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) 2007-2010 Paul Mundt <lethal@linux-sh.org>
14 */
15
16#include <linux/export.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/blkdev.h>
26#include <linux/backing-dev.h>
27#include <linux/mount.h>
28#include <linux/personality.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
31#include <linux/audit.h>
32#include <linux/sched/sysctl.h>
33
34#include <asm/uaccess.h>
35#include <asm/tlb.h>
36#include <asm/tlbflush.h>
37#include <asm/mmu_context.h>
38#include "internal.h"
39
40#if 0
41#define kenter(FMT, ...) \
42    printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
43#define kleave(FMT, ...) \
44    printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
45#define kdebug(FMT, ...) \
46    printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
47#else
48#define kenter(FMT, ...) \
49    no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
50#define kleave(FMT, ...) \
51    no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
52#define kdebug(FMT, ...) \
53    no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
54#endif
55
56void *high_memory;
57struct page *mem_map;
58unsigned long max_mapnr;
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;
65unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
66unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
67int heap_stack_gap = 0;
68
69atomic_long_t mmap_pages_allocated;
70
71/*
72 * The global memory commitment made in the system can be a metric
73 * that can be used to drive ballooning decisions when Linux is hosted
74 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
75 * balancing memory across competing virtual machines that are hosted.
76 * Several metrics drive this policy engine including the guest reported
77 * memory commitment.
78 */
79unsigned long vm_memory_committed(void)
80{
81    return percpu_counter_read_positive(&vm_committed_as);
82}
83
84EXPORT_SYMBOL_GPL(vm_memory_committed);
85
86EXPORT_SYMBOL(mem_map);
87
88/* list of mapped, potentially shareable regions */
89static struct kmem_cache *vm_region_jar;
90struct rb_root nommu_region_tree = RB_ROOT;
91DECLARE_RWSEM(nommu_region_sem);
92
93const struct vm_operations_struct generic_file_vm_ops = {
94};
95
96/*
97 * Return the total memory allocated for this pointer, not
98 * just what the caller asked for.
99 *
100 * Doesn't have to be accurate, i.e. may have races.
101 */
102unsigned int kobjsize(const void *objp)
103{
104    struct page *page;
105
106    /*
107     * If the object we have should not have ksize performed on it,
108     * return size of 0
109     */
110    if (!objp || !virt_addr_valid(objp))
111        return 0;
112
113    page = virt_to_head_page(objp);
114
115    /*
116     * If the allocator sets PageSlab, we know the pointer came from
117     * kmalloc().
118     */
119    if (PageSlab(page))
120        return ksize(objp);
121
122    /*
123     * If it's not a compound page, see if we have a matching VMA
124     * region. This test is intentionally done in reverse order,
125     * so if there's no VMA, we still fall through and hand back
126     * PAGE_SIZE for 0-order pages.
127     */
128    if (!PageCompound(page)) {
129        struct vm_area_struct *vma;
130
131        vma = find_vma(current->mm, (unsigned long)objp);
132        if (vma)
133            return vma->vm_end - vma->vm_start;
134    }
135
136    /*
137     * The ksize() function is only guaranteed to work for pointers
138     * returned by kmalloc(). So handle arbitrary pointers here.
139     */
140    return PAGE_SIZE << compound_order(page);
141}
142
143long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
144              unsigned long start, unsigned long nr_pages,
145              unsigned int foll_flags, struct page **pages,
146              struct vm_area_struct **vmas, int *nonblocking)
147{
148    struct vm_area_struct *vma;
149    unsigned long vm_flags;
150    int i;
151
152    /* calculate required read or write permissions.
153     * If FOLL_FORCE is set, we only require the "MAY" flags.
154     */
155    vm_flags = (foll_flags & FOLL_WRITE) ?
156            (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
157    vm_flags &= (foll_flags & FOLL_FORCE) ?
158            (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
159
160    for (i = 0; i < nr_pages; i++) {
161        vma = find_vma(mm, start);
162        if (!vma)
163            goto finish_or_fault;
164
165        /* protect what we can, including chardevs */
166        if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
167            !(vm_flags & vma->vm_flags))
168            goto finish_or_fault;
169
170        if (pages) {
171            pages[i] = virt_to_page(start);
172            if (pages[i])
173                page_cache_get(pages[i]);
174        }
175        if (vmas)
176            vmas[i] = vma;
177        start = (start + PAGE_SIZE) & PAGE_MASK;
178    }
179
180    return i;
181
182finish_or_fault:
183    return i ? : -EFAULT;
184}
185
186/*
187 * get a list of pages in an address range belonging to the specified process
188 * and indicate the VMA that covers each page
189 * - this is potentially dodgy as we may end incrementing the page count of a
190 * slab page or a secondary page from a compound page
191 * - don't permit access to VMAs that don't support it, such as I/O mappings
192 */
193long get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
194            unsigned long start, unsigned long nr_pages,
195            int write, int force, struct page **pages,
196            struct vm_area_struct **vmas)
197{
198    int flags = 0;
199
200    if (write)
201        flags |= FOLL_WRITE;
202    if (force)
203        flags |= FOLL_FORCE;
204
205    return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
206                NULL);
207}
208EXPORT_SYMBOL(get_user_pages);
209
210/**
211 * follow_pfn - look up PFN at a user virtual address
212 * @vma: memory mapping
213 * @address: user virtual address
214 * @pfn: location to store found PFN
215 *
216 * Only IO mappings and raw PFN mappings are allowed.
217 *
218 * Returns zero and the pfn at @pfn on success, -ve otherwise.
219 */
220int follow_pfn(struct vm_area_struct *vma, unsigned long address,
221    unsigned long *pfn)
222{
223    if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
224        return -EINVAL;
225
226    *pfn = address >> PAGE_SHIFT;
227    return 0;
228}
229EXPORT_SYMBOL(follow_pfn);
230
231LIST_HEAD(vmap_area_list);
232
233void vfree(const void *addr)
234{
235    kfree(addr);
236}
237EXPORT_SYMBOL(vfree);
238
239void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
240{
241    /*
242     * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
243     * returns only a logical address.
244     */
245    return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM);
246}
247EXPORT_SYMBOL(__vmalloc);
248
249void *vmalloc_user(unsigned long size)
250{
251    void *ret;
252
253    ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
254            PAGE_KERNEL);
255    if (ret) {
256        struct vm_area_struct *vma;
257
258        down_write(&current->mm->mmap_sem);
259        vma = find_vma(current->mm, (unsigned long)ret);
260        if (vma)
261            vma->vm_flags |= VM_USERMAP;
262        up_write(&current->mm->mmap_sem);
263    }
264
265    return ret;
266}
267EXPORT_SYMBOL(vmalloc_user);
268
269struct page *vmalloc_to_page(const void *addr)
270{
271    return virt_to_page(addr);
272}
273EXPORT_SYMBOL(vmalloc_to_page);
274
275unsigned long vmalloc_to_pfn(const void *addr)
276{
277    return page_to_pfn(virt_to_page(addr));
278}
279EXPORT_SYMBOL(vmalloc_to_pfn);
280
281long vread(char *buf, char *addr, unsigned long count)
282{
283    /* Don't allow overflow */
284    if ((unsigned long) buf + count < count)
285        count = -(unsigned long) buf;
286
287    memcpy(buf, addr, count);
288    return count;
289}
290
291long vwrite(char *buf, char *addr, unsigned long count)
292{
293    /* Don't allow overflow */
294    if ((unsigned long) addr + count < count)
295        count = -(unsigned long) addr;
296
297    memcpy(addr, buf, count);
298    return(count);
299}
300
301/*
302 * vmalloc - allocate virtually continguos memory
303 *
304 * @size: allocation size
305 *
306 * Allocate enough pages to cover @size from the page level
307 * allocator and map them into continguos kernel virtual space.
308 *
309 * For tight control over page level allocator and protection flags
310 * use __vmalloc() instead.
311 */
312void *vmalloc(unsigned long size)
313{
314       return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
315}
316EXPORT_SYMBOL(vmalloc);
317
318/*
319 * vzalloc - allocate virtually continguos memory with zero fill
320 *
321 * @size: allocation size
322 *
323 * Allocate enough pages to cover @size from the page level
324 * allocator and map them into continguos kernel virtual space.
325 * The memory allocated is set to zero.
326 *
327 * For tight control over page level allocator and protection flags
328 * use __vmalloc() instead.
329 */
330void *vzalloc(unsigned long size)
331{
332    return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
333            PAGE_KERNEL);
334}
335EXPORT_SYMBOL(vzalloc);
336
337/**
338 * vmalloc_node - allocate memory on a specific node
339 * @size: allocation size
340 * @node: numa node
341 *
342 * Allocate enough pages to cover @size from the page level
343 * allocator and map them into contiguous kernel virtual space.
344 *
345 * For tight control over page level allocator and protection flags
346 * use __vmalloc() instead.
347 */
348void *vmalloc_node(unsigned long size, int node)
349{
350    return vmalloc(size);
351}
352EXPORT_SYMBOL(vmalloc_node);
353
354/**
355 * vzalloc_node - allocate memory on a specific node with zero fill
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 * The memory allocated is set to zero.
362 *
363 * For tight control over page level allocator and protection flags
364 * use __vmalloc() instead.
365 */
366void *vzalloc_node(unsigned long size, int node)
367{
368    return vzalloc(size);
369}
370EXPORT_SYMBOL(vzalloc_node);
371
372#ifndef PAGE_KERNEL_EXEC
373# define PAGE_KERNEL_EXEC PAGE_KERNEL
374#endif
375
376/**
377 * vmalloc_exec - allocate virtually contiguous, executable memory
378 * @size: allocation size
379 *
380 * Kernel-internal function to allocate enough pages to cover @size
381 * the page level allocator and map them into contiguous and
382 * executable kernel virtual space.
383 *
384 * For tight control over page level allocator and protection flags
385 * use __vmalloc() instead.
386 */
387
388void *vmalloc_exec(unsigned long size)
389{
390    return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
391}
392
393/**
394 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
395 * @size: allocation size
396 *
397 * Allocate enough 32bit PA addressable pages to cover @size from the
398 * page level allocator and map them into continguos kernel virtual space.
399 */
400void *vmalloc_32(unsigned long size)
401{
402    return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
403}
404EXPORT_SYMBOL(vmalloc_32);
405
406/**
407 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
408 * @size: allocation size
409 *
410 * The resulting memory area is 32bit addressable and zeroed so it can be
411 * mapped to userspace without leaking data.
412 *
413 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
414 * remap_vmalloc_range() are permissible.
415 */
416void *vmalloc_32_user(unsigned long size)
417{
418    /*
419     * We'll have to sort out the ZONE_DMA bits for 64-bit,
420     * but for now this can simply use vmalloc_user() directly.
421     */
422    return vmalloc_user(size);
423}
424EXPORT_SYMBOL(vmalloc_32_user);
425
426void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
427{
428    BUG();
429    return NULL;
430}
431EXPORT_SYMBOL(vmap);
432
433void vunmap(const void *addr)
434{
435    BUG();
436}
437EXPORT_SYMBOL(vunmap);
438
439void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
440{
441    BUG();
442    return NULL;
443}
444EXPORT_SYMBOL(vm_map_ram);
445
446void vm_unmap_ram(const void *mem, unsigned int count)
447{
448    BUG();
449}
450EXPORT_SYMBOL(vm_unmap_ram);
451
452void vm_unmap_aliases(void)
453{
454}
455EXPORT_SYMBOL_GPL(vm_unmap_aliases);
456
457/*
458 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
459 * have one.
460 */
461void __attribute__((weak)) vmalloc_sync_all(void)
462{
463}
464
465/**
466 * alloc_vm_area - allocate a range of kernel address space
467 * @size: size of the area
468 *
469 * Returns: NULL on failure, vm_struct on success
470 *
471 * This function reserves a range of kernel address space, and
472 * allocates pagetables to map that range. No actual mappings
473 * are created. If the kernel address space is not shared
474 * between processes, it syncs the pagetable across all
475 * processes.
476 */
477struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
478{
479    BUG();
480    return NULL;
481}
482EXPORT_SYMBOL_GPL(alloc_vm_area);
483
484void free_vm_area(struct vm_struct *area)
485{
486    BUG();
487}
488EXPORT_SYMBOL_GPL(free_vm_area);
489
490int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
491           struct page *page)
492{
493    return -EINVAL;
494}
495EXPORT_SYMBOL(vm_insert_page);
496
497/*
498 * sys_brk() for the most part doesn't need the global kernel
499 * lock, except when an application is doing something nasty
500 * like trying to un-brk an area that has already been mapped
501 * to a regular file. in this case, the unmapping will need
502 * to invoke file system routines that need the global lock.
503 */
504SYSCALL_DEFINE1(brk, unsigned long, brk)
505{
506    struct mm_struct *mm = current->mm;
507
508    if (brk < mm->start_brk || brk > mm->context.end_brk)
509        return mm->brk;
510
511    if (mm->brk == brk)
512        return mm->brk;
513
514    /*
515     * Always allow shrinking brk
516     */
517    if (brk <= mm->brk) {
518        mm->brk = brk;
519        return brk;
520    }
521
522    /*
523     * Ok, looks good - let it rip.
524     */
525    flush_icache_range(mm->brk, brk);
526    return mm->brk = brk;
527}
528
529/*
530 * initialise the VMA and region record slabs
531 */
532void __init mmap_init(void)
533{
534    int ret;
535
536    ret = percpu_counter_init(&vm_committed_as, 0);
537    VM_BUG_ON(ret);
538    vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
539}
540
541/*
542 * validate the region tree
543 * - the caller must hold the region lock
544 */
545#ifdef CONFIG_DEBUG_NOMMU_REGIONS
546static noinline void validate_nommu_regions(void)
547{
548    struct vm_region *region, *last;
549    struct rb_node *p, *lastp;
550
551    lastp = rb_first(&nommu_region_tree);
552    if (!lastp)
553        return;
554
555    last = rb_entry(lastp, struct vm_region, vm_rb);
556    BUG_ON(unlikely(last->vm_end <= last->vm_start));
557    BUG_ON(unlikely(last->vm_top < last->vm_end));
558
559    while ((p = rb_next(lastp))) {
560        region = rb_entry(p, struct vm_region, vm_rb);
561        last = rb_entry(lastp, struct vm_region, vm_rb);
562
563        BUG_ON(unlikely(region->vm_end <= region->vm_start));
564        BUG_ON(unlikely(region->vm_top < region->vm_end));
565        BUG_ON(unlikely(region->vm_start < last->vm_top));
566
567        lastp = p;
568    }
569}
570#else
571static void validate_nommu_regions(void)
572{
573}
574#endif
575
576/*
577 * add a region into the global tree
578 */
579static void add_nommu_region(struct vm_region *region)
580{
581    struct vm_region *pregion;
582    struct rb_node **p, *parent;
583
584    validate_nommu_regions();
585
586    parent = NULL;
587    p = &nommu_region_tree.rb_node;
588    while (*p) {
589        parent = *p;
590        pregion = rb_entry(parent, struct vm_region, vm_rb);
591        if (region->vm_start < pregion->vm_start)
592            p = &(*p)->rb_left;
593        else if (region->vm_start > pregion->vm_start)
594            p = &(*p)->rb_right;
595        else if (pregion == region)
596            return;
597        else
598            BUG();
599    }
600
601    rb_link_node(&region->vm_rb, parent, p);
602    rb_insert_color(&region->vm_rb, &nommu_region_tree);
603
604    validate_nommu_regions();
605}
606
607/*
608 * delete a region from the global tree
609 */
610static void delete_nommu_region(struct vm_region *region)
611{
612    BUG_ON(!nommu_region_tree.rb_node);
613
614    validate_nommu_regions();
615    rb_erase(&region->vm_rb, &nommu_region_tree);
616    validate_nommu_regions();
617}
618
619/*
620 * free a contiguous series of pages
621 */
622static void free_page_series(unsigned long from, unsigned long to)
623{
624    for (; from < to; from += PAGE_SIZE) {
625        struct page *page = virt_to_page(from);
626
627        kdebug("- free %lx", from);
628        atomic_long_dec(&mmap_pages_allocated);
629        if (page_count(page) != 1)
630            kdebug("free page %p: refcount not one: %d",
631                   page, page_count(page));
632        put_page(page);
633    }
634}
635
636/*
637 * release a reference to a region
638 * - the caller must hold the region semaphore for writing, which this releases
639 * - the region may not have been added to the tree yet, in which case vm_top
640 * will equal vm_start
641 */
642static void __put_nommu_region(struct vm_region *region)
643    __releases(nommu_region_sem)
644{
645    kenter("%p{%d}", region, region->vm_usage);
646
647    BUG_ON(!nommu_region_tree.rb_node);
648
649    if (--region->vm_usage == 0) {
650        if (region->vm_top > region->vm_start)
651            delete_nommu_region(region);
652        up_write(&nommu_region_sem);
653
654        if (region->vm_file)
655            fput(region->vm_file);
656
657        /* IO memory and memory shared directly out of the pagecache
658         * from ramfs/tmpfs mustn't be released here */
659        if (region->vm_flags & VM_MAPPED_COPY) {
660            kdebug("free series");
661            free_page_series(region->vm_start, region->vm_top);
662        }
663        kmem_cache_free(vm_region_jar, region);
664    } else {
665        up_write(&nommu_region_sem);
666    }
667}
668
669/*
670 * release a reference to a region
671 */
672static void put_nommu_region(struct vm_region *region)
673{
674    down_write(&nommu_region_sem);
675    __put_nommu_region(region);
676}
677
678/*
679 * update protection on a vma
680 */
681static void protect_vma(struct vm_area_struct *vma, unsigned long flags)
682{
683#ifdef CONFIG_MPU
684    struct mm_struct *mm = vma->vm_mm;
685    long start = vma->vm_start & PAGE_MASK;
686    while (start < vma->vm_end) {
687        protect_page(mm, start, flags);
688        start += PAGE_SIZE;
689    }
690    update_protections(mm);
691#endif
692}
693
694/*
695 * add a VMA into a process's mm_struct in the appropriate place in the list
696 * and tree and add to the address space's page tree also if not an anonymous
697 * page
698 * - should be called with mm->mmap_sem held writelocked
699 */
700static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma)
701{
702    struct vm_area_struct *pvma, *prev;
703    struct address_space *mapping;
704    struct rb_node **p, *parent, *rb_prev;
705
706    kenter(",%p", vma);
707
708    BUG_ON(!vma->vm_region);
709
710    mm->map_count++;
711    vma->vm_mm = mm;
712
713    protect_vma(vma, vma->vm_flags);
714
715    /* add the VMA to the mapping */
716    if (vma->vm_file) {
717        mapping = vma->vm_file->f_mapping;
718
719        mutex_lock(&mapping->i_mmap_mutex);
720        flush_dcache_mmap_lock(mapping);
721        vma_interval_tree_insert(vma, &mapping->i_mmap);
722        flush_dcache_mmap_unlock(mapping);
723        mutex_unlock(&mapping->i_mmap_mutex);
724    }
725
726    /* add the VMA to the tree */
727    parent = rb_prev = NULL;
728    p = &mm->mm_rb.rb_node;
729    while (*p) {
730        parent = *p;
731        pvma = rb_entry(parent, struct vm_area_struct, vm_rb);
732
733        /* sort by: start addr, end addr, VMA struct addr in that order
734         * (the latter is necessary as we may get identical VMAs) */
735        if (vma->vm_start < pvma->vm_start)
736            p = &(*p)->rb_left;
737        else if (vma->vm_start > pvma->vm_start) {
738            rb_prev = parent;
739            p = &(*p)->rb_right;
740        } else if (vma->vm_end < pvma->vm_end)
741            p = &(*p)->rb_left;
742        else if (vma->vm_end > pvma->vm_end) {
743            rb_prev = parent;
744            p = &(*p)->rb_right;
745        } else if (vma < pvma)
746            p = &(*p)->rb_left;
747        else if (vma > pvma) {
748            rb_prev = parent;
749            p = &(*p)->rb_right;
750        } else
751            BUG();
752    }
753
754    rb_link_node(&vma->vm_rb, parent, p);
755    rb_insert_color(&vma->vm_rb, &mm->mm_rb);
756
757    /* add VMA to the VMA list also */
758    prev = NULL;
759    if (rb_prev)
760        prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
761
762    __vma_link_list(mm, vma, prev, parent);
763}
764
765/*
766 * delete a VMA from its owning mm_struct and address space
767 */
768static void delete_vma_from_mm(struct vm_area_struct *vma)
769{
770    struct address_space *mapping;
771    struct mm_struct *mm = vma->vm_mm;
772
773    kenter("%p", vma);
774
775    protect_vma(vma, 0);
776
777    mm->map_count--;
778    if (mm->mmap_cache == vma)
779        mm->mmap_cache = NULL;
780
781    /* remove the VMA from the mapping */
782    if (vma->vm_file) {
783        mapping = vma->vm_file->f_mapping;
784
785        mutex_lock(&mapping->i_mmap_mutex);
786        flush_dcache_mmap_lock(mapping);
787        vma_interval_tree_remove(vma, &mapping->i_mmap);
788        flush_dcache_mmap_unlock(mapping);
789        mutex_unlock(&mapping->i_mmap_mutex);
790    }
791
792    /* remove from the MM's tree and list */
793    rb_erase(&vma->vm_rb, &mm->mm_rb);
794
795    if (vma->vm_prev)
796        vma->vm_prev->vm_next = vma->vm_next;
797    else
798        mm->mmap = vma->vm_next;
799
800    if (vma->vm_next)
801        vma->vm_next->vm_prev = vma->vm_prev;
802}
803
804/*
805 * destroy a VMA record
806 */
807static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
808{
809    kenter("%p", vma);
810    if (vma->vm_ops && vma->vm_ops->close)
811        vma->vm_ops->close(vma);
812    if (vma->vm_file)
813        fput(vma->vm_file);
814    put_nommu_region(vma->vm_region);
815    kmem_cache_free(vm_area_cachep, vma);
816}
817
818/*
819 * look up the first VMA in which addr resides, NULL if none
820 * - should be called with mm->mmap_sem at least held readlocked
821 */
822struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
823{
824    struct vm_area_struct *vma;
825
826    /* check the cache first */
827    vma = ACCESS_ONCE(mm->mmap_cache);
828    if (vma && vma->vm_start <= addr && vma->vm_end > addr)
829        return vma;
830
831    /* trawl the list (there may be multiple mappings in which addr
832     * resides) */
833    for (vma = mm->mmap; vma; vma = vma->vm_next) {
834        if (vma->vm_start > addr)
835            return NULL;
836        if (vma->vm_end > addr) {
837            mm->mmap_cache = vma;
838            return vma;
839        }
840    }
841
842    return NULL;
843}
844EXPORT_SYMBOL(find_vma);
845
846/*
847 * find a VMA
848 * - we don't extend stack VMAs under NOMMU conditions
849 */
850struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
851{
852    return find_vma(mm, addr);
853}
854
855/*
856 * expand a stack to a given address
857 * - not supported under NOMMU conditions
858 */
859int expand_stack(struct vm_area_struct *vma, unsigned long address)
860{
861    return -ENOMEM;
862}
863
864/*
865 * look up the first VMA exactly that exactly matches addr
866 * - should be called with mm->mmap_sem at least held readlocked
867 */
868static struct vm_area_struct *find_vma_exact(struct mm_struct *mm,
869                         unsigned long addr,
870                         unsigned long len)
871{
872    struct vm_area_struct *vma;
873    unsigned long end = addr + len;
874
875    /* check the cache first */
876    vma = mm->mmap_cache;
877    if (vma && vma->vm_start == addr && vma->vm_end == end)
878        return vma;
879
880    /* trawl the list (there may be multiple mappings in which addr
881     * resides) */
882    for (vma = mm->mmap; vma; vma = vma->vm_next) {
883        if (vma->vm_start < addr)
884            continue;
885        if (vma->vm_start > addr)
886            return NULL;
887        if (vma->vm_end == end) {
888            mm->mmap_cache = vma;
889            return vma;
890        }
891    }
892
893    return NULL;
894}
895
896/*
897 * determine whether a mapping should be permitted and, if so, what sort of
898 * mapping we're capable of supporting
899 */
900static int validate_mmap_request(struct file *file,
901                 unsigned long addr,
902                 unsigned long len,
903                 unsigned long prot,
904                 unsigned long flags,
905                 unsigned long pgoff,
906                 unsigned long *_capabilities)
907{
908    unsigned long capabilities, rlen;
909    int ret;
910
911    /* do the simple checks first */
912    if (flags & MAP_FIXED) {
913        printk(KERN_DEBUG
914               "%d: Can't do fixed-address/overlay mmap of RAM\n",
915               current->pid);
916        return -EINVAL;
917    }
918
919    if ((flags & MAP_TYPE) != MAP_PRIVATE &&
920        (flags & MAP_TYPE) != MAP_SHARED)
921        return -EINVAL;
922
923    if (!len)
924        return -EINVAL;
925
926    /* Careful about overflows.. */
927    rlen = PAGE_ALIGN(len);
928    if (!rlen || rlen > TASK_SIZE)
929        return -ENOMEM;
930
931    /* offset overflow? */
932    if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff)
933        return -EOVERFLOW;
934
935    if (file) {
936        /* validate file mapping requests */
937        struct address_space *mapping;
938
939        /* files must support mmap */
940        if (!file->f_op || !file->f_op->mmap)
941            return -ENODEV;
942
943        /* work out if what we've got could possibly be shared
944         * - we support chardevs that provide their own "memory"
945         * - we support files/blockdevs that are memory backed
946         */
947        mapping = file->f_mapping;
948        if (!mapping)
949            mapping = file_inode(file)->i_mapping;
950
951        capabilities = 0;
952        if (mapping && mapping->backing_dev_info)
953            capabilities = mapping->backing_dev_info->capabilities;
954
955        if (!capabilities) {
956            /* no explicit capabilities set, so assume some
957             * defaults */
958            switch (file_inode(file)->i_mode & S_IFMT) {
959            case S_IFREG:
960            case S_IFBLK:
961                capabilities = BDI_CAP_MAP_COPY;
962                break;
963
964            case S_IFCHR:
965                capabilities =
966                    BDI_CAP_MAP_DIRECT |
967                    BDI_CAP_READ_MAP |
968                    BDI_CAP_WRITE_MAP;
969                break;
970
971            default:
972                return -EINVAL;
973            }
974        }
975
976        /* eliminate any capabilities that we can't support on this
977         * device */
978        if (!file->f_op->get_unmapped_area)
979            capabilities &= ~BDI_CAP_MAP_DIRECT;
980        if (!file->f_op->read)
981            capabilities &= ~BDI_CAP_MAP_COPY;
982
983        /* The file shall have been opened with read permission. */
984        if (!(file->f_mode & FMODE_READ))
985            return -EACCES;
986
987        if (flags & MAP_SHARED) {
988            /* do checks for writing, appending and locking */
989            if ((prot & PROT_WRITE) &&
990                !(file->f_mode & FMODE_WRITE))
991                return -EACCES;
992
993            if (IS_APPEND(file_inode(file)) &&
994                (file->f_mode & FMODE_WRITE))
995                return -EACCES;
996
997            if (locks_verify_locked(file_inode(file)))
998                return -EAGAIN;
999
1000            if (!(capabilities & BDI_CAP_MAP_DIRECT))
1001                return -ENODEV;
1002
1003            /* we mustn't privatise shared mappings */
1004            capabilities &= ~BDI_CAP_MAP_COPY;
1005        }
1006        else {
1007            /* we're going to read the file into private memory we
1008             * allocate */
1009            if (!(capabilities & BDI_CAP_MAP_COPY))
1010                return -ENODEV;
1011
1012            /* we don't permit a private writable mapping to be
1013             * shared with the backing device */
1014            if (prot & PROT_WRITE)
1015                capabilities &= ~BDI_CAP_MAP_DIRECT;
1016        }
1017
1018        if (capabilities & BDI_CAP_MAP_DIRECT) {
1019            if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) ||
1020                ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) ||
1021                ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP))
1022                ) {
1023                capabilities &= ~BDI_CAP_MAP_DIRECT;
1024                if (flags & MAP_SHARED) {
1025                    printk(KERN_WARNING
1026                           "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 (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1035            if (prot & PROT_EXEC)
1036                return -EPERM;
1037        }
1038        else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) {
1039            /* handle implication of PROT_EXEC by PROT_READ */
1040            if (current->personality & READ_IMPLIES_EXEC) {
1041                if (capabilities & BDI_CAP_EXEC_MAP)
1042                    prot |= PROT_EXEC;
1043            }
1044        }
1045        else if ((prot & PROT_READ) &&
1046             (prot & PROT_EXEC) &&
1047             !(capabilities & BDI_CAP_EXEC_MAP)
1048             ) {
1049            /* backing file is not executable, try to copy */
1050            capabilities &= ~BDI_CAP_MAP_DIRECT;
1051        }
1052    }
1053    else {
1054        /* anonymous mappings are always memory backed and can be
1055         * privately mapped
1056         */
1057        capabilities = BDI_CAP_MAP_COPY;
1058
1059        /* handle PROT_EXEC implication by PROT_READ */
1060        if ((prot & PROT_READ) &&
1061            (current->personality & READ_IMPLIES_EXEC))
1062            prot |= PROT_EXEC;
1063    }
1064
1065    /* allow the security API to have its say */
1066    ret = security_mmap_addr(addr);
1067    if (ret < 0)
1068        return ret;
1069
1070    /* looks okay */
1071    *_capabilities = capabilities;
1072    return 0;
1073}
1074
1075/*
1076 * we've determined that we can make the mapping, now translate what we
1077 * now know into VMA flags
1078 */
1079static unsigned long determine_vm_flags(struct file *file,
1080                    unsigned long prot,
1081                    unsigned long flags,
1082                    unsigned long capabilities)
1083{
1084    unsigned long vm_flags;
1085
1086    vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags);
1087    /* vm_flags |= mm->def_flags; */
1088
1089    if (!(capabilities & BDI_CAP_MAP_DIRECT)) {
1090        /* attempt to share read-only copies of mapped file chunks */
1091        vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1092        if (file && !(prot & PROT_WRITE))
1093            vm_flags |= VM_MAYSHARE;
1094    } else {
1095        /* overlay a shareable mapping on the backing device or inode
1096         * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1097         * romfs/cramfs */
1098        vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS);
1099        if (flags & MAP_SHARED)
1100            vm_flags |= VM_SHARED;
1101    }
1102
1103    /* refuse to let anyone share private mappings with this process if
1104     * it's being traced - otherwise breakpoints set in it may interfere
1105     * with another untraced process
1106     */
1107    if ((flags & MAP_PRIVATE) && current->ptrace)
1108        vm_flags &= ~VM_MAYSHARE;
1109
1110    return vm_flags;
1111}
1112
1113/*
1114 * set up a shared mapping on a file (the driver or filesystem provides and
1115 * pins the storage)
1116 */
1117static int do_mmap_shared_file(struct vm_area_struct *vma)
1118{
1119    int ret;
1120
1121    ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1122    if (ret == 0) {
1123        vma->vm_region->vm_top = vma->vm_region->vm_end;
1124        return 0;
1125    }
1126    if (ret != -ENOSYS)
1127        return ret;
1128
1129    /* getting -ENOSYS indicates that direct mmap isn't possible (as
1130     * opposed to tried but failed) so we can only give a suitable error as
1131     * it's not possible to make a private copy if MAP_SHARED was given */
1132    return -ENODEV;
1133}
1134
1135/*
1136 * set up a private mapping or an anonymous shared mapping
1137 */
1138static int do_mmap_private(struct vm_area_struct *vma,
1139               struct vm_region *region,
1140               unsigned long len,
1141               unsigned long capabilities)
1142{
1143    struct page *pages;
1144    unsigned long total, point, n;
1145    void *base;
1146    int ret, order;
1147
1148    /* invoke the file's mapping function so that it can keep track of
1149     * shared mappings on devices or memory
1150     * - VM_MAYSHARE will be set if it may attempt to share
1151     */
1152    if (capabilities & BDI_CAP_MAP_DIRECT) {
1153        ret = vma->vm_file->f_op->mmap(vma->vm_file, vma);
1154        if (ret == 0) {
1155            /* shouldn't return success if we're not sharing */
1156            BUG_ON(!(vma->vm_flags & VM_MAYSHARE));
1157            vma->vm_region->vm_top = vma->vm_region->vm_end;
1158            return 0;
1159        }
1160        if (ret != -ENOSYS)
1161            return ret;
1162
1163        /* getting an ENOSYS error indicates that direct mmap isn't
1164         * possible (as opposed to tried but failed) so we'll try to
1165         * make a private copy of the data and map that instead */
1166    }
1167
1168
1169    /* allocate some memory to hold the mapping
1170     * - note that this may not return a page-aligned address if the object
1171     * we're allocating is smaller than a page
1172     */
1173    order = get_order(len);
1174    kdebug("alloc order %d for %lx", order, len);
1175
1176    pages = alloc_pages(GFP_KERNEL, order);
1177    if (!pages)
1178        goto enomem;
1179
1180    total = 1 << order;
1181    atomic_long_add(total, &mmap_pages_allocated);
1182
1183    point = len >> PAGE_SHIFT;
1184
1185    /* we allocated a power-of-2 sized page set, so we may want to trim off
1186     * the excess */
1187    if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) {
1188        while (total > point) {
1189            order = ilog2(total - point);
1190            n = 1 << order;
1191            kdebug("shave %lu/%lu @%lu", n, total - point, total);
1192            atomic_long_sub(n, &mmap_pages_allocated);
1193            total -= n;
1194            set_page_refcounted(pages + total);
1195            __free_pages(pages + total, order);
1196        }
1197    }
1198
1199    for (point = 1; point < total; point++)
1200        set_page_refcounted(&pages[point]);
1201
1202    base = page_address(pages);
1203    region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY;
1204    region->vm_start = (unsigned long) base;
1205    region->vm_end = region->vm_start + len;
1206    region->vm_top = region->vm_start + (total << PAGE_SHIFT);
1207
1208    vma->vm_start = region->vm_start;
1209    vma->vm_end = region->vm_start + len;
1210
1211    if (vma->vm_file) {
1212        /* read the contents of a file into the copy */
1213        mm_segment_t old_fs;
1214        loff_t fpos;
1215
1216        fpos = vma->vm_pgoff;
1217        fpos <<= PAGE_SHIFT;
1218
1219        old_fs = get_fs();
1220        set_fs(KERNEL_DS);
1221        ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos);
1222        set_fs(old_fs);
1223
1224        if (ret < 0)
1225            goto error_free;
1226
1227        /* clear the last little bit */
1228        if (ret < len)
1229            memset(base + ret, 0, len - ret);
1230
1231    }
1232
1233    return 0;
1234
1235error_free:
1236    free_page_series(region->vm_start, region->vm_top);
1237    region->vm_start = vma->vm_start = 0;
1238    region->vm_end = vma->vm_end = 0;
1239    region->vm_top = 0;
1240    return ret;
1241
1242enomem:
1243    printk("Allocation of length %lu from process %d (%s) failed\n",
1244           len, current->pid, current->comm);
1245    show_free_areas(0);
1246    return -ENOMEM;
1247}
1248
1249/*
1250 * handle mapping creation for uClinux
1251 */
1252unsigned long do_mmap_pgoff(struct file *file,
1253                unsigned long addr,
1254                unsigned long len,
1255                unsigned long prot,
1256                unsigned long flags,
1257                unsigned long pgoff,
1258                unsigned long *populate)
1259{
1260    struct vm_area_struct *vma;
1261    struct vm_region *region;
1262    struct rb_node *rb;
1263    unsigned long capabilities, vm_flags, result;
1264    int ret;
1265
1266    kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff);
1267
1268    *populate = 0;
1269
1270    /* decide whether we should attempt the mapping, and if so what sort of
1271     * mapping */
1272    ret = validate_mmap_request(file, addr, len, prot, flags, pgoff,
1273                    &capabilities);
1274    if (ret < 0) {
1275        kleave(" = %d [val]", ret);
1276        return ret;
1277    }
1278
1279    /* we ignore the address hint */
1280    addr = 0;
1281    len = PAGE_ALIGN(len);
1282
1283    /* we've determined that we can make the mapping, now translate what we
1284     * now know into VMA flags */
1285    vm_flags = determine_vm_flags(file, prot, flags, capabilities);
1286
1287    /* we're going to need to record the mapping */
1288    region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL);
1289    if (!region)
1290        goto error_getting_region;
1291
1292    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1293    if (!vma)
1294        goto error_getting_vma;
1295
1296    region->vm_usage = 1;
1297    region->vm_flags = vm_flags;
1298    region->vm_pgoff = pgoff;
1299
1300    INIT_LIST_HEAD(&vma->anon_vma_chain);
1301    vma->vm_flags = vm_flags;
1302    vma->vm_pgoff = pgoff;
1303
1304    if (file) {
1305        region->vm_file = get_file(file);
1306        vma->vm_file = get_file(file);
1307    }
1308
1309    down_write(&nommu_region_sem);
1310
1311    /* if we want to share, we need to check for regions created by other
1312     * mmap() calls that overlap with our proposed mapping
1313     * - we can only share with a superset match on most regular files
1314     * - shared mappings on character devices and memory backed files are
1315     * permitted to overlap inexactly as far as we are concerned for in
1316     * these cases, sharing is handled in the driver or filesystem rather
1317     * than here
1318     */
1319    if (vm_flags & VM_MAYSHARE) {
1320        struct vm_region *pregion;
1321        unsigned long pglen, rpglen, pgend, rpgend, start;
1322
1323        pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1324        pgend = pgoff + pglen;
1325
1326        for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) {
1327            pregion = rb_entry(rb, struct vm_region, vm_rb);
1328
1329            if (!(pregion->vm_flags & VM_MAYSHARE))
1330                continue;
1331
1332            /* search for overlapping mappings on the same file */
1333            if (file_inode(pregion->vm_file) !=
1334                file_inode(file))
1335                continue;
1336
1337            if (pregion->vm_pgoff >= pgend)
1338                continue;
1339
1340            rpglen = pregion->vm_end - pregion->vm_start;
1341            rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1342            rpgend = pregion->vm_pgoff + rpglen;
1343            if (pgoff >= rpgend)
1344                continue;
1345
1346            /* handle inexactly overlapping matches between
1347             * mappings */
1348            if ((pregion->vm_pgoff != pgoff || rpglen != pglen) &&
1349                !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) {
1350                /* new mapping is not a subset of the region */
1351                if (!(capabilities & BDI_CAP_MAP_DIRECT))
1352                    goto sharing_violation;
1353                continue;
1354            }
1355
1356            /* we've found a region we can share */
1357            pregion->vm_usage++;
1358            vma->vm_region = pregion;
1359            start = pregion->vm_start;
1360            start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT;
1361            vma->vm_start = start;
1362            vma->vm_end = start + len;
1363
1364            if (pregion->vm_flags & VM_MAPPED_COPY) {
1365                kdebug("share copy");
1366                vma->vm_flags |= VM_MAPPED_COPY;
1367            } else {
1368                kdebug("share mmap");
1369                ret = do_mmap_shared_file(vma);
1370                if (ret < 0) {
1371                    vma->vm_region = NULL;
1372                    vma->vm_start = 0;
1373                    vma->vm_end = 0;
1374                    pregion->vm_usage--;
1375                    pregion = NULL;
1376                    goto error_just_free;
1377                }
1378            }
1379            fput(region->vm_file);
1380            kmem_cache_free(vm_region_jar, region);
1381            region = pregion;
1382            result = start;
1383            goto share;
1384        }
1385
1386        /* obtain the address at which to make a shared mapping
1387         * - this is the hook for quasi-memory character devices to
1388         * tell us the location of a shared mapping
1389         */
1390        if (capabilities & BDI_CAP_MAP_DIRECT) {
1391            addr = file->f_op->get_unmapped_area(file, addr, len,
1392                                 pgoff, flags);
1393            if (IS_ERR_VALUE(addr)) {
1394                ret = addr;
1395                if (ret != -ENOSYS)
1396                    goto error_just_free;
1397
1398                /* the driver refused to tell us where to site
1399                 * the mapping so we'll have to attempt to copy
1400                 * it */
1401                ret = -ENODEV;
1402                if (!(capabilities & BDI_CAP_MAP_COPY))
1403                    goto error_just_free;
1404
1405                capabilities &= ~BDI_CAP_MAP_DIRECT;
1406            } else {
1407                vma->vm_start = region->vm_start = addr;
1408                vma->vm_end = region->vm_end = addr + len;
1409            }
1410        }
1411    }
1412
1413    vma->vm_region = region;
1414
1415    /* set up the mapping
1416     * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1417     */
1418    if (file && vma->vm_flags & VM_SHARED)
1419        ret = do_mmap_shared_file(vma);
1420    else
1421        ret = do_mmap_private(vma, region, len, capabilities);
1422    if (ret < 0)
1423        goto error_just_free;
1424    add_nommu_region(region);
1425
1426    /* clear anonymous mappings that don't ask for uninitialized data */
1427    if (!vma->vm_file && !(flags & MAP_UNINITIALIZED))
1428        memset((void *)region->vm_start, 0,
1429               region->vm_end - region->vm_start);
1430
1431    /* okay... we have a mapping; now we have to register it */
1432    result = vma->vm_start;
1433
1434    current->mm->total_vm += len >> PAGE_SHIFT;
1435
1436share:
1437    add_vma_to_mm(current->mm, vma);
1438
1439    /* we flush the region from the icache only when the first executable
1440     * mapping of it is made */
1441    if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) {
1442        flush_icache_range(region->vm_start, region->vm_end);
1443        region->vm_icache_flushed = true;
1444    }
1445
1446    up_write(&nommu_region_sem);
1447
1448    kleave(" = %lx", result);
1449    return result;
1450
1451error_just_free:
1452    up_write(&nommu_region_sem);
1453error:
1454    if (region->vm_file)
1455        fput(region->vm_file);
1456    kmem_cache_free(vm_region_jar, region);
1457    if (vma->vm_file)
1458        fput(vma->vm_file);
1459    kmem_cache_free(vm_area_cachep, vma);
1460    kleave(" = %d", ret);
1461    return ret;
1462
1463sharing_violation:
1464    up_write(&nommu_region_sem);
1465    printk(KERN_WARNING "Attempt to share mismatched mappings\n");
1466    ret = -EINVAL;
1467    goto error;
1468
1469error_getting_vma:
1470    kmem_cache_free(vm_region_jar, region);
1471    printk(KERN_WARNING "Allocation of vma for %lu byte allocation"
1472           " from process %d failed\n",
1473           len, current->pid);
1474    show_free_areas(0);
1475    return -ENOMEM;
1476
1477error_getting_region:
1478    printk(KERN_WARNING "Allocation of vm region for %lu byte allocation"
1479           " from process %d failed\n",
1480           len, current->pid);
1481    show_free_areas(0);
1482    return -ENOMEM;
1483}
1484
1485SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1486        unsigned long, prot, unsigned long, flags,
1487        unsigned long, fd, unsigned long, pgoff)
1488{
1489    struct file *file = NULL;
1490    unsigned long retval = -EBADF;
1491
1492    audit_mmap_fd(fd, flags);
1493    if (!(flags & MAP_ANONYMOUS)) {
1494        file = fget(fd);
1495        if (!file)
1496            goto out;
1497    }
1498
1499    flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1500
1501    retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1502
1503    if (file)
1504        fput(file);
1505out:
1506    return retval;
1507}
1508
1509#ifdef __ARCH_WANT_SYS_OLD_MMAP
1510struct mmap_arg_struct {
1511    unsigned long addr;
1512    unsigned long len;
1513    unsigned long prot;
1514    unsigned long flags;
1515    unsigned long fd;
1516    unsigned long offset;
1517};
1518
1519SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1520{
1521    struct mmap_arg_struct a;
1522
1523    if (copy_from_user(&a, arg, sizeof(a)))
1524        return -EFAULT;
1525    if (a.offset & ~PAGE_MASK)
1526        return -EINVAL;
1527
1528    return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1529                  a.offset >> PAGE_SHIFT);
1530}
1531#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1532
1533/*
1534 * split a vma into two pieces at address 'addr', a new vma is allocated either
1535 * for the first part or the tail.
1536 */
1537int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
1538          unsigned long addr, int new_below)
1539{
1540    struct vm_area_struct *new;
1541    struct vm_region *region;
1542    unsigned long npages;
1543
1544    kenter("");
1545
1546    /* we're only permitted to split anonymous regions (these should have
1547     * only a single usage on the region) */
1548    if (vma->vm_file)
1549        return -ENOMEM;
1550
1551    if (mm->map_count >= sysctl_max_map_count)
1552        return -ENOMEM;
1553
1554    region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL);
1555    if (!region)
1556        return -ENOMEM;
1557
1558    new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1559    if (!new) {
1560        kmem_cache_free(vm_region_jar, region);
1561        return -ENOMEM;
1562    }
1563
1564    /* most fields are the same, copy all, and then fixup */
1565    *new = *vma;
1566    *region = *vma->vm_region;
1567    new->vm_region = region;
1568
1569    npages = (addr - vma->vm_start) >> PAGE_SHIFT;
1570
1571    if (new_below) {
1572        region->vm_top = region->vm_end = new->vm_end = addr;
1573    } else {
1574        region->vm_start = new->vm_start = addr;
1575        region->vm_pgoff = new->vm_pgoff += npages;
1576    }
1577
1578    if (new->vm_ops && new->vm_ops->open)
1579        new->vm_ops->open(new);
1580
1581    delete_vma_from_mm(vma);
1582    down_write(&nommu_region_sem);
1583    delete_nommu_region(vma->vm_region);
1584    if (new_below) {
1585        vma->vm_region->vm_start = vma->vm_start = addr;
1586        vma->vm_region->vm_pgoff = vma->vm_pgoff += npages;
1587    } else {
1588        vma->vm_region->vm_end = vma->vm_end = addr;
1589        vma->vm_region->vm_top = addr;
1590    }
1591    add_nommu_region(vma->vm_region);
1592    add_nommu_region(new->vm_region);
1593    up_write(&nommu_region_sem);
1594    add_vma_to_mm(mm, vma);
1595    add_vma_to_mm(mm, new);
1596    return 0;
1597}
1598
1599/*
1600 * shrink a VMA by removing the specified chunk from either the beginning or
1601 * the end
1602 */
1603static int shrink_vma(struct mm_struct *mm,
1604              struct vm_area_struct *vma,
1605              unsigned long from, unsigned long to)
1606{
1607    struct vm_region *region;
1608
1609    kenter("");
1610
1611    /* adjust the VMA's pointers, which may reposition it in the MM's tree
1612     * and list */
1613    delete_vma_from_mm(vma);
1614    if (from > vma->vm_start)
1615        vma->vm_end = from;
1616    else
1617        vma->vm_start = to;
1618    add_vma_to_mm(mm, vma);
1619
1620    /* cut the backing region down to size */
1621    region = vma->vm_region;
1622    BUG_ON(region->vm_usage != 1);
1623
1624    down_write(&nommu_region_sem);
1625    delete_nommu_region(region);
1626    if (from > region->vm_start) {
1627        to = region->vm_top;
1628        region->vm_top = region->vm_end = from;
1629    } else {
1630        region->vm_start = to;
1631    }
1632    add_nommu_region(region);
1633    up_write(&nommu_region_sem);
1634
1635    free_page_series(from, to);
1636    return 0;
1637}
1638
1639/*
1640 * release a mapping
1641 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1642 * VMA, though it need not cover the whole VMA
1643 */
1644int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1645{
1646    struct vm_area_struct *vma;
1647    unsigned long end;
1648    int ret;
1649
1650    kenter(",%lx,%zx", start, len);
1651
1652    len = PAGE_ALIGN(len);
1653    if (len == 0)
1654        return -EINVAL;
1655
1656    end = start + len;
1657
1658    /* find the first potentially overlapping VMA */
1659    vma = find_vma(mm, start);
1660    if (!vma) {
1661        static int limit = 0;
1662        if (limit < 5) {
1663            printk(KERN_WARNING
1664                   "munmap of memory not mmapped by process %d"
1665                   " (%s): 0x%lx-0x%lx\n",
1666                   current->pid, current->comm,
1667                   start, start + len - 1);
1668            limit++;
1669        }
1670        return -EINVAL;
1671    }
1672
1673    /* we're allowed to split an anonymous VMA but not a file-backed one */
1674    if (vma->vm_file) {
1675        do {
1676            if (start > vma->vm_start) {
1677                kleave(" = -EINVAL [miss]");
1678                return -EINVAL;
1679            }
1680            if (end == vma->vm_end)
1681                goto erase_whole_vma;
1682            vma = vma->vm_next;
1683        } while (vma);
1684        kleave(" = -EINVAL [split file]");
1685        return -EINVAL;
1686    } else {
1687        /* the chunk must be a subset of the VMA found */
1688        if (start == vma->vm_start && end == vma->vm_end)
1689            goto erase_whole_vma;
1690        if (start < vma->vm_start || end > vma->vm_end) {
1691            kleave(" = -EINVAL [superset]");
1692            return -EINVAL;
1693        }
1694        if (start & ~PAGE_MASK) {
1695            kleave(" = -EINVAL [unaligned start]");
1696            return -EINVAL;
1697        }
1698        if (end != vma->vm_end && end & ~PAGE_MASK) {
1699            kleave(" = -EINVAL [unaligned split]");
1700            return -EINVAL;
1701        }
1702        if (start != vma->vm_start && end != vma->vm_end) {
1703            ret = split_vma(mm, vma, start, 1);
1704            if (ret < 0) {
1705                kleave(" = %d [split]", ret);
1706                return ret;
1707            }
1708        }
1709        return shrink_vma(mm, vma, start, end);
1710    }
1711
1712erase_whole_vma:
1713    delete_vma_from_mm(vma);
1714    delete_vma(mm, vma);
1715    kleave(" = 0");
1716    return 0;
1717}
1718EXPORT_SYMBOL(do_munmap);
1719
1720int vm_munmap(unsigned long addr, size_t len)
1721{
1722    struct mm_struct *mm = current->mm;
1723    int ret;
1724
1725    down_write(&mm->mmap_sem);
1726    ret = do_munmap(mm, addr, len);
1727    up_write(&mm->mmap_sem);
1728    return ret;
1729}
1730EXPORT_SYMBOL(vm_munmap);
1731
1732SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1733{
1734    return vm_munmap(addr, len);
1735}
1736
1737/*
1738 * release all the mappings made in a process's VM space
1739 */
1740void exit_mmap(struct mm_struct *mm)
1741{
1742    struct vm_area_struct *vma;
1743
1744    if (!mm)
1745        return;
1746
1747    kenter("");
1748
1749    mm->total_vm = 0;
1750
1751    while ((vma = mm->mmap)) {
1752        mm->mmap = vma->vm_next;
1753        delete_vma_from_mm(vma);
1754        delete_vma(mm, vma);
1755        cond_resched();
1756    }
1757
1758    kleave("");
1759}
1760
1761unsigned long vm_brk(unsigned long addr, unsigned long len)
1762{
1763    return -ENOMEM;
1764}
1765
1766/*
1767 * expand (or shrink) an existing mapping, potentially moving it at the same
1768 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1769 *
1770 * under NOMMU conditions, we only permit changing a mapping's size, and only
1771 * as long as it stays within the region allocated by do_mmap_private() and the
1772 * block is not shareable
1773 *
1774 * MREMAP_FIXED is not supported under NOMMU conditions
1775 */
1776static unsigned long do_mremap(unsigned long addr,
1777            unsigned long old_len, unsigned long new_len,
1778            unsigned long flags, unsigned long new_addr)
1779{
1780    struct vm_area_struct *vma;
1781
1782    /* insanity checks first */
1783    old_len = PAGE_ALIGN(old_len);
1784    new_len = PAGE_ALIGN(new_len);
1785    if (old_len == 0 || new_len == 0)
1786        return (unsigned long) -EINVAL;
1787
1788    if (addr & ~PAGE_MASK)
1789        return -EINVAL;
1790
1791    if (flags & MREMAP_FIXED && new_addr != addr)
1792        return (unsigned long) -EINVAL;
1793
1794    vma = find_vma_exact(current->mm, addr, old_len);
1795    if (!vma)
1796        return (unsigned long) -EINVAL;
1797
1798    if (vma->vm_end != vma->vm_start + old_len)
1799        return (unsigned long) -EFAULT;
1800
1801    if (vma->vm_flags & VM_MAYSHARE)
1802        return (unsigned long) -EPERM;
1803
1804    if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start)
1805        return (unsigned long) -ENOMEM;
1806
1807    /* all checks complete - do it */
1808    vma->vm_end = vma->vm_start + new_len;
1809    return vma->vm_start;
1810}
1811
1812SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
1813        unsigned long, new_len, unsigned long, flags,
1814        unsigned long, new_addr)
1815{
1816    unsigned long ret;
1817
1818    down_write(&current->mm->mmap_sem);
1819    ret = do_mremap(addr, old_len, new_len, flags, new_addr);
1820    up_write(&current->mm->mmap_sem);
1821    return ret;
1822}
1823
1824struct page *follow_page_mask(struct vm_area_struct *vma,
1825                  unsigned long address, unsigned int flags,
1826                  unsigned int *page_mask)
1827{
1828    *page_mask = 0;
1829    return NULL;
1830}
1831
1832int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1833        unsigned long pfn, unsigned long size, pgprot_t prot)
1834{
1835    if (addr != (pfn << PAGE_SHIFT))
1836        return -EINVAL;
1837
1838    vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1839    return 0;
1840}
1841EXPORT_SYMBOL(remap_pfn_range);
1842
1843int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len)
1844{
1845    unsigned long pfn = start >> PAGE_SHIFT;
1846    unsigned long vm_len = vma->vm_end - vma->vm_start;
1847
1848    pfn += vma->vm_pgoff;
1849    return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot);
1850}
1851EXPORT_SYMBOL(vm_iomap_memory);
1852
1853int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1854            unsigned long pgoff)
1855{
1856    unsigned int size = vma->vm_end - vma->vm_start;
1857
1858    if (!(vma->vm_flags & VM_USERMAP))
1859        return -EINVAL;
1860
1861    vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT));
1862    vma->vm_end = vma->vm_start + size;
1863
1864    return 0;
1865}
1866EXPORT_SYMBOL(remap_vmalloc_range);
1867
1868unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
1869    unsigned long len, unsigned long pgoff, unsigned long flags)
1870{
1871    return -ENOMEM;
1872}
1873
1874void unmap_mapping_range(struct address_space *mapping,
1875             loff_t const holebegin, loff_t const holelen,
1876             int even_cows)
1877{
1878}
1879EXPORT_SYMBOL(unmap_mapping_range);
1880
1881/*
1882 * Check that a process has enough memory to allocate a new virtual
1883 * mapping. 0 means there is enough memory for the allocation to
1884 * succeed and -ENOMEM implies there is not.
1885 *
1886 * We currently support three overcommit policies, which are set via the
1887 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1888 *
1889 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1890 * Additional code 2002 Jul 20 by Robert Love.
1891 *
1892 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1893 *
1894 * Note this is a helper function intended to be used by LSMs which
1895 * wish to use this logic.
1896 */
1897int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
1898{
1899    unsigned long free, allowed, reserve;
1900
1901    vm_acct_memory(pages);
1902
1903    /*
1904     * Sometimes we want to use more memory than we have
1905     */
1906    if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
1907        return 0;
1908
1909    if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
1910        free = global_page_state(NR_FREE_PAGES);
1911        free += global_page_state(NR_FILE_PAGES);
1912
1913        /*
1914         * shmem pages shouldn't be counted as free in this
1915         * case, they can't be purged, only swapped out, and
1916         * that won't affect the overall amount of available
1917         * memory in the system.
1918         */
1919        free -= global_page_state(NR_SHMEM);
1920
1921        free += get_nr_swap_pages();
1922
1923        /*
1924         * Any slabs which are created with the
1925         * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1926         * which are reclaimable, under pressure. The dentry
1927         * cache and most inode caches should fall into this
1928         */
1929        free += global_page_state(NR_SLAB_RECLAIMABLE);
1930
1931        /*
1932         * Leave reserved pages. The pages are not for anonymous pages.
1933         */
1934        if (free <= totalreserve_pages)
1935            goto error;
1936        else
1937            free -= totalreserve_pages;
1938
1939        /*
1940         * Reserve some for root
1941         */
1942        if (!cap_sys_admin)
1943            free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1944
1945        if (free > pages)
1946            return 0;
1947
1948        goto error;
1949    }
1950
1951    allowed = totalram_pages * sysctl_overcommit_ratio / 100;
1952    /*
1953     * Reserve some 3% for root
1954     */
1955    if (!cap_sys_admin)
1956        allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
1957    allowed += total_swap_pages;
1958
1959    /*
1960     * Don't let a single process grow so big a user can't recover
1961     */
1962    if (mm) {
1963        reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
1964        allowed -= min(mm->total_vm / 32, reserve);
1965    }
1966
1967    if (percpu_counter_read_positive(&vm_committed_as) < allowed)
1968        return 0;
1969
1970error:
1971    vm_unacct_memory(pages);
1972
1973    return -ENOMEM;
1974}
1975
1976int in_gate_area_no_mm(unsigned long addr)
1977{
1978    return 0;
1979}
1980
1981int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1982{
1983    BUG();
1984    return 0;
1985}
1986EXPORT_SYMBOL(filemap_fault);
1987
1988int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr,
1989                 unsigned long size, pgoff_t pgoff)
1990{
1991    BUG();
1992    return 0;
1993}
1994EXPORT_SYMBOL(generic_file_remap_pages);
1995
1996static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
1997        unsigned long addr, void *buf, int len, int write)
1998{
1999    struct vm_area_struct *vma;
2000
2001    down_read(&mm->mmap_sem);
2002
2003    /* the access must start within one of the target process's mappings */
2004    vma = find_vma(mm, addr);
2005    if (vma) {
2006        /* don't overrun this mapping */
2007        if (addr + len >= vma->vm_end)
2008            len = vma->vm_end - addr;
2009
2010        /* only read or write mappings where it is permitted */
2011        if (write && vma->vm_flags & VM_MAYWRITE)
2012            copy_to_user_page(vma, NULL, addr,
2013                     (void *) addr, buf, len);
2014        else if (!write && vma->vm_flags & VM_MAYREAD)
2015            copy_from_user_page(vma, NULL, addr,
2016                        buf, (void *) addr, len);
2017        else
2018            len = 0;
2019    } else {
2020        len = 0;
2021    }
2022
2023    up_read(&mm->mmap_sem);
2024
2025    return len;
2026}
2027
2028/**
2029 * @access_remote_vm - access another process' address space
2030 * @mm: the mm_struct of the target address space
2031 * @addr: start address to access
2032 * @buf: source or destination buffer
2033 * @len: number of bytes to transfer
2034 * @write: whether the access is a write
2035 *
2036 * The caller must hold a reference on @mm.
2037 */
2038int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2039        void *buf, int len, int write)
2040{
2041    return __access_remote_vm(NULL, mm, addr, buf, len, write);
2042}
2043
2044/*
2045 * Access another process' address space.
2046 * - source/target buffer must be kernel space
2047 */
2048int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2049{
2050    struct mm_struct *mm;
2051
2052    if (addr + len < addr)
2053        return 0;
2054
2055    mm = get_task_mm(tsk);
2056    if (!mm)
2057        return 0;
2058
2059    len = __access_remote_vm(tsk, mm, addr, buf, len, write);
2060
2061    mmput(mm);
2062    return len;
2063}
2064
2065/**
2066 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2067 * @inode: The inode to check
2068 * @size: The current filesize of the inode
2069 * @newsize: The proposed filesize of the inode
2070 *
2071 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2072 * make sure that that any outstanding VMAs aren't broken and then shrink the
2073 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2074 * automatically grant mappings that are too large.
2075 */
2076int nommu_shrink_inode_mappings(struct inode *inode, size_t size,
2077                size_t newsize)
2078{
2079    struct vm_area_struct *vma;
2080    struct vm_region *region;
2081    pgoff_t low, high;
2082    size_t r_size, r_top;
2083
2084    low = newsize >> PAGE_SHIFT;
2085    high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2086
2087    down_write(&nommu_region_sem);
2088    mutex_lock(&inode->i_mapping->i_mmap_mutex);
2089
2090    /* search for VMAs that fall within the dead zone */
2091    vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) {
2092        /* found one - only interested if it's shared out of the page
2093         * cache */
2094        if (vma->vm_flags & VM_SHARED) {
2095            mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2096            up_write(&nommu_region_sem);
2097            return -ETXTBSY; /* not quite true, but near enough */
2098        }
2099    }
2100
2101    /* reduce any regions that overlap the dead zone - if in existence,
2102     * these will be pointed to by VMAs that don't overlap the dead zone
2103     *
2104     * we don't check for any regions that start beyond the EOF as there
2105     * shouldn't be any
2106     */
2107    vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap,
2108                  0, ULONG_MAX) {
2109        if (!(vma->vm_flags & VM_SHARED))
2110            continue;
2111
2112        region = vma->vm_region;
2113        r_size = region->vm_top - region->vm_start;
2114        r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size;
2115
2116        if (r_top > newsize) {
2117            region->vm_top -= r_top - newsize;
2118            if (region->vm_end > region->vm_top)
2119                region->vm_end = region->vm_top;
2120        }
2121    }
2122
2123    mutex_unlock(&inode->i_mapping->i_mmap_mutex);
2124    up_write(&nommu_region_sem);
2125    return 0;
2126}
2127
2128/*
2129 * Initialise sysctl_user_reserve_kbytes.
2130 *
2131 * This is intended to prevent a user from starting a single memory hogging
2132 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2133 * mode.
2134 *
2135 * The default value is min(3% of free memory, 128MB)
2136 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2137 */
2138static int __meminit init_user_reserve(void)
2139{
2140    unsigned long free_kbytes;
2141
2142    free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2143
2144    sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
2145    return 0;
2146}
2147module_init(init_user_reserve)
2148
2149/*
2150 * Initialise sysctl_admin_reserve_kbytes.
2151 *
2152 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2153 * to log in and kill a memory hogging process.
2154 *
2155 * Systems with more than 256MB will reserve 8MB, enough to recover
2156 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2157 * only reserve 3% of free pages by default.
2158 */
2159static int __meminit init_admin_reserve(void)
2160{
2161    unsigned long free_kbytes;
2162
2163    free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
2164
2165    sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
2166    return 0;
2167}
2168module_init(init_admin_reserve)
2169

Archive Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9



interactive