Root/mm/nommu.c

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

Archive Download this file



interactive