT_SYMBOL(__get_user_pages_unlocked);
217
get_user_pages_unlocked(struct task_struct * tsk,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,struct page ** pages,unsigned int gup_flags)218 long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
219 unsigned long start, unsigned long nr_pages,
220 struct page **pages, unsigned int gup_flags)
221 {
222 return __get_user_pages_unlocked(tsk, mm, start, nr_pages,
223 pages, gup_flags);
224 }
225 EXPORT_SYMBOL(get_user_pages_unlocked);
226
227 /**
228 * follow_pfn - look up PFN at a user virtual address
229 * @vma: memory mapping
230 * @address: user virtual address
231 * @pfn: location to store found PFN
232 *
233 * Only IO mappings and raw PFN mappings are allowed.
234 *
235 * Returns zero and the pfn at @pfn on success, -ve otherwise.
236 */
follow_pfn(struct vm_area_struct * vma,unsigned long address,unsigned long * pfn)237 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
238 unsigned long *pfn)
239 {
240 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
241 return -EINVAL;
242
243 *pfn = address >> PAGE_SHIFT;
244 return 0;
245 }
246 EXPORT_SYMBOL(follow_pfn);
247
248 LIST_HEAD(vmap_area_list);
249
vfree(const void * addr)250 void vfree(const void *addr)
251 {
252 kfree(addr);
253 }
254 EXPORT_SYMBOL(vfree);
255
__vmalloc(unsigned long size,gfp_t gfp_mask,pgprot_t prot)256 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
257 {
258 /*
259 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
260 * returns only a logical address.
261 */
262 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
263 }
264 EXPORT_SYMBOL(__vmalloc);
265
vmalloc_user(unsigned long size)266 void *vmalloc_user(unsigned long size)
267 {
268 void *ret;
269
270 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
271 PAGE_KERNEL);
272 if (ret) {
273 struct vm_area_struct *vma;
274
275 down_write(¤t->mm->mmap_sem);
276 vma = find_vma(current->mm, (unsigned long)ret);
277 if (vma)
278 vma->vm_flags |= VM_USERMAP;
279 up_write(¤t->mm->mmap_sem);
280 }
281
282 return ret;
283 }
284 EXPORT_SYMBOL(vmalloc_user);
285
vmalloc_to_page(const void * addr)286 struct page *vmalloc_to_page(const void *addr)
287 {
288 return virt_to_page(addr);
289 }
290 EXPORT_SYMBOL(vmalloc_to_page);
291
vmalloc_to_pfn(const void * addr)292 unsigned long vmalloc_to_pfn(const void *addr)
293 {
294 return page_to_pfn(virt_to_page(addr));
295 }
296 EXPORT_SYMBOL(vmalloc_to_pfn);
297
vread(char * buf,char * addr,unsigned long count)298 long vread(char *buf, char *addr, unsigned long count)
299 {
300 /* Don't allow overflow */
301 if ((unsigned long) buf + count < count)
302 count = -(unsigned long) buf;
303
304 memcpy(buf, addr, count);
305 return count;
306 }
307
vwrite(char * buf,char * addr,unsigned long count)308 long vwrite(char *buf, char *addr, unsigned long count)
309 {
310 /* Don't allow overflow */
311 if ((unsigned long) addr + count < count)
312 count = -(unsigned long) addr;
313
314 memcpy(addr, buf, count);
315 return count;
316 }
317
318 /*
319 * vmalloc - allocate virtually contiguous memory
320 *
321 * @size: allocation size
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 */
vmalloc(unsigned long size)329 void *vmalloc(unsigned long size)
330 {
331 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
332 }
333 EXPORT_SYMBOL(vmalloc);
334
335 /*
336 * vzalloc - allocate virtually contiguous memory with zero fill
337 *
338 * @size: allocation size
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 */
vzalloc(unsigned long size)347 void *vzalloc(unsigned long size)
348 {
349 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
350 PAGE_KERNEL);
351 }
352 EXPORT_SYMBOL(vzalloc);
353
354 /**
355 * vmalloc_node - allocate memory on a specific node
356 * @size: allocation size
357 * @node: numa node
358 *
359 * Allocate enough pages to cover @size from the page level
360 * allocator and map them into contiguous kernel virtual space.
361 *
362 * For tight control over page level allocator and protection flags
363 * use __vmalloc() instead.
364 */
vmalloc_node(unsigned long size,int node)365 void *vmalloc_node(unsigned long size, int node)
366 {
367 return vmalloc(size);
368 }
369 EXPORT_SYMBOL(vmalloc_node);
370
371 /**
372 * vzalloc_node - allocate memory on a specific node with zero fill
373 * @size: allocation size
374 * @node: numa node
375 *
376 * Allocate enough pages to cover @size from the page level
377 * allocator and map them into contiguous kernel virtual space.
378 * The memory allocated is set to zero.
379 *
380 * For tight control over page level allocator and protection flags
381 * use __vmalloc() instead.
382 */
vzalloc_node(unsigned long size,int node)383 void *vzalloc_node(unsigned long size, int node)
384 {
385 return vzalloc(size);
386 }
387 EXPORT_SYMBOL(vzalloc_node);
388
389 #ifndef PAGE_KERNEL_EXEC
390 # define PAGE_KERNEL_EXEC PAGE_KERNEL
391 #endif
392
393 /**
394 * vmalloc_exec - allocate virtually contiguous, executable memory
395 * @size: allocation size
396 *
397 * Kernel-internal function to allocate enough pages to cover @size
398 * the page level allocator and map them into contiguous and
399 * executable kernel virtual space.
400 *
401 * For tight control over page level allocator and protection flags
402 * use __vmalloc() instead.
403 */
404
vmalloc_exec(unsigned long size)405 void *vmalloc_exec(unsigned long size)
406 {
407 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
408 }
409
410 /**
411 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
412 * @size: allocation size
413 *
414 * Allocate enough 32bit PA addressable pages to cover @size from the
415 * page level allocator and map them into contiguous kernel virtual space.
416 */
vmalloc_32(unsigned long size)417 void *vmalloc_32(unsigned long size)
418 {
419 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
420 }
421 EXPORT_SYMBOL(vmalloc_32);
422
423 /**
424 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
425 * @size: allocation size
426 *
427 * The resulting memory area is 32bit addressable and zeroed so it can be
428 * mapped to userspace without leaking data.
429 *
430 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
431 * remap_vmalloc_range() are permissible.
432 */
vmalloc_32_user(unsigned long size)433 void *vmalloc_32_user(unsigned long size)
434 {
435 /*
436 * We'll have to sort out the ZONE_DMA bits for 64-bit,
437 * but for now this can simply use vmalloc_user() directly.
438 */
439 return vmalloc_user(size);
440 }
441 EXPORT_SYMBOL(vmalloc_32_user);
442
vmap(struct page ** pages,unsigned int count,unsigned long flags,pgprot_t prot)443 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
444 {
445 BUG();
446 return NULL;
447 }
448 EXPORT_SYMBOL(vmap);
449
vunmap(const void * addr)450 void vunmap(const void *addr)
451 {
452 BUG();
453 }
454 EXPORT_SYMBOL(vunmap);
455
vm_map_ram(struct page ** pages,unsigned int count,int node,pgprot_t prot)456 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
457 {
458 BUG();
459 return NULL;
460 }
461 EXPORT_SYMBOL(vm_map_ram);
462
vm_unmap_ram(const void * mem,unsigned int count)463 void vm_unmap_ram(const void *mem, unsigned int count)
464 {
465 BUG();
466 }
467 EXPORT_SYMBOL(vm_unmap_ram);
468
vm_unmap_aliases(void)469 void vm_unmap_aliases(void)
470 {
471 }
472 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
473
474 /*
475 * Implement a stub for vmalloc_sync_[un]mapping() if the architecture
476 * chose not to have one.
477 */
vmalloc_sync_mappings(void)478 void __weak vmalloc_sync_mappings(void)
479 {
480 }
481
vmalloc_sync_unmappings(void)482 void __weak vmalloc_sync_unmappings(void)
483 {
484 }
485
486 /**
487 * alloc_vm_area - allocate a range of kernel address space
488 * @size: size of the area
489 *
490 * Returns: NULL on failure, vm_struct on success
491 *
492 * This function reserves a range of kernel address space, and
493 * allocates pagetables to map that range. No actual mappings
494 * are created. If the kernel address space is not shared
495 * between processes, it syncs the pagetable across all
496 * processes.
497 */
alloc_vm_area(size_t size,pte_t ** ptes)498 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
499 {
500 BUG();
501 return NULL;
502 }
503 EXPORT_SYMBOL_GPL(alloc_vm_area);
504
free_vm_area(struct vm_struct * area)505 void free_vm_area(struct vm_struct *area)
506 {
507 BUG();
508 }
509 EXPORT_SYMBOL_GPL(free_vm_area);
510
vm_insert_page(struct vm_area_struct * vma,unsigned long addr,struct page * page)511 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
512 struct page *page)
513 {
514 return -EINVAL;
515 }
516 EXPORT_SYMBOL(vm_insert_page);
517
518 /*
519 * sys_brk() for the most part doesn't need the global kernel
520 * lock, except when an application is doing something nasty
521 * like trying to un-brk an area that has already been mapped
522 * to a regular file. in this case, the unmapping will need
523 * to invoke file system routines that need the global lock.
524 */
SYSCALL_DEFINE1(brk,unsigned long,brk)525 SYSCALL_DEFINE1(brk, unsigned long, brk)
526 {
527 struct mm_struct *mm = current->mm;
528
529 if (brk < mm->start_brk || brk > mm->context.end_brk)
530 return mm->brk;
531
532 if (mm->brk == brk)
533 return mm->brk;
534
535 /*
536 * Always allow shrinking brk
537 */
538 if (brk <= mm->brk) {
539 mm->brk = brk;
540 return brk;
541 }
542
543 /*
544 * Ok, looks good - let it rip.
545 */
546 flush_icache_range(mm->brk, brk);
547 return mm->brk = brk;
548 }
549
550 /*
551 * initialise the VMA and region record slabs
552 */
mmap_init(void)553 void __init mmap_init(void)
554 {
555 int ret;
556
557 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
558 VM_BUG_ON(ret);
559 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
560 }
561
562 /*
563 * validate the region tree
564 * - the caller must hold the region lock
565 */
566 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
validate_nommu_regions(void)567 static noinline void validate_nommu_regions(void)
568 {
569 struct vm_region *region, *last;
570 struct rb_node *p, *lastp;
571
572 lastp = rb_first(&nommu_region_tree);
573 if (!lastp)
574 return;
575
576 last = rb_entry(lastp, struct vm_region, vm_rb);
577 BUG_ON(last->vm_end <= last->vm_start);
578 BUG_ON(last->vm_top < last->vm_end);
579
580 while ((p = rb_next(lastp))) {
581 region = rb_entry(p, struct vm_region, vm_rb);
582 last = rb_entry(lastp, struct vm_region, vm_rb);
583
584 BUG_ON(region->vm_end <= region->vm_start);
585 BUG_ON(region->vm_top < region->vm_end);
586 BUG_ON(region->vm_start < last->vm_top);
587
588 lastp = p;
589 }
590 }
591 #else
validate_nommu_regions(void)592 static void validate_nommu_regions(void)
593 {
594 }
595 #endif
596
597 /*
598 * add a region into the global tree
599 */
add_nommu_region(struct vm_region * region)600 static void add_nommu_region(struct vm_region *region)
601 {
602 struct vm_region *pregion;
603 struct rb_node **p, *parent;
604
605 validate_nommu_regions();
606
607 parent = NULL;
608 p = &nommu_region_tree.rb_node;
609 while (*p) {
610 parent = *p;
611 pregion = rb_entry(parent, struct vm_region, vm_rb);
612 if (region->vm_start < pregion->vm_start)
613 p = &(*p)->rb_left;
614 else if (region->vm_start > pregion->vm_start)
615 p = &(*p)->rb_right;
616 else if (pregion == region)
617 return;
618 else
619 BUG();
620 }
621
622 rb_link_node(®ion->vm_rb, parent, p);
623 rb_insert_color(®ion->vm_rb, &nommu_region_tree);
624
625 validate_nommu_regions();
626 }
627
628 /*
629 * delete a region from the global tree
630 */
delete_nommu_region(struct vm_region * region)631 static void delete_nommu_region(struct vm_region *region)
632 {
633 BUG_ON(!nommu_region_tree.rb_node);
634
635 validate_nommu_regions();
636 rb_erase(®ion->vm_rb, &nommu_region_tree);
637 validate_nommu_regions();
638 }
639
640 /*
641 * free a contiguous series of pages
642 */
free_page_series(unsigned long from,unsigned long to)643 static void free_page_series(unsigned long from, unsigned long to)
644 {
645 for (; from < to; from += PAGE_SIZE) {
646 struct page *page = virt_to_page(from);
647
648 atomic_long_dec(&mmap_pages_allocated);
649 put_page(page);
650 }
651 }
652
653 /*
654 * release a reference to a region
655 * - the caller must hold the region semaphore for writing, which this releases
656 * - the region may not have been added to the tree yet, in which case vm_top
657 * will equal vm_start
658 */
__put_nommu_region(struct vm_region * region)659 static void __put_nommu_region(struct vm_region *region)
660 __releases(nommu_region_sem)
661 {
662 BUG_ON(!nommu_region_tree.rb_node);
663
664 if (--region->vm_usage == 0) {
665 if (region->vm_top > region->vm_start)
666 delete_nommu_region(region);
667 up_write(&nommu_region_sem);
668
669 if (region->vm_file)
670 fput(region->vm_file);
671
672 /* IO memory and memory shared directly out of the pagecache
673 * from ramfs/tmpfs mustn't be released here */
674 if (region->vm_flags & VM_MAPPED_COPY)
675 free_page_series(region->vm_start, region->vm_top);
676 kmem_cache_free(vm_region_jar, region);
677 } else {
678 up_write(&nommu_region_sem);
679 }
680 }
681
682 /*
683 * release a reference to a region
684 */
put_nommu_region(struct vm_region * region)685 static void put_nommu_region(struct vm_region *region)
686 {
687 down_write(&nommu_region_sem);
688 __put_nommu_region(region);
689 }
690
691 /*
692 * update protection on a vma
693 */
protect_vma(struct vm_area_struct * vma,unsigned long flags)694 static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
695 {
696 #ifdef CONFIG_MPU
697 struct mm_struct *mm = vma->vm_mm;
698 long start = vma->vm_start & PAGE_MASK;
699 while (start < vma->vm_end) {
700 protect_page(mm, start, flags);
701 start += PAGE_SIZE;
702 }
703 update_protections(mm);
704 #endif
705 }
706
707 /*
708 * add a VMA into a process's mm_struct in the appropriate place in the list
709 * and tree and add to the address space's page tree also if not an anonymous
710 * page
711 * - should be called with mm->mmap_sem held writelocked
712 */
add_vma_to_mm(struct mm_struct * mm,struct vm_area_struct * vma)713 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
714 {
715 struct vm_area_struct *pvma, *prev;
716 struct address_space *mapping;
717 struct rb_node **p, *parent, *rb_prev;
718
719 BUG_ON(!vma->vm_region);
720
721 mm->map_count++;
722 vma->vm_mm = mm;
723
724 protect_vma(vma, vma->vm_flags);
725
726 /* add the VMA to the mapping */
727 if (vma->vm_file) {
728 mapping = vma->vm_file->f_mapping;
729
730 i_mmap_lock_write(mapping);
731 flush_dcache_mmap_lock(mapping);
732 vma_interval_tree_insert(vma, &mapping->i_mmap);
733 flush_dcache_mmap_unlock(mapping);
734 i_mmap_unlock_write(mapping);
735 }
736
737 /* add the VMA to the tree */
738 parent = rb_prev = NULL;
739 p = &mm->mm_rb.rb_node;
740 while (*p) {
741 parent = *p;
742 pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
743
744 /* sort by: start addr, end addr, VMA struct addr in that order
745 * (the latter is necessary as we may get identical VMAs) */
746 if (vma->vm_start < pvma->vm_start)
747 p = &(*p)->rb_left;
748 else if (vma->vm_start > pvma->vm_start) {
749 rb_prev = parent;
750 p = &(*p)->rb_right;
751 } else if (vma->vm_end < pvma->vm_end)
752 p = &(*p)->rb_left;
753 else if (vma->vm_end > pvma->vm_end) {
754 rb_prev = parent;
755 p = &(*p)->rb_right;
756 } else if (vma < pvma)
757 p = &(*p)->rb_left;
758 else if (vma > pvma) {
759 rb_prev = parent;
760 p = &(*p)->rb_right;
761 } else
762 BUG();
763 }
764
765 rb_link_node(&vma->vm_rb, parent, p);
766 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
767
768 /* add VMA to the VMA list also */
769 prev = NULL;
770 if (rb_prev)
771 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
772
773 __vma_link_list(mm, vma, prev, parent);
774 }
775
776 /*
777 * delete a VMA from its owning mm_struct and address space
778 */
delete_vma_from_mm(struct vm_area_struct * vma)779 static void delete_vma_from_mm(struct vm_area_struct *vma)
780 {
781 int i;
782 struct address_space *mapping;
783 struct mm_struct *mm = vma->vm_mm;
784 struct task_struct *curr = current;
785
786 protect_vma(vma, 0);
787
788 mm->map_count--;
789 for (i = 0; i < VMACACHE_SIZE; i++) {
790 /* if the vma is cached, invalidate the entire cache */
791 if (curr->vmacache[i] == vma) {
792 vmacache_invalidate(mm);
793 break;
794 }
795 }
796
797 /* remove the VMA from the mapping */
798 if (vma->vm_file) {
799 mapping = vma->vm_file->f_mapping;
800
801 i_mmap_lock_write(mapping);
802 flush_dcache_mmap_lock(mapping);
803 vma_interval_tree_remove(vma, &mapping->i_mmap);
804 flush_dcache_mmap_unlock(mapping);
805 i_mmap_unlock_write(mapping);
806 }
807
808 /* remove from the MM's tree and list */
809 rb_erase(&vma->vm_rb, &mm->mm_rb);
810
811 if (vma->vm_prev)
812 vma->vm_prev->vm_next = vma->vm_next;
813 else
814 mm->mmap = vma->vm_next;
815
816 if (vma->vm_next)
817 vma->vm_next->vm_prev = vma->vm_prev;
818 }
819
820 /*
821 * destroy a VMA record
822 */
delete_vma(struct mm_struct * mm,struct vm_area_struct * vma)823 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
824 {
825 if (vma->vm_ops && vma->vm_ops->close)
826 vma->vm_ops->close(vma);
827 if (vma->vm_file)
828 fput(vma->vm_file);
829 put_nommu_region(vma->vm_region);
830 kmem_cache_free(vm_area_cachep, vma);
831 }
832
833 /*
834 * look up the first VMA in which addr resides, NULL if none
835 * - should be called with mm->mmap_sem at least held readlocked
836 */
find_vma(struct mm_struct * mm,unsigned long addr)837 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
838 {
839 struct vm_area_struct *vma;
840
841 /* check the cache first */
842 vma = vmacache_find(mm, addr);
843 if (likely(vma))
844 return vma;
845
846 /* trawl the list (there may be multiple mappings in which addr
847 * resides) */
848 for (vma = mm->mmap; vma; vma = vma->vm_next) {
849 if (vma->vm_start > addr)
850 return NULL;
851 if (vma->vm_end > addr) {
852 vmacache_update(addr, vma);
853 return vma;
854 }
855 }
856
857 return NULL;
858 }
859 EXPORT_SYMBOL(find_vma);
860
861 /*
862 * find a VMA
863 * - we don't extend stack VMAs under NOMMU conditions
864 */
find_extend_vma(struct mm_struct * mm,unsigned long addr)865 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
866 {
867 return find_vma(mm, addr);
868 }
869
870 /*
871 * expand a stack to a given address
872 * - not supported under NOMMU conditions
873 */
expand_stack(struct vm_area_struct * vma,unsigned long address)874 int expand_stack(struct vm_area_struct *vma, unsigned long address)
875 {
876 return -ENOMEM;
877 }
878
879 /*
880 * look up the first VMA exactly that exactly matches addr
881 * - should be called with mm->mmap_sem at least held readlocked
882 */
find_vma_exact(struct mm_struct * mm,unsigned long addr,unsigned long len)883 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
884 unsigned long addr,
885 unsigned long len)
886 {
887 struct vm_area_struct *vma;
888 unsigned long end = addr + len;
889
890 /* check the cache first */
891 vma = vmacache_find_exact(mm, addr, end);
892 if (vma)
893 return vma;
894
895 /* trawl the list (there may be multiple mappings in which addr
896 * resides) */
897 for (vma = mm->mmap; vma; vma = vma->vm_next) {
898 if (vma->vm_start < addr)
899 continue;
900 if (vma->vm_start > addr)
901 return NULL;
902 if (vma->vm_end == end) {
903 vmacache_update(addr, vma);
904 return vma;
905 }
906 }
907
908 return NULL;
909 }
910
911 /*
912 * determine whether a mapping should be permitted and, if so, what sort of
913 * mapping we're capable of supporting
914 */
validate_mmap_request(struct file * file,unsigned long addr,unsigned long len,unsigned long prot,unsigned long flags,unsigned long pgoff,unsigned long * _capabilities)915 static int validate_mmap_request(struct file *file,
916 unsigned long addr,
917 unsigned long len,
918 unsigned long prot,
919 unsigned long flags,
920 unsigned long pgoff,
921 unsigned long *_capabilities)
922 {
923 unsigned long capabilities, rlen;
924 int ret;
925
926 /* do the simple checks first */
927 if (flags & MAP_FIXED)
928 return -EINVAL;
929
930 if ((flags & MAP_TYPE) != MAP_PRIVATE &&
931 (flags & MAP_TYPE) != MAP_SHARED)
932 return -EINVAL;
933
934 if (!len)
935 return -EINVAL;
936
937 /* Careful about overflows.. */
938 rlen = PAGE_ALIGN(len);
939 if (!rlen || rlen > TASK_SIZE)
940 return -ENOMEM;
941
942 /* offset overflow? */
943 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
944 return -EOVERFLOW;
945
946 if (file) {
947 /* files must support mmap */
948 if (!file->f_op->mmap)
949 return -ENODEV;
950
951 /* work out if what we've got could possibly be shared
952 * - we support chardevs that provide their own "memory"
953 * - we support files/blockdevs that are memory backed
954 */
955 if (file->f_op->mmap_capabilities) {
956 capabilities = file->f_op->mmap_capabilities(file);
957 } else {
958 /* no explicit capabilities set, so assume some
959 * defaults */
960 switch (file_inode(file)->i_mode & S_IFMT) {
961 case S_IFREG:
962 case S_IFBLK:
963 capabilities = NOMMU_MAP_COPY;
964 break;
965
966 case S_IFCHR:
967 capabilities =
968 NOMMU_MAP_DIRECT |
969 NOMMU_MAP_READ |
970 NOMMU_MAP_WRITE;
971 break;
972
973 default:
974 return -EINVAL;
975 }
976 }
977
978 /* eliminate any capabilities that we can't support on this
979 * device */
980 if (!file->f_op->get_unmapped_area)
981 capabilities &= ~NOMMU_MAP_DIRECT;
982 if (!(file->f_mode & FMODE_CAN_READ))
983 capabilities &= ~NOMMU_MAP_COPY;
984
985 /* The file shall have been opened with read permission. */
986 if (!(file->f_mode & FMODE_READ))
987 return -EACCES;
988
989 if (flags & MAP_SHARED) {
990 /* do checks for writing, appending and locking */
991 if ((prot & PROT_WRITE) &&
992 !(file->f_mode & FMODE_WRITE))
993 return -EACCES;
994
995 if (IS_APPEND(file_inode(file)) &&
996 (file->f_mode & FMODE_WRITE))
997 return -EACCES;
998
999 if (locks_verify_locked(file))
1000 return -EAGAIN;
1001
1002 if (!(capabilities & NOMMU_MAP_DIRECT))
1003 return -ENODEV;
1004
1005 /* we mustn't privatise shared mappings */
1006 capabilities &= ~NOMMU_MAP_COPY;
1007 } else {
1008 /* we're going to read the file into private memory we
1009 * allocate */
1010 if (!(capabilities & NOMMU_MAP_COPY))
1011 return -ENODEV;
1012
1013 /* we don't permit a private writable mapping to be
1014 * shared with the backing device */
1015 if (prot & PROT_WRITE)
1016 capabilities &= ~NOMMU_MAP_DIRECT;
1017 }
1018
1019 if (capabilities & NOMMU_MAP_DIRECT) {
1020 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) ||
1021 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) ||
1022 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC))
1023 ) {
1024 capabilities &= ~NOMMU_MAP_DIRECT;
1025 if (flags & MAP_SHARED) {
1026 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1027 return -EINVAL;
1028 }
1029 }
1030 }
1031
1032 /* handle executable mappings and implied executable
1033 * mappings */
1034 if (path_noexec(&file->f_path)) {
1035 if (prot & PROT_EXEC)
1036 return -EPERM;
1037 } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1038 /* handle implication of PROT_EXEC by PROT_READ */
1039 if (current->personality & READ_IMPLIES_EXEC) {
1040 if (capabilities & NOMMU_MAP_EXEC)
1041 prot |= PROT_EXEC;
1042 }
1043 } else if ((prot & PROT_READ) &&
1044 (prot & PROT_EXEC) &&
1045 !(capabilities & NOMMU_MAP_EXEC)
1046 ) {
1047 /* backing file is not executable, try to copy */
1048 capabilities &= ~NOMMU_MAP_DIRECT;
1049 }
1050 } else {
1051 /* anonymous mappings are always memory backed and can be
1052 * privately mapped
1053 */
1054 capabilities = NOMMU_MAP_COPY;
1055
1056 /* handle PROT_EXEC implication by PROT_READ */
1057 if ((prot & PROT_READ) &&
1058 (current->personality & READ_IMPLIES_EXEC))
1059 prot |= PROT_EXEC;
1060 }
1061
1062 /* allow the security API to have its say */
1063 ret = security_mmap_addr(addr);
1064 if (ret < 0)
1065 return ret;
1066
1067 /* looks okay */
1068 *_capabilities = capabilities;
1069 return 0;
1070 }
1071
1072 /*
1073 * we've determined that we can make the mapping, now translate what we
1074 * now know into VMA flags
1075 */
determine_vm_flags(struct file * file,unsigned long prot,unsigned long flags,unsigned long capabilities)1076 static unsigned long determine_vm_flags(struct file *file,
1077 unsigned long prot,
1078 unsigned long flags,
1079 unsigned long capabilities)
1080 {
1081 unsigned long vm_flags;
1082
1083 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1084 /* vm_flags |= mm->def_flags; */
1085
1086 if (!(capabilities & NOMMU_MAP_DIRECT)) {
1087 /* attempt to share read-only copies of mapped file chunks */
1088 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1089 if (file && !(prot & PROT_WRITE))
1090 vm_flags |= VM_MAYSHARE;
1091 } else {
1092 /* overlay a shareable mapping on the backing device or inode
1093 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1094 * romfs/cramfs */
1095 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS);
1096 if (flags & MAP_SHARED)
1097 vm_flags |= VM_SHARED;
1098 }
1099
1100 /* refuse to let anyone share private mappings with this process if
1101 * it's being traced - otherwise breakpoints set in it may interfere
1102 * with another untraced process
1103 */
1104 if ((flags & MAP_PRIVATE) && current->ptrace)
1105 vm_flags &= ~VM_MAYSHARE;
1106
1107 return vm_flags;
1108 }
1109
1110 /*
1111 * set up a shared mapping on a file (the driver or filesystem provides and
1112 * pins the storage)
1113 */
do_mmap_shared_file(struct vm_area_struct * vma)1114 static int do_mmap_shared_file(struct vm_area_struct *vma)
1115 {
1116 int ret;
1117
1118 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1119 if (ret == 0) {
1120 vma->vm_region->vm_top = vma->vm_region->vm_end;
1121 return 0;
1122 }
1123 if (ret != -ENOSYS)
1124 return ret;
1125
1126 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1127 * opposed to tried but failed) so we can only give a suitable error as
1128 * it's not possible to make a private copy if MAP_SHARED was given */
1129 return -ENODEV;
1130 }
1131
1132 /*
1133 * set up a private mapping or an anonymous shared mapping
1134 */
do_mmap_private(struct vm_area_struct * vma,struct vm_region * region,unsigned long len,unsigned long capabilities)1135 static int do_mmap_private(struct vm_area_struct *vma,
1136 struct vm_region *region,
1137 unsigned long len,
1138 unsigned long capabilities)
1139 {
1140 unsigned long total, point;
1141 void *base;
1142 int ret, order;
1143
1144 /* invoke the file's mapping function so that it can keep track of
1145 * shared mappings on devices or memory
1146 * - VM_MAYSHARE will be set if it may attempt to share
1147 */
1148 if (capabilities & NOMMU_MAP_DIRECT) {
1149 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1150 if (ret == 0) {
1151 /* shouldn't return success if we're not sharing */
1152 BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1153 vma->vm_region->vm_top = vma->vm_region->vm_end;
1154 return 0;
1155 }
1156 if (ret != -ENOSYS)
1157 return ret;
1158
1159 /* getting an ENOSYS error indicates that direct mmap isn't
1160 * possible (as opposed to tried but failed) so we'll try to
1161 * make a private copy of the data and map that instead */
1162 }
1163
1164
1165 /* allocate some memory to hold the mapping
1166 * - note that this may not return a page-aligned address if the object
1167 * we're allocating is smaller than a page
1168 */
1169 order = get_order(len);
1170 total = 1 << order;
1171 point = len >> PAGE_SHIFT;
1172
1173 /* we don't want to allocate a power-of-2 sized page set */
1174 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages)
1175 total = point;
1176
1177 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL);
1178 if (!base)
1179 goto enomem;
1180
1181 atomic_long_add(total, &mmap_pages_allocated);
1182
1183 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1184 region->vm_start = (unsigned long) base;
1185 region->vm_end = region->vm_start + len;
1186 region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1187
1188 vma->vm_start = region->vm_start;
1189 vma->vm_end = region->vm_start + len;
1190
1191 if (vma->vm_file) {
1192 /* read the contents of a file into the copy */
1193 mm_segment_t old_fs;
1194 loff_t fpos;
1195
1196 fpos = vma->vm_pgoff;
1197 fpos <<= PAGE_SHIFT;
1198
1199 old_fs = get_fs();
1200 set_fs(KERNEL_DS);
1201 ret = __vfs_read(vma->vm_file, base, len, &fpos);
1202 set_fs(old_fs);
1203
1204 if (ret < 0)
1205 goto error_free;
1206
1207 /* clear the last little bit */
1208 if (ret < len)
1209 memset(base + ret, 0, len - ret);
1210
1211 }
1212
1213 return 0;
1214
1215 error_free:
1216 free_page_series(region->vm_start, region->vm_top);
1217 region->vm_start = vma->vm_start = 0;
1218 region->vm_end = vma->vm_end = 0;
1219 region->vm_top = 0;
1220 return ret;
1221
1222 enomem:
1223 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1224 len, current->pid, current->comm);
1225 show_free_areas(0);
1226 return -ENOMEM;
1227 }
1228
1229 /*
1230 * handle mapping creation for uClinux
1231 */
do_mmap(struct file * file,unsigned long addr,unsigned long len,unsigned long prot,unsigned long flags,vm_flags_t vm_flags,unsigned long pgoff,unsigned long * populate)1232 unsigned long do_mmap(struct file *file,
1233 unsigned long addr,
1234 unsigned long len,
1235 unsigned long prot,
1236 unsigned long flags,
1237 vm_flags_t vm_flags,
1238 unsigned long pgoff,
1239 unsigned long *populate)
1240 {
1241 struct vm_area_struct *vma;
1242 struct vm_region *region;
1243 struct rb_node *rb;
1244 unsigned long capabilities, result;
1245 int ret;
1246
1247 *populate = 0;
1248
1249 /* decide whether we should attempt the mapping, and if so what sort of
1250 * mapping */
1251 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1252 &capabilities);
1253 if (ret < 0)
1254 return ret;
1255
1256 /* we ignore the address hint */
1257 addr = 0;
1258 len = PAGE_ALIGN(len);
1259
1260 /* we've determined that we can make the mapping, now translate what we
1261 * now know into VMA flags */
1262 vm_flags |= determine_vm_flags(file, prot, flags, capabilities);
1263
1264 /* we're going to need to record the mapping */
1265 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1266 if (!region)
1267 goto error_getting_region;
1268
1269 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1270 if (!vma)
1271 goto error_getting_vma;
1272
1273 region->vm_usage = 1;
1274 region->vm_flags = vm_flags;
1275 region->vm_pgoff = pgoff;
1276
1277 INIT_LIST_HEAD(&vma->anon_vma_chain);
1278 vma->vm_flags = vm_flags;
1279 vma->vm_pgoff = pgoff;
1280
1281 if (file) {
1282 region->vm_file = get_file(file);
1283 vma->vm_file = get_file(file);
1284 }
1285
1286 down_write(&nommu_region_sem);
1287
1288 /* if we want to share, we need to check for regions created by other
1289 * mmap() calls that overlap with our proposed mapping
1290 * - we can only share with a superset match on most regular files
1291 * - shared mappings on character devices and memory backed files are
1292 * permitted to overlap inexactly as far as we are concerned for in
1293 * these cases, sharing is handled in the driver or filesystem rather
1294 * than here
1295 */
1296 if (vm_flags & VM_MAYSHARE) {
1297 struct vm_region *pregion;
1298 unsigned long pglen, rpglen, pgend, rpgend, start;
1299
1300 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1301 pgend = pgoff + pglen;
1302
1303 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1304 pregion = rb_entry(rb, struct vm_region, vm_rb);
1305
1306 if (!(pregion->vm_flags & VM_MAYSHARE))
1307 continue;
1308
1309 /* search for overlapping mappings on the same file */
1310 if (file_inode(pregion->vm_file) !=
1311 file_inode(file))
1312 continue;
1313
1314 if (pregion->vm_pgoff >= pgend)
1315 continue;
1316
1317 rpglen = pregion->vm_end - pregion->vm_start;
1318 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1319 rpgend = pregion->vm_pgoff + rpglen;
1320 if (pgoff >= rpgend)
1321 continue;
1322
1323 /* handle inexactly overlapping matches between
1324 * mappings */
1325 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1326 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1327 /* new mapping is not a subset of the region */
1328 if (!(capabilities & NOMMU_MAP_DIRECT))
1329 goto sharing_violation;
1330 continue;
1331 }
1332
1333 /* we've found a region we can share */
1334 pregion->vm_usage++;
1335 vma->vm_region = pregion;
1336 start = pregion->vm_start;
1337 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1338 vma->vm_start = start;
1339 vma->vm_end = start + len;
1340
1341 if (pregion->vm_flags & VM_MAPPED_COPY)
1342 vma->vm_flags |= VM_MAPPED_COPY;
1343 else {
1344 ret = do_mmap_shared_file(vma);
1345 if (ret < 0) {
1346 vma->vm_region = NULL;
1347 vma->vm_start = 0;
1348 vma->vm_end = 0;
1349 pregion->vm_usage--;
1350 pregion = NULL;
1351 goto error_just_free;
1352 }
1353 }
1354 fput(region->vm_file);
1355 kmem_cache_free(vm_region_jar, region);
1356 region = pregion;
1357 result = start;
1358 goto share;
1359 }
1360
1361 /* obtain the address at which to make a shared mapping
1362 * - this is the hook for quasi-memory character devices to
1363 * tell us the location of a shared mapping
1364 */
1365 if (capabilities & NOMMU_MAP_DIRECT) {
1366 addr = file->f_op->get_unmapped_area(file, addr, len,
1367 pgoff, flags);
1368 if (IS_ERR_VALUE(addr)) {
1369 ret = addr;
1370 if (ret != -ENOSYS)
1371 goto error_just_free;
1372
1373 /* the driver refused to tell us where to site
1374 * the mapping so we'll have to attempt to copy
1375 * it */
1376 ret = -ENODEV;
1377 if (!(capabilities & NOMMU_MAP_COPY))
1378 goto error_just_free;
1379
1380 capabilities &= ~NOMMU_MAP_DIRECT;
1381 } else {
1382 vma->vm_start = region->vm_start = addr;
1383 vma->vm_end = region->vm_end = addr + len;
1384 }
1385 }
1386 }
1387
1388 vma->vm_region = region;
1389
1390 /* set up the mapping
1391 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1392 */
1393 if (file && vma->vm_flags & VM_SHARED)
1394 ret = do_mmap_shared_file(vma);
1395 else
1396 ret = do_mmap_private(vma, region, len, capabilities);
1397 if (ret < 0)
1398 goto error_just_free;
1399 add_nommu_region(region);
1400
1401 /* clear anonymous mappings that don't ask for uninitialized data */
1402 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1403 memset((void *)region->vm_start, 0,
1404 region->vm_end - region->vm_start);
1405
1406 /* okay... we have a mapping; now we have to register it */
1407 result = vma->vm_start;
1408
1409 current->mm->total_vm += len >> PAGE_SHIFT;
1410
1411 share:
1412 add_vma_to_mm(current->mm, vma);
1413
1414 /* we flush the region from the icache only when the first executable
1415 * mapping of it is made */
1416 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1417 flush_icache_range(region->vm_start, region->vm_end);
1418 region->vm_icache_flushed = true;
1419 }
1420
1421 up_write(&nommu_region_sem);
1422
1423 return result;
1424
1425 error_just_free:
1426 up_write(&nommu_region_sem);
1427 error:
1428 if (region->vm_file)
1429 fput(region->vm_file);
1430 kmem_cache_free(vm_region_jar, region);
1431 if (vma->vm_file)
1432 fput(vma->vm_file);
1433 kmem_cache_free(vm_area_cachep, vma);
1434 return ret;
1435
1436 sharing_violation:
1437 up_write(&nommu_region_sem);
1438 pr_warn("Attempt to share mismatched mappings\n");
1439 ret = -EINVAL;
1440 goto error;
1441
1442 error_getting_vma:
1443 kmem_cache_free(vm_region_jar, region);
1444 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1445 len, current->pid);
1446 show_free_areas(0);
1447 return -ENOMEM;
1448
1449 error_getting_region:
1450 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1451 len, current->pid);
1452 show_free_areas(0);
1453 return -ENOMEM;
1454 }
1455
SYSCALL_DEFINE6(mmap_pgoff,unsigned long,addr,unsigned long,len,unsigned long,prot,unsigned long,flags,unsigned long,fd,unsigned long,pgoff)1456 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1457 unsigned long, prot, unsigned long, flags,
1458 unsigned long, fd, unsigned long, pgoff)
1459 {
1460 struct file *file = NULL;
1461 unsigned long retval = -EBADF;
1462
1463 audit_mmap_fd(fd, flags);
1464 if (!(flags & MAP_ANONYMOUS)) {
1465 file = fget(fd);
1466 if (!file)
1467 goto out;
1468 }
1469
1470 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1471
1472 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1473
1474 if (file)
1475 fput(file);
1476 out:
1477 return retval;
1478 }
1479
1480 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1481 struct mmap_arg_struct {
1482 unsigned long addr;
1483 unsigned long len;
1484 unsigned long prot;
1485 unsigned long flags;
1486 unsigned long fd;
1487 unsigned long offset;
1488 };
1489
SYSCALL_DEFINE1(old_mmap,struct mmap_arg_struct __user *,arg)1490 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1491 {
1492 struct mmap_arg_struct a;
1493
1494 if (copy_from_user(&a, arg, sizeof(a)))
1495 return -EFAULT;
1496 if (offset_in_page(a.offset))
1497 return -EINVAL;
1498
1499 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1500 a.offset >> PAGE_SHIFT);
1501 }
1502 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1503
1504 /*
1505 * split a vma into two pieces at address 'addr', a new vma is allocated either
1506 * for the first part or the tail.
1507 */
split_vma(struct mm_struct * mm,struct vm_area_struct * vma,unsigned long addr,int new_below)1508 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1509 unsigned long addr, int new_below)
1510 {
1511 struct vm_area_struct *new;
1512 struct vm_region *region;
1513 unsigned long npages;
1514
1515 /* we're only permitted to split anonymous regions (these should have
1516 * only a single usage on the region) */
1517 if (vma->vm_file)
1518 return -ENOMEM;
1519
1520 if (mm->map_count >= sysctl_max_map_count)
1521 return -ENOMEM;
1522
1523 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1524 if (!region)
1525 return -ENOMEM;
1526
1527 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1528 if (!new) {
1529 kmem_cache_free(vm_region_jar, region);
1530 return -ENOMEM;
1531 }
1532
1533 /* most fields are the same, copy all, and then fixup */
1534 *new = *vma;
1535 *region = *vma->vm_region;
1536 new->vm_region = region;
1537
1538 npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1539
1540 if (new_below) {
1541 region->vm_top = region->vm_end = new->vm_end = addr;
1542 } else {
1543 region->vm_start = new->vm_start = addr;
1544 region->vm_pgoff = new->vm_pgoff += npages;
1545 }
1546
1547 if (new->vm_ops && new->vm_ops->open)
1548 new->vm_ops->open(new);
1549
1550 delete_vma_from_mm(vma);
1551 down_write(&nommu_region_sem);
1552 delete_nommu_region(vma->vm_region);
1553 if (new_below) {
1554 vma->vm_region->vm_start = vma->vm_start = addr;
1555 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1556 } else {
1557 vma->vm_region->vm_end = vma->vm_end = addr;
1558 vma->vm_region->vm_top = addr;
1559 }
1560 add_nommu_region(vma->vm_region);
1561 add_nommu_region(new->vm_region);
1562 up_write(&nommu_region_sem);
1563 add_vma_to_mm(mm, vma);
1564 add_vma_to_mm(mm, new);
1565 return 0;
1566 }
1567
1568 /*
1569 * shrink a VMA by removing the specified chunk from either the beginning or
1570 * the end
1571 */
shrink_vma(struct mm_struct * mm,struct vm_area_struct * vma,unsigned long from,unsigned long to)1572 static int shrink_vma(struct mm_struct *mm,
1573 struct vm_area_struct *vma,
1574 unsigned long from, unsigned long to)
1575 {
1576 struct vm_region *region;
1577
1578 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1579 * and list */
1580 delete_vma_from_mm(vma);
1581 if (from > vma->vm_start)
1582 vma->vm_end = from;
1583 else
1584 vma->vm_start = to;
1585 add_vma_to_mm(mm, vma);
1586
1587 /* cut the backing region down to size */
1588 region = vma->vm_region;
1589 BUG_ON(region->vm_usage != 1);
1590
1591 down_write(&nommu_region_sem);
1592 delete_nommu_region(region);
1593 if (from > region->vm_start) {
1594 to = region->vm_top;
1595 region->vm_top = region->vm_end = from;
1596 } else {
1597 region->vm_start = to;
1598 }
1599 add_nommu_region(region);
1600 up_write(&nommu_region_sem);
1601
1602 free_page_series(from, to);
1603 return 0;
1604 }
1605
1606 /*
1607 * release a mapping
1608 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1609 * VMA, though it need not cover the whole VMA
1610 */
do_munmap(struct mm_struct * mm,unsigned long start,size_t len)1611 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1612 {
1613 struct vm_area_struct *vma;
1614 unsigned long end;
1615 int ret;
1616
1617 len = PAGE_ALIGN(len);
1618 if (len == 0)
1619 return -EINVAL;
1620
1621 end = start + len;
1622
1623 /* find the first potentially overlapping VMA */
1624 vma = find_vma(mm, start);
1625 if (!vma) {
1626 static int limit;
1627 if (limit < 5) {
1628 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1629 current->pid, current->comm,
1630 start, start + len - 1);
1631 limit++;
1632 }
1633 return -EINVAL;
1634 }
1635
1636 /* we're allowed to split an anonymous VMA but not a file-backed one */
1637 if (vma->vm_file) {
1638 do {
1639 if (start > vma->vm_start)
1640 return -EINVAL;
1641 if (end == vma->vm_end)
1642 goto erase_whole_vma;
1643 vma = vma->vm_next;
1644 } while (vma);
1645 return -EINVAL;
1646 } else {
1647 /* the chunk must be a subset of the VMA found */
1648 if (start == vma->vm_start && end == vma->vm_end)
1649 goto erase_whole_vma;
1650 if (start < vma->vm_start || end > vma->vm_end)
1651 return -EINVAL;
1652 if (offset_in_page(start))
1653 return -EINVAL;
1654 if (end != vma->vm_end && offset_in_page(end))
1655 return -EINVAL;
1656 if (start != vma->vm_start && end != vma->vm_end) {
1657 ret = split_vma(mm, vma, start, 1);
1658 if (ret < 0)
1659 return ret;
1660 }
1661 return shrink_vma(mm, vma, start, end);
1662 }
1663
1664 erase_whole_vma:
1665 delete_vma_from_mm(vma);
1666 delete_vma(mm, vma);
1667 return 0;
1668 }
1669 EXPORT_SYMBOL(do_munmap);
1670
vm_munmap(unsigned long addr,size_t len)1671 int vm_munmap(unsigned long addr, size_t len)
1672 {
1673 struct mm_struct *mm = current->mm;
1674 int ret;
1675
1676 down_write(&mm->mmap_sem);
1677 ret = do_munmap(mm, addr, len);
1678 up_write(&mm->mmap_sem);
1679 return ret;
1680 }
1681 EXPORT_SYMBOL(vm_munmap);
1682
SYSCALL_DEFINE2(munmap,unsigned long,addr,size_t,len)1683 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1684 {
1685 return vm_munmap(addr, len);
1686 }
1687
1688 /*
1689 * release all the mappings made in a process's VM space
1690 */
exit_mmap(struct mm_struct * mm)1691 void exit_mmap(struct mm_struct *mm)
1692 {
1693 struct vm_area_struct *vma;
1694
1695 if (!mm)
1696 return;
1697
1698 mm->total_vm = 0;
1699
1700 while ((vma = mm->mmap)) {
1701 mm->mmap = vma->vm_next;
1702 delete_vma_from_mm(vma);
1703 delete_vma(mm, vma);
1704 cond_resched();
1705 }
1706 }
1707
vm_brk(unsigned long addr,unsigned long len)1708 unsigned long vm_brk(unsigned long addr, unsigned long len)
1709 {
1710 return -ENOMEM;
1711 }
1712
1713 /*
1714 * expand (or shrink) an existing mapping, potentially moving it at the same
1715 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1716 *
1717 * under NOMMU conditions, we only permit changing a mapping's size, and only
1718 * as long as it stays within the region allocated by do_mmap_private() and the
1719 * block is not shareable
1720 *
1721 * MREMAP_FIXED is not supported under NOMMU conditions
1722 */
do_mremap(unsigned long addr,unsigned long old_len,unsigned long new_len,unsigned long flags,unsigned long new_addr)1723 static unsigned long do_mremap(unsigned long addr,
1724 unsigned long old_len, unsigned long new_len,
1725 unsigned long flags, unsigned long new_addr)
1726 {
1727 struct vm_area_struct *vma;
1728
1729 /* insanity checks first */
1730 old_len = PAGE_ALIGN(old_len);
1731 new_len = PAGE_ALIGN(new_len);
1732 if (old_len == 0 || new_len == 0)
1733 return (unsigned long) -EINVAL;
1734
1735 if (offset_in_page(addr))
1736 return -EINVAL;
1737
1738 if (flags & MREMAP_FIXED && new_addr != addr)
1739 return (unsigned long) -EINVAL;
1740
1741 vma = find_vma_exact(current->mm, addr, old_len);
1742 if (!vma)
1743 return (unsigned long) -EINVAL;
1744
1745 if (vma->vm_end != vma->vm_start + old_len)
1746 return (unsigned long) -EFAULT;
1747
1748 if (vma->vm_flags & VM_MAYSHARE)
1749 return (unsigned long) -EPERM;
1750
1751 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1752 return (unsigned long) -ENOMEM;
1753
1754 /* all checks complete - do it */
1755 vma->vm_end = vma->vm_start + new_len;
1756 return vma->vm_start;
1757 }
1758
SYSCALL_DEFINE5(mremap,unsigned long,addr,unsigned long,old_len,unsigned long,new_len,unsigned long,flags,unsigned long,new_addr)1759 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1760 unsigned long, new_len, unsigned long, flags,
1761 unsigned long, new_addr)
1762 {
1763 unsigned long ret;
1764
1765 down_write(¤t->mm->mmap_sem);
1766 ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1767 up_write(¤t->mm->mmap_sem);
1768 return ret;
1769 }
1770
follow_page_mask(struct vm_area_struct * vma,unsigned long address,unsigned int flags,unsigned int * page_mask)1771 struct page *follow_page_mask(struct vm_area_struct *vma,
1772 unsigned long address, unsigned int flags,
1773 unsigned int *page_mask)
1774 {
1775 *page_mask = 0;
1776 return NULL;
1777 }
1778
remap_pfn_range(struct vm_area_struct * vma,unsigned long addr,unsigned long pfn,unsigned long size,pgprot_t prot)1779 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1780 unsigned long pfn, unsigned long size, pgprot_t prot)
1781 {
1782 if (addr != (pfn << PAGE_SHIFT))
1783 return -EINVAL;
1784
1785 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1786 return 0;
1787 }
1788 EXPORT_SYMBOL(remap_pfn_range);
1789
vm_iomap_memory(struct vm_area_struct * vma,phys_addr_t start,unsigned long len)1790 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1791 {
1792 unsigned long pfn = start >> PAGE_SHIFT;
1793 unsigned long vm_len = vma->vm_end - vma->vm_start;
1794
1795 pfn += vma->vm_pgoff;
1796 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1797 }
1798 EXPORT_SYMBOL(vm_iomap_memory);
1799
remap_vmalloc_range(struct vm_area_struct * vma,void * addr,unsigned long pgoff)1800 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1801 unsigned long pgoff)
1802 {
1803 unsigned int size = vma->vm_end - vma->vm_start;
1804
1805 if (!(vma->vm_flags & VM_USERMAP))
1806 return -EINVAL;
1807
1808 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1809 vma->vm_end = vma->vm_start + size;
1810
1811 return 0;
1812 }
1813 EXPORT_SYMBOL(remap_vmalloc_range);
1814
arch_get_unmapped_area(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)1815 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1816 unsigned long len, unsigned long pgoff, unsigned long flags)
1817 {
1818 return -ENOMEM;
1819 }
1820
unmap_mapping_range(struct address_space * mapping,loff_t const holebegin,loff_t const holelen,int even_cows)1821 void unmap_mapping_range(struct address_space *mapping,
1822 loff_t const holebegin, loff_t const holelen,
1823 int even_cows)
1824 {
1825 }
1826 EXPORT_SYMBOL(unmap_mapping_range);
1827
1828 /*
1829 * Check that a process has enough memory to allocate a new virtual
1830 * mapping. 0 means there is enough memory for the allocation to
1831 * succeed and -ENOMEM implies there is not.
1832 *
1833 * We currently support three overcommit policies, which are set via the
1834 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1835 *
1836 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1837 * Additional code 2002 Jul 20 by Robert Love.
1838 *
1839 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1840 *
1841 * Note this is a helper function intended to be used by LSMs which
1842 * wish to use this logic.
1843 */
__vm_enough_memory(struct mm_struct * mm,long pages,int cap_sys_admin)1844 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1845 {
1846 long free, allowed, reserve;
1847
1848 vm_acct_memory(pages);
1849
1850 /*
1851 * Sometimes we want to use more memory than we have
1852 */
1853 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1854 return 0;
1855
1856 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1857 free = global_page_state(NR_FREE_PAGES);
1858 free += global_page_state(NR_FILE_PAGES);
1859
1860 /*
1861 * shmem pages shouldn't be counted as free in this
1862 * case, they can't be purged, only swapped out, and
1863 * that won't affect the overall amount of available
1864 * memory in the system.
1865 */
1866 free -= global_page_state(NR_SHMEM);
1867
1868 free += get_nr_swap_pages();
1869
1870 /*
1871 * Any slabs which are created with the
1872 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1873 * which are reclaimable, under pressure. The dentry
1874 * cache and most inode caches should fall into this
1875 */
1876 free += global_page_state(NR_SLAB_RECLAIMABLE);
1877
1878 /*
1879 * Leave reserved pages. The pages are not for anonymous pages.
1880 */
1881 if (free <= totalreserve_pages)
1882 goto error;
1883 else
1884 free -= totalreserve_pages;
1885
1886 /*
1887 * Reserve some for root
1888 */
1889 if (!cap_sys_admin)
1890 free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1891
1892 if (free > pages)
1893 return 0;
1894
1895 goto error;
1896 }
1897
1898 allowed = vm_commit_limit();
1899 /*
1900 * Reserve some 3% for root
1901 */
1902 if (!cap_sys_admin)
1903 allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1904
1905 /*
1906 * Don't let a single process grow so big a user can't recover
1907 */
1908 if (mm) {
1909 reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1910 allowed -= min_t(long, mm->total_vm / 32, reserve);
1911 }
1912
1913 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1914 return 0;
1915
1916 error:
1917 vm_unacct_memory(pages);
1918
1919 return -ENOMEM;
1920 }
1921
filemap_fault(struct vm_area_struct * vma,struct vm_fault * vmf)1922 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1923 {
1924 BUG();
1925 return 0;
1926 }
1927 EXPORT_SYMBOL(filemap_fault);
1928
filemap_map_pages(struct vm_area_struct * vma,struct vm_fault * vmf)1929 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1930 {
1931 BUG();
1932 }
1933 EXPORT_SYMBOL(filemap_map_pages);
1934
__access_remote_vm(struct task_struct * tsk,struct mm_struct * mm,unsigned long addr,void * buf,int len,unsigned int gup_flags)1935 static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1936 unsigned long addr, void *buf, int len, unsigned int gup_flags)
1937 {
1938 struct vm_area_struct *vma;
1939 int write = gup_flags & FOLL_WRITE;
1940
1941 down_read(&mm->mmap_sem);
1942
1943 /* the access must start within one of the target process's mappings */
1944 vma = find_vma(mm, addr);
1945 if (vma) {
1946 /* don't overrun this mapping */
1947 if (addr + len >= vma->vm_end)
1948 len = vma->vm_end - addr;
1949
1950 /* only read or write mappings where it is permitted */
1951 if (write && vma->vm_flags & VM_MAYWRITE)
1952 copy_to_user_page(vma, NULL, addr,
1953 (void *) addr, buf, len);
1954 else if (!write && vma->vm_flags & VM_MAYREAD)
1955 copy_from_user_page(vma, NULL, addr,
1956 buf, (void *) addr, len);
1957 else
1958 len = 0;
1959 } else {
1960 len = 0;
1961 }
1962
1963 up_read(&mm->mmap_sem);
1964
1965 return len;
1966 }
1967
1968 /**
1969 * @access_remote_vm - access another process' address space
1970 * @mm: the mm_struct of the target address space
1971 * @addr: start address to access
1972 * @buf: source or destination buffer
1973 * @len: number of bytes to transfer
1974 * @gup_flags: flags modifying lookup behaviour
1975 *
1976 * The caller must hold a reference on @mm.
1977 */
access_remote_vm(struct mm_struct * mm,unsigned long addr,void * buf,int len,unsigned int gup_flags)1978 int access_remote_vm(struct mm_struct *mm, unsigned long addr,
1979 void *buf, int len, unsigned int gup_flags)
1980 {
1981 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags);
1982 }
1983
1984 /*
1985 * Access another process' address space.
1986 * - source/target buffer must be kernel space
1987 */
access_process_vm(struct task_struct * tsk,unsigned long addr,void * buf,int len,int write)1988 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
1989 {
1990 struct mm_struct *mm;
1991
1992 if (addr + len < addr)
1993 return 0;
1994
1995 mm = get_task_mm(tsk);
1996 if (!mm)
1997 return 0;
1998
1999 len = __access_remote_vm(tsk, mm, addr, buf, len,
2000 write ? FOLL_WRITE : 0);
2001
2002 mmput(mm);
2003 return len;
2004 }
2005
2006 /**
2007 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2008 * @inode: The inode to check
2009 * @size: The current filesize of the inode
2010 * @newsize: The proposed filesize of the inode
2011 *
2012 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2013 * make sure that that any outstanding VMAs aren't broken and then shrink the
2014 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2015 * automatically grant mappings that are too large.
2016 */
nommu_shrink_inode_mappings(struct inode * inode,size_t size,size_t newsize)2017 int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2018 size_t newsize)
2019 {
2020 struct vm_area_struct *vma;
2021 struct vm_region *region;
2022 pgoff_t low, high;
2023 size_t r_size, r_top;
2024
2025 low = newsize >> PAGE_SHIFT;
2026 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2027
2028 down_write(&nommu_region_sem);
2029 i_mmap_lock_read(inode->i_mapping);
2030
2031 /* search for VMAs that fall within the dead zone */
2032 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2033 /* found one - only interested if it's shared out of the page
2034 * cache */
2035 if (vma->vm_flags & VM_SHARED) {
2036 i_mmap_unlock_read(inode->i_mapping);
2037 up_write(&nommu_region_sem);
2038 return -ETXTBSY; /* not quite true, but near enough */
2039 }
2040 }
2041
2042 /* reduce any regions that overlap the dead zone - if in existence,
2043 * these will be pointed to by VMAs that don't overlap the dead zone
2044 *
2045 * we don't check for any regions that start beyond the EOF as there
2046 * shouldn't be any
2047 */
2048 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) {
2049 if (!(vma->vm_flags & VM_SHARED))
2050 continue;
2051
2052 region = vma->vm_region;
2053 r_size = region->vm_top - region->vm_start;
2054 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2055
2056 if (r_top > newsize) {
2057 region->vm_top -= r_top - newsize;
2058 if (region->vm_end > region->vm_top)
2059 region->vm_end = region->vm_top;
2060 }
2061 }
2062
2063 i_mmap_unlock_read(inode->i_mapping);
2064 up_write(&nommu_region_sem);
2065 return 0;
2066 }
2067
2068 /*
2069 * Initialise sysctl_user_reserve_kbytes.
2070 *
2071 * This is intended to prevent a user from starting a single memory hogging
2072 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2073 * mode.
2074 *
2075 * The default value is min(3% of free memory, 128MB)
2076 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2077 */
init_user_reserve(void)2078 static int __meminit init_user_reserve(void)
2079 {
2080 unsigned long free_kbytes;
2081
2082 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2083
2084 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2085 return 0;
2086 }
2087 subsys_initcall(init_user_reserve);
2088
2089 /*
2090 * Initialise sysctl_admin_reserve_kbytes.
2091 *
2092 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2093 * to log in and kill a memory hogging process.
2094 *
2095 * Systems with more than 256MB will reserve 8MB, enough to recover
2096 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2097 * only reserve 3% of free pages by default.
2098 */
init_admin_reserve(void)2099 static int __meminit init_admin_reserve(void)
2100 {
2101 unsigned long free_kbytes;
2102
2103 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2104
2105 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2106 return 0;
2107 }
2108 subsys_initcall(init_admin_reserve);
2109