Root/mm/mmap.c

1/*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
7 */
8
9#include <linux/slab.h>
10#include <linux/backing-dev.h>
11#include <linux/mm.h>
12#include <linux/shm.h>
13#include <linux/mman.h>
14#include <linux/pagemap.h>
15#include <linux/swap.h>
16#include <linux/syscalls.h>
17#include <linux/capability.h>
18#include <linux/init.h>
19#include <linux/file.h>
20#include <linux/fs.h>
21#include <linux/personality.h>
22#include <linux/security.h>
23#include <linux/hugetlb.h>
24#include <linux/profile.h>
25#include <linux/module.h>
26#include <linux/mount.h>
27#include <linux/mempolicy.h>
28#include <linux/rmap.h>
29#include <linux/mmu_notifier.h>
30#include <linux/perf_event.h>
31
32#include <asm/uaccess.h>
33#include <asm/cacheflush.h>
34#include <asm/tlb.h>
35#include <asm/mmu_context.h>
36
37#include "internal.h"
38
39#ifndef arch_mmap_check
40#define arch_mmap_check(addr, len, flags) (0)
41#endif
42
43#ifndef arch_rebalance_pgtables
44#define arch_rebalance_pgtables(addr, len) (addr)
45#endif
46
47static void unmap_region(struct mm_struct *mm,
48        struct vm_area_struct *vma, struct vm_area_struct *prev,
49        unsigned long start, unsigned long end);
50
51/*
52 * WARNING: the debugging will use recursive algorithms so never enable this
53 * unless you know what you are doing.
54 */
55#undef DEBUG_MM_RB
56
57/* description of effects of mapping type and prot in current implementation.
58 * this is due to the limited x86 page protection hardware. The expected
59 * behavior is in parens:
60 *
61 * map_type prot
62 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
63 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
64 * w: (no) no w: (no) no w: (yes) yes w: (no) no
65 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
66 *
67 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
68 * w: (no) no w: (no) no w: (copy) copy w: (no) no
69 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 *
71 */
72pgprot_t protection_map[16] = {
73    __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
74    __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
75};
76
77pgprot_t vm_get_page_prot(unsigned long vm_flags)
78{
79    return __pgprot(pgprot_val(protection_map[vm_flags &
80                (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
81            pgprot_val(arch_vm_get_page_prot(vm_flags)));
82}
83EXPORT_SYMBOL(vm_get_page_prot);
84
85int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
86int sysctl_overcommit_ratio = 50; /* default is 50% */
87int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
88struct percpu_counter vm_committed_as;
89
90/*
91 * Check that a process has enough memory to allocate a new virtual
92 * mapping. 0 means there is enough memory for the allocation to
93 * succeed and -ENOMEM implies there is not.
94 *
95 * We currently support three overcommit policies, which are set via the
96 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
97 *
98 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
99 * Additional code 2002 Jul 20 by Robert Love.
100 *
101 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
102 *
103 * Note this is a helper function intended to be used by LSMs which
104 * wish to use this logic.
105 */
106int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
107{
108    unsigned long free, allowed;
109
110    vm_acct_memory(pages);
111
112    /*
113     * Sometimes we want to use more memory than we have
114     */
115    if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
116        return 0;
117
118    if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
119        unsigned long n;
120
121        free = global_page_state(NR_FILE_PAGES);
122        free += nr_swap_pages;
123
124        /*
125         * Any slabs which are created with the
126         * SLAB_RECLAIM_ACCOUNT flag claim to have contents
127         * which are reclaimable, under pressure. The dentry
128         * cache and most inode caches should fall into this
129         */
130        free += global_page_state(NR_SLAB_RECLAIMABLE);
131
132        /*
133         * Leave the last 3% for root
134         */
135        if (!cap_sys_admin)
136            free -= free / 32;
137
138        if (free > pages)
139            return 0;
140
141        /*
142         * nr_free_pages() is very expensive on large systems,
143         * only call if we're about to fail.
144         */
145        n = nr_free_pages();
146
147        /*
148         * Leave reserved pages. The pages are not for anonymous pages.
149         */
150        if (n <= totalreserve_pages)
151            goto error;
152        else
153            n -= totalreserve_pages;
154
155        /*
156         * Leave the last 3% for root
157         */
158        if (!cap_sys_admin)
159            n -= n / 32;
160        free += n;
161
162        if (free > pages)
163            return 0;
164
165        goto error;
166    }
167
168    allowed = (totalram_pages - hugetlb_total_pages())
169               * sysctl_overcommit_ratio / 100;
170    /*
171     * Leave the last 3% for root
172     */
173    if (!cap_sys_admin)
174        allowed -= allowed / 32;
175    allowed += total_swap_pages;
176
177    /* Don't let a single process grow too big:
178       leave 3% of the size of this process for other processes */
179    if (mm)
180        allowed -= mm->total_vm / 32;
181
182    if (percpu_counter_read_positive(&vm_committed_as) < allowed)
183        return 0;
184error:
185    vm_unacct_memory(pages);
186
187    return -ENOMEM;
188}
189
190/*
191 * Requires inode->i_mapping->i_mmap_lock
192 */
193static void __remove_shared_vm_struct(struct vm_area_struct *vma,
194        struct file *file, struct address_space *mapping)
195{
196    if (vma->vm_flags & VM_DENYWRITE)
197        atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
198    if (vma->vm_flags & VM_SHARED)
199        mapping->i_mmap_writable--;
200
201    flush_dcache_mmap_lock(mapping);
202    if (unlikely(vma->vm_flags & VM_NONLINEAR))
203        list_del_init(&vma->shared.vm_set.list);
204    else
205        vma_prio_tree_remove(vma, &mapping->i_mmap);
206    flush_dcache_mmap_unlock(mapping);
207}
208
209/*
210 * Unlink a file-based vm structure from its prio_tree, to hide
211 * vma from rmap and vmtruncate before freeing its page tables.
212 */
213void unlink_file_vma(struct vm_area_struct *vma)
214{
215    struct file *file = vma->vm_file;
216
217    if (file) {
218        struct address_space *mapping = file->f_mapping;
219        spin_lock(&mapping->i_mmap_lock);
220        __remove_shared_vm_struct(vma, file, mapping);
221        spin_unlock(&mapping->i_mmap_lock);
222    }
223}
224
225/*
226 * Close a vm structure and free it, returning the next.
227 */
228static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
229{
230    struct vm_area_struct *next = vma->vm_next;
231
232    might_sleep();
233    if (vma->vm_ops && vma->vm_ops->close)
234        vma->vm_ops->close(vma);
235    if (vma->vm_file) {
236        fput(vma->vm_file);
237        if (vma->vm_flags & VM_EXECUTABLE)
238            removed_exe_file_vma(vma->vm_mm);
239    }
240    mpol_put(vma_policy(vma));
241    kmem_cache_free(vm_area_cachep, vma);
242    return next;
243}
244
245SYSCALL_DEFINE1(brk, unsigned long, brk)
246{
247    unsigned long rlim, retval;
248    unsigned long newbrk, oldbrk;
249    struct mm_struct *mm = current->mm;
250    unsigned long min_brk;
251
252    down_write(&mm->mmap_sem);
253
254#ifdef CONFIG_COMPAT_BRK
255    min_brk = mm->end_code;
256#else
257    min_brk = mm->start_brk;
258#endif
259    if (brk < min_brk)
260        goto out;
261
262    /*
263     * Check against rlimit here. If this check is done later after the test
264     * of oldbrk with newbrk then it can escape the test and let the data
265     * segment grow beyond its set limit the in case where the limit is
266     * not page aligned -Ram Gupta
267     */
268    rlim = rlimit(RLIMIT_DATA);
269    if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
270            (mm->end_data - mm->start_data) > rlim)
271        goto out;
272
273    newbrk = PAGE_ALIGN(brk);
274    oldbrk = PAGE_ALIGN(mm->brk);
275    if (oldbrk == newbrk)
276        goto set_brk;
277
278    /* Always allow shrinking brk. */
279    if (brk <= mm->brk) {
280        if (!do_munmap(mm, newbrk, oldbrk-newbrk))
281            goto set_brk;
282        goto out;
283    }
284
285    /* Check against existing mmap mappings. */
286    if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
287        goto out;
288
289    /* Ok, looks good - let it rip. */
290    if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
291        goto out;
292set_brk:
293    mm->brk = brk;
294out:
295    retval = mm->brk;
296    up_write(&mm->mmap_sem);
297    return retval;
298}
299
300#ifdef DEBUG_MM_RB
301static int browse_rb(struct rb_root *root)
302{
303    int i = 0, j;
304    struct rb_node *nd, *pn = NULL;
305    unsigned long prev = 0, pend = 0;
306
307    for (nd = rb_first(root); nd; nd = rb_next(nd)) {
308        struct vm_area_struct *vma;
309        vma = rb_entry(nd, struct vm_area_struct, vm_rb);
310        if (vma->vm_start < prev)
311            printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
312        if (vma->vm_start < pend)
313            printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
314        if (vma->vm_start > vma->vm_end)
315            printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
316        i++;
317        pn = nd;
318        prev = vma->vm_start;
319        pend = vma->vm_end;
320    }
321    j = 0;
322    for (nd = pn; nd; nd = rb_prev(nd)) {
323        j++;
324    }
325    if (i != j)
326        printk("backwards %d, forwards %d\n", j, i), i = 0;
327    return i;
328}
329
330void validate_mm(struct mm_struct *mm)
331{
332    int bug = 0;
333    int i = 0;
334    struct vm_area_struct *tmp = mm->mmap;
335    while (tmp) {
336        tmp = tmp->vm_next;
337        i++;
338    }
339    if (i != mm->map_count)
340        printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
341    i = browse_rb(&mm->mm_rb);
342    if (i != mm->map_count)
343        printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
344    BUG_ON(bug);
345}
346#else
347#define validate_mm(mm) do { } while (0)
348#endif
349
350static struct vm_area_struct *
351find_vma_prepare(struct mm_struct *mm, unsigned long addr,
352        struct vm_area_struct **pprev, struct rb_node ***rb_link,
353        struct rb_node ** rb_parent)
354{
355    struct vm_area_struct * vma;
356    struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
357
358    __rb_link = &mm->mm_rb.rb_node;
359    rb_prev = __rb_parent = NULL;
360    vma = NULL;
361
362    while (*__rb_link) {
363        struct vm_area_struct *vma_tmp;
364
365        __rb_parent = *__rb_link;
366        vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
367
368        if (vma_tmp->vm_end > addr) {
369            vma = vma_tmp;
370            if (vma_tmp->vm_start <= addr)
371                break;
372            __rb_link = &__rb_parent->rb_left;
373        } else {
374            rb_prev = __rb_parent;
375            __rb_link = &__rb_parent->rb_right;
376        }
377    }
378
379    *pprev = NULL;
380    if (rb_prev)
381        *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
382    *rb_link = __rb_link;
383    *rb_parent = __rb_parent;
384    return vma;
385}
386
387static inline void
388__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
389        struct vm_area_struct *prev, struct rb_node *rb_parent)
390{
391    if (prev) {
392        vma->vm_next = prev->vm_next;
393        prev->vm_next = vma;
394    } else {
395        mm->mmap = vma;
396        if (rb_parent)
397            vma->vm_next = rb_entry(rb_parent,
398                    struct vm_area_struct, vm_rb);
399        else
400            vma->vm_next = NULL;
401    }
402}
403
404void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
405        struct rb_node **rb_link, struct rb_node *rb_parent)
406{
407    rb_link_node(&vma->vm_rb, rb_parent, rb_link);
408    rb_insert_color(&vma->vm_rb, &mm->mm_rb);
409}
410
411static void __vma_link_file(struct vm_area_struct *vma)
412{
413    struct file *file;
414
415    file = vma->vm_file;
416    if (file) {
417        struct address_space *mapping = file->f_mapping;
418
419        if (vma->vm_flags & VM_DENYWRITE)
420            atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
421        if (vma->vm_flags & VM_SHARED)
422            mapping->i_mmap_writable++;
423
424        flush_dcache_mmap_lock(mapping);
425        if (unlikely(vma->vm_flags & VM_NONLINEAR))
426            vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
427        else
428            vma_prio_tree_insert(vma, &mapping->i_mmap);
429        flush_dcache_mmap_unlock(mapping);
430    }
431}
432
433static void
434__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
435    struct vm_area_struct *prev, struct rb_node **rb_link,
436    struct rb_node *rb_parent)
437{
438    __vma_link_list(mm, vma, prev, rb_parent);
439    __vma_link_rb(mm, vma, rb_link, rb_parent);
440}
441
442static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
443            struct vm_area_struct *prev, struct rb_node **rb_link,
444            struct rb_node *rb_parent)
445{
446    struct address_space *mapping = NULL;
447
448    if (vma->vm_file)
449        mapping = vma->vm_file->f_mapping;
450
451    if (mapping) {
452        spin_lock(&mapping->i_mmap_lock);
453        vma->vm_truncate_count = mapping->truncate_count;
454    }
455    anon_vma_lock(vma);
456
457    __vma_link(mm, vma, prev, rb_link, rb_parent);
458    __vma_link_file(vma);
459
460    anon_vma_unlock(vma);
461    if (mapping)
462        spin_unlock(&mapping->i_mmap_lock);
463
464    mm->map_count++;
465    validate_mm(mm);
466}
467
468/*
469 * Helper for vma_adjust in the split_vma insert case:
470 * insert vm structure into list and rbtree and anon_vma,
471 * but it has already been inserted into prio_tree earlier.
472 */
473static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
474{
475    struct vm_area_struct *__vma, *prev;
476    struct rb_node **rb_link, *rb_parent;
477
478    __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
479    BUG_ON(__vma && __vma->vm_start < vma->vm_end);
480    __vma_link(mm, vma, prev, rb_link, rb_parent);
481    mm->map_count++;
482}
483
484static inline void
485__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
486        struct vm_area_struct *prev)
487{
488    prev->vm_next = vma->vm_next;
489    rb_erase(&vma->vm_rb, &mm->mm_rb);
490    if (mm->mmap_cache == vma)
491        mm->mmap_cache = prev;
492}
493
494/*
495 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
496 * is already present in an i_mmap tree without adjusting the tree.
497 * The following helper function should be used when such adjustments
498 * are necessary. The "insert" vma (if any) is to be inserted
499 * before we drop the necessary locks.
500 */
501int vma_adjust(struct vm_area_struct *vma, unsigned long start,
502    unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
503{
504    struct mm_struct *mm = vma->vm_mm;
505    struct vm_area_struct *next = vma->vm_next;
506    struct vm_area_struct *importer = NULL;
507    struct address_space *mapping = NULL;
508    struct prio_tree_root *root = NULL;
509    struct file *file = vma->vm_file;
510    long adjust_next = 0;
511    int remove_next = 0;
512
513    if (next && !insert) {
514        struct vm_area_struct *exporter = NULL;
515
516        if (end >= next->vm_end) {
517            /*
518             * vma expands, overlapping all the next, and
519             * perhaps the one after too (mprotect case 6).
520             */
521again: remove_next = 1 + (end > next->vm_end);
522            end = next->vm_end;
523            exporter = next;
524            importer = vma;
525        } else if (end > next->vm_start) {
526            /*
527             * vma expands, overlapping part of the next:
528             * mprotect case 5 shifting the boundary up.
529             */
530            adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
531            exporter = next;
532            importer = vma;
533        } else if (end < vma->vm_end) {
534            /*
535             * vma shrinks, and !insert tells it's not
536             * split_vma inserting another: so it must be
537             * mprotect case 4 shifting the boundary down.
538             */
539            adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
540            exporter = vma;
541            importer = next;
542        }
543
544        /*
545         * Easily overlooked: when mprotect shifts the boundary,
546         * make sure the expanding vma has anon_vma set if the
547         * shrinking vma had, to cover any anon pages imported.
548         */
549        if (exporter && exporter->anon_vma && !importer->anon_vma) {
550            if (anon_vma_clone(importer, exporter))
551                return -ENOMEM;
552            importer->anon_vma = exporter->anon_vma;
553        }
554    }
555
556    if (file) {
557        mapping = file->f_mapping;
558        if (!(vma->vm_flags & VM_NONLINEAR))
559            root = &mapping->i_mmap;
560        spin_lock(&mapping->i_mmap_lock);
561        if (importer &&
562            vma->vm_truncate_count != next->vm_truncate_count) {
563            /*
564             * unmap_mapping_range might be in progress:
565             * ensure that the expanding vma is rescanned.
566             */
567            importer->vm_truncate_count = 0;
568        }
569        if (insert) {
570            insert->vm_truncate_count = vma->vm_truncate_count;
571            /*
572             * Put into prio_tree now, so instantiated pages
573             * are visible to arm/parisc __flush_dcache_page
574             * throughout; but we cannot insert into address
575             * space until vma start or end is updated.
576             */
577            __vma_link_file(insert);
578        }
579    }
580
581    if (root) {
582        flush_dcache_mmap_lock(mapping);
583        vma_prio_tree_remove(vma, root);
584        if (adjust_next)
585            vma_prio_tree_remove(next, root);
586    }
587
588    vma->vm_start = start;
589    vma->vm_end = end;
590    vma->vm_pgoff = pgoff;
591    if (adjust_next) {
592        next->vm_start += adjust_next << PAGE_SHIFT;
593        next->vm_pgoff += adjust_next;
594    }
595
596    if (root) {
597        if (adjust_next)
598            vma_prio_tree_insert(next, root);
599        vma_prio_tree_insert(vma, root);
600        flush_dcache_mmap_unlock(mapping);
601    }
602
603    if (remove_next) {
604        /*
605         * vma_merge has merged next into vma, and needs
606         * us to remove next before dropping the locks.
607         */
608        __vma_unlink(mm, next, vma);
609        if (file)
610            __remove_shared_vm_struct(next, file, mapping);
611    } else if (insert) {
612        /*
613         * split_vma has split insert from vma, and needs
614         * us to insert it before dropping the locks
615         * (it may either follow vma or precede it).
616         */
617        __insert_vm_struct(mm, insert);
618    }
619
620    if (mapping)
621        spin_unlock(&mapping->i_mmap_lock);
622
623    if (remove_next) {
624        if (file) {
625            fput(file);
626            if (next->vm_flags & VM_EXECUTABLE)
627                removed_exe_file_vma(mm);
628        }
629        if (next->anon_vma)
630            anon_vma_merge(vma, next);
631        mm->map_count--;
632        mpol_put(vma_policy(next));
633        kmem_cache_free(vm_area_cachep, next);
634        /*
635         * In mprotect's case 6 (see comments on vma_merge),
636         * we must remove another next too. It would clutter
637         * up the code too much to do both in one go.
638         */
639        if (remove_next == 2) {
640            next = vma->vm_next;
641            goto again;
642        }
643    }
644
645    validate_mm(mm);
646
647    return 0;
648}
649
650/*
651 * If the vma has a ->close operation then the driver probably needs to release
652 * per-vma resources, so we don't attempt to merge those.
653 */
654static inline int is_mergeable_vma(struct vm_area_struct *vma,
655            struct file *file, unsigned long vm_flags)
656{
657    /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
658    if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
659        return 0;
660    if (vma->vm_file != file)
661        return 0;
662    if (vma->vm_ops && vma->vm_ops->close)
663        return 0;
664    return 1;
665}
666
667static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
668                    struct anon_vma *anon_vma2)
669{
670    return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
671}
672
673/*
674 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
675 * in front of (at a lower virtual address and file offset than) the vma.
676 *
677 * We cannot merge two vmas if they have differently assigned (non-NULL)
678 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
679 *
680 * We don't check here for the merged mmap wrapping around the end of pagecache
681 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
682 * wrap, nor mmaps which cover the final page at index -1UL.
683 */
684static int
685can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
686    struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
687{
688    if (is_mergeable_vma(vma, file, vm_flags) &&
689        is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
690        if (vma->vm_pgoff == vm_pgoff)
691            return 1;
692    }
693    return 0;
694}
695
696/*
697 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
698 * beyond (at a higher virtual address and file offset than) the vma.
699 *
700 * We cannot merge two vmas if they have differently assigned (non-NULL)
701 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
702 */
703static int
704can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
705    struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
706{
707    if (is_mergeable_vma(vma, file, vm_flags) &&
708        is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
709        pgoff_t vm_pglen;
710        vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
711        if (vma->vm_pgoff + vm_pglen == vm_pgoff)
712            return 1;
713    }
714    return 0;
715}
716
717/*
718 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
719 * whether that can be merged with its predecessor or its successor.
720 * Or both (it neatly fills a hole).
721 *
722 * In most cases - when called for mmap, brk or mremap - [addr,end) is
723 * certain not to be mapped by the time vma_merge is called; but when
724 * called for mprotect, it is certain to be already mapped (either at
725 * an offset within prev, or at the start of next), and the flags of
726 * this area are about to be changed to vm_flags - and the no-change
727 * case has already been eliminated.
728 *
729 * The following mprotect cases have to be considered, where AAAA is
730 * the area passed down from mprotect_fixup, never extending beyond one
731 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
732 *
733 * AAAA AAAA AAAA AAAA
734 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
735 * cannot merge might become might become might become
736 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
737 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
738 * mremap move: PPPPNNNNNNNN 8
739 * AAAA
740 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
741 * might become case 1 below case 2 below case 3 below
742 *
743 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
744 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
745 */
746struct vm_area_struct *vma_merge(struct mm_struct *mm,
747            struct vm_area_struct *prev, unsigned long addr,
748            unsigned long end, unsigned long vm_flags,
749                 struct anon_vma *anon_vma, struct file *file,
750            pgoff_t pgoff, struct mempolicy *policy)
751{
752    pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
753    struct vm_area_struct *area, *next;
754    int err;
755
756    /*
757     * We later require that vma->vm_flags == vm_flags,
758     * so this tests vma->vm_flags & VM_SPECIAL, too.
759     */
760    if (vm_flags & VM_SPECIAL)
761        return NULL;
762
763    if (prev)
764        next = prev->vm_next;
765    else
766        next = mm->mmap;
767    area = next;
768    if (next && next->vm_end == end) /* cases 6, 7, 8 */
769        next = next->vm_next;
770
771    /*
772     * Can it merge with the predecessor?
773     */
774    if (prev && prev->vm_end == addr &&
775              mpol_equal(vma_policy(prev), policy) &&
776            can_vma_merge_after(prev, vm_flags,
777                        anon_vma, file, pgoff)) {
778        /*
779         * OK, it can. Can we now merge in the successor as well?
780         */
781        if (next && end == next->vm_start &&
782                mpol_equal(policy, vma_policy(next)) &&
783                can_vma_merge_before(next, vm_flags,
784                    anon_vma, file, pgoff+pglen) &&
785                is_mergeable_anon_vma(prev->anon_vma,
786                              next->anon_vma)) {
787                            /* cases 1, 6 */
788            err = vma_adjust(prev, prev->vm_start,
789                next->vm_end, prev->vm_pgoff, NULL);
790        } else /* cases 2, 5, 7 */
791            err = vma_adjust(prev, prev->vm_start,
792                end, prev->vm_pgoff, NULL);
793        if (err)
794            return NULL;
795        return prev;
796    }
797
798    /*
799     * Can this new request be merged in front of next?
800     */
801    if (next && end == next->vm_start &&
802             mpol_equal(policy, vma_policy(next)) &&
803            can_vma_merge_before(next, vm_flags,
804                    anon_vma, file, pgoff+pglen)) {
805        if (prev && addr < prev->vm_end) /* case 4 */
806            err = vma_adjust(prev, prev->vm_start,
807                addr, prev->vm_pgoff, NULL);
808        else /* cases 3, 8 */
809            err = vma_adjust(area, addr, next->vm_end,
810                next->vm_pgoff - pglen, NULL);
811        if (err)
812            return NULL;
813        return area;
814    }
815
816    return NULL;
817}
818
819/*
820 * Rough compatbility check to quickly see if it's even worth looking
821 * at sharing an anon_vma.
822 *
823 * They need to have the same vm_file, and the flags can only differ
824 * in things that mprotect may change.
825 *
826 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
827 * we can merge the two vma's. For example, we refuse to merge a vma if
828 * there is a vm_ops->close() function, because that indicates that the
829 * driver is doing some kind of reference counting. But that doesn't
830 * really matter for the anon_vma sharing case.
831 */
832static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
833{
834    return a->vm_end == b->vm_start &&
835        mpol_equal(vma_policy(a), vma_policy(b)) &&
836        a->vm_file == b->vm_file &&
837        !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
838        b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
839}
840
841/*
842 * Do some basic sanity checking to see if we can re-use the anon_vma
843 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
844 * the same as 'old', the other will be the new one that is trying
845 * to share the anon_vma.
846 *
847 * NOTE! This runs with mm_sem held for reading, so it is possible that
848 * the anon_vma of 'old' is concurrently in the process of being set up
849 * by another page fault trying to merge _that_. But that's ok: if it
850 * is being set up, that automatically means that it will be a singleton
851 * acceptable for merging, so we can do all of this optimistically. But
852 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
853 *
854 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
855 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
856 * is to return an anon_vma that is "complex" due to having gone through
857 * a fork).
858 *
859 * We also make sure that the two vma's are compatible (adjacent,
860 * and with the same memory policies). That's all stable, even with just
861 * a read lock on the mm_sem.
862 */
863static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
864{
865    if (anon_vma_compatible(a, b)) {
866        struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
867
868        if (anon_vma && list_is_singular(&old->anon_vma_chain))
869            return anon_vma;
870    }
871    return NULL;
872}
873
874/*
875 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
876 * neighbouring vmas for a suitable anon_vma, before it goes off
877 * to allocate a new anon_vma. It checks because a repetitive
878 * sequence of mprotects and faults may otherwise lead to distinct
879 * anon_vmas being allocated, preventing vma merge in subsequent
880 * mprotect.
881 */
882struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
883{
884    struct anon_vma *anon_vma;
885    struct vm_area_struct *near;
886
887    near = vma->vm_next;
888    if (!near)
889        goto try_prev;
890
891    anon_vma = reusable_anon_vma(near, vma, near);
892    if (anon_vma)
893        return anon_vma;
894try_prev:
895    /*
896     * It is potentially slow to have to call find_vma_prev here.
897     * But it's only on the first write fault on the vma, not
898     * every time, and we could devise a way to avoid it later
899     * (e.g. stash info in next's anon_vma_node when assigning
900     * an anon_vma, or when trying vma_merge). Another time.
901     */
902    BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
903    if (!near)
904        goto none;
905
906    anon_vma = reusable_anon_vma(near, near, vma);
907    if (anon_vma)
908        return anon_vma;
909none:
910    /*
911     * There's no absolute need to look only at touching neighbours:
912     * we could search further afield for "compatible" anon_vmas.
913     * But it would probably just be a waste of time searching,
914     * or lead to too many vmas hanging off the same anon_vma.
915     * We're trying to allow mprotect remerging later on,
916     * not trying to minimize memory used for anon_vmas.
917     */
918    return NULL;
919}
920
921#ifdef CONFIG_PROC_FS
922void vm_stat_account(struct mm_struct *mm, unsigned long flags,
923                        struct file *file, long pages)
924{
925    const unsigned long stack_flags
926        = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
927
928    if (file) {
929        mm->shared_vm += pages;
930        if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
931            mm->exec_vm += pages;
932    } else if (flags & stack_flags)
933        mm->stack_vm += pages;
934    if (flags & (VM_RESERVED|VM_IO))
935        mm->reserved_vm += pages;
936}
937#endif /* CONFIG_PROC_FS */
938
939/*
940 * The caller must hold down_write(&current->mm->mmap_sem).
941 */
942
943unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
944            unsigned long len, unsigned long prot,
945            unsigned long flags, unsigned long pgoff)
946{
947    struct mm_struct * mm = current->mm;
948    struct inode *inode;
949    unsigned int vm_flags;
950    int error;
951    unsigned long reqprot = prot;
952
953    /*
954     * Does the application expect PROT_READ to imply PROT_EXEC?
955     *
956     * (the exception is when the underlying filesystem is noexec
957     * mounted, in which case we dont add PROT_EXEC.)
958     */
959    if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
960        if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
961            prot |= PROT_EXEC;
962
963    if (!len)
964        return -EINVAL;
965
966    if (!(flags & MAP_FIXED))
967        addr = round_hint_to_min(addr);
968
969    /* Careful about overflows.. */
970    len = PAGE_ALIGN(len);
971    if (!len)
972        return -ENOMEM;
973
974    /* offset overflow? */
975    if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
976               return -EOVERFLOW;
977
978    /* Too many mappings? */
979    if (mm->map_count > sysctl_max_map_count)
980        return -ENOMEM;
981
982    /* Obtain the address to map to. we verify (or select) it and ensure
983     * that it represents a valid section of the address space.
984     */
985    addr = get_unmapped_area(file, addr, len, pgoff, flags);
986    if (addr & ~PAGE_MASK)
987        return addr;
988
989    /* Do simple checking here so the lower-level routines won't have
990     * to. we assume access permissions have been handled by the open
991     * of the memory object, so we don't do any here.
992     */
993    vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
994            mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
995
996    if (flags & MAP_LOCKED)
997        if (!can_do_mlock())
998            return -EPERM;
999
1000    /* mlock MCL_FUTURE? */
1001    if (vm_flags & VM_LOCKED) {
1002        unsigned long locked, lock_limit;
1003        locked = len >> PAGE_SHIFT;
1004        locked += mm->locked_vm;
1005        lock_limit = rlimit(RLIMIT_MEMLOCK);
1006        lock_limit >>= PAGE_SHIFT;
1007        if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1008            return -EAGAIN;
1009    }
1010
1011    inode = file ? file->f_path.dentry->d_inode : NULL;
1012
1013    if (file) {
1014        switch (flags & MAP_TYPE) {
1015        case MAP_SHARED:
1016            if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1017                return -EACCES;
1018
1019            /*
1020             * Make sure we don't allow writing to an append-only
1021             * file..
1022             */
1023            if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1024                return -EACCES;
1025
1026            /*
1027             * Make sure there are no mandatory locks on the file.
1028             */
1029            if (locks_verify_locked(inode))
1030                return -EAGAIN;
1031
1032            vm_flags |= VM_SHARED | VM_MAYSHARE;
1033            if (!(file->f_mode & FMODE_WRITE))
1034                vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1035
1036            /* fall through */
1037        case MAP_PRIVATE:
1038            if (!(file->f_mode & FMODE_READ))
1039                return -EACCES;
1040            if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1041                if (vm_flags & VM_EXEC)
1042                    return -EPERM;
1043                vm_flags &= ~VM_MAYEXEC;
1044            }
1045
1046            if (!file->f_op || !file->f_op->mmap)
1047                return -ENODEV;
1048            break;
1049
1050        default:
1051            return -EINVAL;
1052        }
1053    } else {
1054        switch (flags & MAP_TYPE) {
1055        case MAP_SHARED:
1056            /*
1057             * Ignore pgoff.
1058             */
1059            pgoff = 0;
1060            vm_flags |= VM_SHARED | VM_MAYSHARE;
1061            break;
1062        case MAP_PRIVATE:
1063            /*
1064             * Set pgoff according to addr for anon_vma.
1065             */
1066            pgoff = addr >> PAGE_SHIFT;
1067            break;
1068        default:
1069            return -EINVAL;
1070        }
1071    }
1072
1073    error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1074    if (error)
1075        return error;
1076
1077    return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1078}
1079EXPORT_SYMBOL(do_mmap_pgoff);
1080
1081SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1082        unsigned long, prot, unsigned long, flags,
1083        unsigned long, fd, unsigned long, pgoff)
1084{
1085    struct file *file = NULL;
1086    unsigned long retval = -EBADF;
1087
1088    if (!(flags & MAP_ANONYMOUS)) {
1089        if (unlikely(flags & MAP_HUGETLB))
1090            return -EINVAL;
1091        file = fget(fd);
1092        if (!file)
1093            goto out;
1094    } else if (flags & MAP_HUGETLB) {
1095        struct user_struct *user = NULL;
1096        /*
1097         * VM_NORESERVE is used because the reservations will be
1098         * taken when vm_ops->mmap() is called
1099         * A dummy user value is used because we are not locking
1100         * memory so no accounting is necessary
1101         */
1102        len = ALIGN(len, huge_page_size(&default_hstate));
1103        file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1104                        &user, HUGETLB_ANONHUGE_INODE);
1105        if (IS_ERR(file))
1106            return PTR_ERR(file);
1107    }
1108
1109    flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1110
1111    down_write(&current->mm->mmap_sem);
1112    retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1113    up_write(&current->mm->mmap_sem);
1114
1115    if (file)
1116        fput(file);
1117out:
1118    return retval;
1119}
1120
1121#ifdef __ARCH_WANT_SYS_OLD_MMAP
1122struct mmap_arg_struct {
1123    unsigned long addr;
1124    unsigned long len;
1125    unsigned long prot;
1126    unsigned long flags;
1127    unsigned long fd;
1128    unsigned long offset;
1129};
1130
1131SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1132{
1133    struct mmap_arg_struct a;
1134
1135    if (copy_from_user(&a, arg, sizeof(a)))
1136        return -EFAULT;
1137    if (a.offset & ~PAGE_MASK)
1138        return -EINVAL;
1139
1140    return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1141                  a.offset >> PAGE_SHIFT);
1142}
1143#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1144
1145/*
1146 * Some shared mappigns will want the pages marked read-only
1147 * to track write events. If so, we'll downgrade vm_page_prot
1148 * to the private version (using protection_map[] without the
1149 * VM_SHARED bit).
1150 */
1151int vma_wants_writenotify(struct vm_area_struct *vma)
1152{
1153    unsigned int vm_flags = vma->vm_flags;
1154
1155    /* If it was private or non-writable, the write bit is already clear */
1156    if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1157        return 0;
1158
1159    /* The backer wishes to know when pages are first written to? */
1160    if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1161        return 1;
1162
1163    /* The open routine did something to the protections already? */
1164    if (pgprot_val(vma->vm_page_prot) !=
1165        pgprot_val(vm_get_page_prot(vm_flags)))
1166        return 0;
1167
1168    /* Specialty mapping? */
1169    if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1170        return 0;
1171
1172    /* Can the mapping track the dirty pages? */
1173    return vma->vm_file && vma->vm_file->f_mapping &&
1174        mapping_cap_account_dirty(vma->vm_file->f_mapping);
1175}
1176
1177/*
1178 * We account for memory if it's a private writeable mapping,
1179 * not hugepages and VM_NORESERVE wasn't set.
1180 */
1181static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
1182{
1183    /*
1184     * hugetlb has its own accounting separate from the core VM
1185     * VM_HUGETLB may not be set yet so we cannot check for that flag.
1186     */
1187    if (file && is_file_hugepages(file))
1188        return 0;
1189
1190    return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1191}
1192
1193unsigned long mmap_region(struct file *file, unsigned long addr,
1194              unsigned long len, unsigned long flags,
1195              unsigned int vm_flags, unsigned long pgoff)
1196{
1197    struct mm_struct *mm = current->mm;
1198    struct vm_area_struct *vma, *prev;
1199    int correct_wcount = 0;
1200    int error;
1201    struct rb_node **rb_link, *rb_parent;
1202    unsigned long charged = 0;
1203    struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1204
1205    /* Clear old maps */
1206    error = -ENOMEM;
1207munmap_back:
1208    vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1209    if (vma && vma->vm_start < addr + len) {
1210        if (do_munmap(mm, addr, len))
1211            return -ENOMEM;
1212        goto munmap_back;
1213    }
1214
1215    /* Check against address space limit. */
1216    if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1217        return -ENOMEM;
1218
1219    /*
1220     * Set 'VM_NORESERVE' if we should not account for the
1221     * memory use of this mapping.
1222     */
1223    if ((flags & MAP_NORESERVE)) {
1224        /* We honor MAP_NORESERVE if allowed to overcommit */
1225        if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1226            vm_flags |= VM_NORESERVE;
1227
1228        /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1229        if (file && is_file_hugepages(file))
1230            vm_flags |= VM_NORESERVE;
1231    }
1232
1233    /*
1234     * Private writable mapping: check memory availability
1235     */
1236    if (accountable_mapping(file, vm_flags)) {
1237        charged = len >> PAGE_SHIFT;
1238        if (security_vm_enough_memory(charged))
1239            return -ENOMEM;
1240        vm_flags |= VM_ACCOUNT;
1241    }
1242
1243    /*
1244     * Can we just expand an old mapping?
1245     */
1246    vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1247    if (vma)
1248        goto out;
1249
1250    /*
1251     * Determine the object being mapped and call the appropriate
1252     * specific mapper. the address has already been validated, but
1253     * not unmapped, but the maps are removed from the list.
1254     */
1255    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1256    if (!vma) {
1257        error = -ENOMEM;
1258        goto unacct_error;
1259    }
1260
1261    vma->vm_mm = mm;
1262    vma->vm_start = addr;
1263    vma->vm_end = addr + len;
1264    vma->vm_flags = vm_flags;
1265    vma->vm_page_prot = vm_get_page_prot(vm_flags);
1266    vma->vm_pgoff = pgoff;
1267    INIT_LIST_HEAD(&vma->anon_vma_chain);
1268
1269    if (file) {
1270        error = -EINVAL;
1271        if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1272            goto free_vma;
1273        if (vm_flags & VM_DENYWRITE) {
1274            error = deny_write_access(file);
1275            if (error)
1276                goto free_vma;
1277            correct_wcount = 1;
1278        }
1279        vma->vm_file = file;
1280        get_file(file);
1281        error = file->f_op->mmap(file, vma);
1282        if (error)
1283            goto unmap_and_free_vma;
1284        if (vm_flags & VM_EXECUTABLE)
1285            added_exe_file_vma(mm);
1286
1287        /* Can addr have changed??
1288         *
1289         * Answer: Yes, several device drivers can do it in their
1290         * f_op->mmap method. -DaveM
1291         */
1292        addr = vma->vm_start;
1293        pgoff = vma->vm_pgoff;
1294        vm_flags = vma->vm_flags;
1295    } else if (vm_flags & VM_SHARED) {
1296        error = shmem_zero_setup(vma);
1297        if (error)
1298            goto free_vma;
1299    }
1300
1301    if (vma_wants_writenotify(vma)) {
1302        pgprot_t pprot = vma->vm_page_prot;
1303
1304        /* Can vma->vm_page_prot have changed??
1305         *
1306         * Answer: Yes, drivers may have changed it in their
1307         * f_op->mmap method.
1308         *
1309         * Ensures that vmas marked as uncached stay that way.
1310         */
1311        vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1312        if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1313            vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1314    }
1315
1316    vma_link(mm, vma, prev, rb_link, rb_parent);
1317    file = vma->vm_file;
1318
1319    /* Once vma denies write, undo our temporary denial count */
1320    if (correct_wcount)
1321        atomic_inc(&inode->i_writecount);
1322out:
1323    perf_event_mmap(vma);
1324
1325    mm->total_vm += len >> PAGE_SHIFT;
1326    vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1327    if (vm_flags & VM_LOCKED) {
1328        if (!mlock_vma_pages_range(vma, addr, addr + len))
1329            mm->locked_vm += (len >> PAGE_SHIFT);
1330    } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1331        make_pages_present(addr, addr + len);
1332    return addr;
1333
1334unmap_and_free_vma:
1335    if (correct_wcount)
1336        atomic_inc(&inode->i_writecount);
1337    vma->vm_file = NULL;
1338    fput(file);
1339
1340    /* Undo any partial mapping done by a device driver. */
1341    unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1342    charged = 0;
1343free_vma:
1344    kmem_cache_free(vm_area_cachep, vma);
1345unacct_error:
1346    if (charged)
1347        vm_unacct_memory(charged);
1348    return error;
1349}
1350
1351/* Get an address range which is currently unmapped.
1352 * For shmat() with addr=0.
1353 *
1354 * Ugly calling convention alert:
1355 * Return value with the low bits set means error value,
1356 * ie
1357 * if (ret & ~PAGE_MASK)
1358 * error = ret;
1359 *
1360 * This function "knows" that -ENOMEM has the bits set.
1361 */
1362#ifndef HAVE_ARCH_UNMAPPED_AREA
1363unsigned long
1364arch_get_unmapped_area(struct file *filp, unsigned long addr,
1365        unsigned long len, unsigned long pgoff, unsigned long flags)
1366{
1367    struct mm_struct *mm = current->mm;
1368    struct vm_area_struct *vma;
1369    unsigned long start_addr;
1370
1371    if (len > TASK_SIZE)
1372        return -ENOMEM;
1373
1374    if (flags & MAP_FIXED)
1375        return addr;
1376
1377    if (addr) {
1378        addr = PAGE_ALIGN(addr);
1379        vma = find_vma(mm, addr);
1380        if (TASK_SIZE - len >= addr &&
1381            (!vma || addr + len <= vma->vm_start))
1382            return addr;
1383    }
1384    if (len > mm->cached_hole_size) {
1385            start_addr = addr = mm->free_area_cache;
1386    } else {
1387            start_addr = addr = TASK_UNMAPPED_BASE;
1388            mm->cached_hole_size = 0;
1389    }
1390
1391full_search:
1392    for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1393        /* At this point: (!vma || addr < vma->vm_end). */
1394        if (TASK_SIZE - len < addr) {
1395            /*
1396             * Start a new search - just in case we missed
1397             * some holes.
1398             */
1399            if (start_addr != TASK_UNMAPPED_BASE) {
1400                addr = TASK_UNMAPPED_BASE;
1401                    start_addr = addr;
1402                mm->cached_hole_size = 0;
1403                goto full_search;
1404            }
1405            return -ENOMEM;
1406        }
1407        if (!vma || addr + len <= vma->vm_start) {
1408            /*
1409             * Remember the place where we stopped the search:
1410             */
1411            mm->free_area_cache = addr + len;
1412            return addr;
1413        }
1414        if (addr + mm->cached_hole_size < vma->vm_start)
1415                mm->cached_hole_size = vma->vm_start - addr;
1416        addr = vma->vm_end;
1417    }
1418}
1419#endif
1420
1421void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1422{
1423    /*
1424     * Is this a new hole at the lowest possible address?
1425     */
1426    if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1427        mm->free_area_cache = addr;
1428        mm->cached_hole_size = ~0UL;
1429    }
1430}
1431
1432/*
1433 * This mmap-allocator allocates new areas top-down from below the
1434 * stack's low limit (the base):
1435 */
1436#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1437unsigned long
1438arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1439              const unsigned long len, const unsigned long pgoff,
1440              const unsigned long flags)
1441{
1442    struct vm_area_struct *vma;
1443    struct mm_struct *mm = current->mm;
1444    unsigned long addr = addr0;
1445
1446    /* requested length too big for entire address space */
1447    if (len > TASK_SIZE)
1448        return -ENOMEM;
1449
1450    if (flags & MAP_FIXED)
1451        return addr;
1452
1453    /* requesting a specific address */
1454    if (addr) {
1455        addr = PAGE_ALIGN(addr);
1456        vma = find_vma(mm, addr);
1457        if (TASK_SIZE - len >= addr &&
1458                (!vma || addr + len <= vma->vm_start))
1459            return addr;
1460    }
1461
1462    /* check if free_area_cache is useful for us */
1463    if (len <= mm->cached_hole_size) {
1464             mm->cached_hole_size = 0;
1465         mm->free_area_cache = mm->mmap_base;
1466     }
1467
1468    /* either no address requested or can't fit in requested address hole */
1469    addr = mm->free_area_cache;
1470
1471    /* make sure it can fit in the remaining address space */
1472    if (addr > len) {
1473        vma = find_vma(mm, addr-len);
1474        if (!vma || addr <= vma->vm_start)
1475            /* remember the address as a hint for next time */
1476            return (mm->free_area_cache = addr-len);
1477    }
1478
1479    if (mm->mmap_base < len)
1480        goto bottomup;
1481
1482    addr = mm->mmap_base-len;
1483
1484    do {
1485        /*
1486         * Lookup failure means no vma is above this address,
1487         * else if new region fits below vma->vm_start,
1488         * return with success:
1489         */
1490        vma = find_vma(mm, addr);
1491        if (!vma || addr+len <= vma->vm_start)
1492            /* remember the address as a hint for next time */
1493            return (mm->free_area_cache = addr);
1494
1495         /* remember the largest hole we saw so far */
1496         if (addr + mm->cached_hole_size < vma->vm_start)
1497                 mm->cached_hole_size = vma->vm_start - addr;
1498
1499        /* try just below the current vma->vm_start */
1500        addr = vma->vm_start-len;
1501    } while (len < vma->vm_start);
1502
1503bottomup:
1504    /*
1505     * A failed mmap() very likely causes application failure,
1506     * so fall back to the bottom-up function here. This scenario
1507     * can happen with large stack limits and large mmap()
1508     * allocations.
1509     */
1510    mm->cached_hole_size = ~0UL;
1511      mm->free_area_cache = TASK_UNMAPPED_BASE;
1512    addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1513    /*
1514     * Restore the topdown base:
1515     */
1516    mm->free_area_cache = mm->mmap_base;
1517    mm->cached_hole_size = ~0UL;
1518
1519    return addr;
1520}
1521#endif
1522
1523void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1524{
1525    /*
1526     * Is this a new hole at the highest possible address?
1527     */
1528    if (addr > mm->free_area_cache)
1529        mm->free_area_cache = addr;
1530
1531    /* dont allow allocations above current base */
1532    if (mm->free_area_cache > mm->mmap_base)
1533        mm->free_area_cache = mm->mmap_base;
1534}
1535
1536unsigned long
1537get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1538        unsigned long pgoff, unsigned long flags)
1539{
1540    unsigned long (*get_area)(struct file *, unsigned long,
1541                  unsigned long, unsigned long, unsigned long);
1542
1543    unsigned long error = arch_mmap_check(addr, len, flags);
1544    if (error)
1545        return error;
1546
1547    /* Careful about overflows.. */
1548    if (len > TASK_SIZE)
1549        return -ENOMEM;
1550
1551    get_area = current->mm->get_unmapped_area;
1552    if (file && file->f_op && file->f_op->get_unmapped_area)
1553        get_area = file->f_op->get_unmapped_area;
1554    addr = get_area(file, addr, len, pgoff, flags);
1555    if (IS_ERR_VALUE(addr))
1556        return addr;
1557
1558    if (addr > TASK_SIZE - len)
1559        return -ENOMEM;
1560    if (addr & ~PAGE_MASK)
1561        return -EINVAL;
1562
1563    return arch_rebalance_pgtables(addr, len);
1564}
1565
1566EXPORT_SYMBOL(get_unmapped_area);
1567
1568/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1569struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1570{
1571    struct vm_area_struct *vma = NULL;
1572
1573    if (mm) {
1574        /* Check the cache first. */
1575        /* (Cache hit rate is typically around 35%.) */
1576        vma = mm->mmap_cache;
1577        if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1578            struct rb_node * rb_node;
1579
1580            rb_node = mm->mm_rb.rb_node;
1581            vma = NULL;
1582
1583            while (rb_node) {
1584                struct vm_area_struct * vma_tmp;
1585
1586                vma_tmp = rb_entry(rb_node,
1587                        struct vm_area_struct, vm_rb);
1588
1589                if (vma_tmp->vm_end > addr) {
1590                    vma = vma_tmp;
1591                    if (vma_tmp->vm_start <= addr)
1592                        break;
1593                    rb_node = rb_node->rb_left;
1594                } else
1595                    rb_node = rb_node->rb_right;
1596            }
1597            if (vma)
1598                mm->mmap_cache = vma;
1599        }
1600    }
1601    return vma;
1602}
1603
1604EXPORT_SYMBOL(find_vma);
1605
1606/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1607struct vm_area_struct *
1608find_vma_prev(struct mm_struct *mm, unsigned long addr,
1609            struct vm_area_struct **pprev)
1610{
1611    struct vm_area_struct *vma = NULL, *prev = NULL;
1612    struct rb_node *rb_node;
1613    if (!mm)
1614        goto out;
1615
1616    /* Guard against addr being lower than the first VMA */
1617    vma = mm->mmap;
1618
1619    /* Go through the RB tree quickly. */
1620    rb_node = mm->mm_rb.rb_node;
1621
1622    while (rb_node) {
1623        struct vm_area_struct *vma_tmp;
1624        vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1625
1626        if (addr < vma_tmp->vm_end) {
1627            rb_node = rb_node->rb_left;
1628        } else {
1629            prev = vma_tmp;
1630            if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1631                break;
1632            rb_node = rb_node->rb_right;
1633        }
1634    }
1635
1636out:
1637    *pprev = prev;
1638    return prev ? prev->vm_next : vma;
1639}
1640
1641/*
1642 * Verify that the stack growth is acceptable and
1643 * update accounting. This is shared with both the
1644 * grow-up and grow-down cases.
1645 */
1646static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1647{
1648    struct mm_struct *mm = vma->vm_mm;
1649    struct rlimit *rlim = current->signal->rlim;
1650    unsigned long new_start;
1651
1652    /* address space limit tests */
1653    if (!may_expand_vm(mm, grow))
1654        return -ENOMEM;
1655
1656    /* Stack limit test */
1657    if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1658        return -ENOMEM;
1659
1660    /* mlock limit tests */
1661    if (vma->vm_flags & VM_LOCKED) {
1662        unsigned long locked;
1663        unsigned long limit;
1664        locked = mm->locked_vm + grow;
1665        limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1666        limit >>= PAGE_SHIFT;
1667        if (locked > limit && !capable(CAP_IPC_LOCK))
1668            return -ENOMEM;
1669    }
1670
1671    /* Check to ensure the stack will not grow into a hugetlb-only region */
1672    new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1673            vma->vm_end - size;
1674    if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1675        return -EFAULT;
1676
1677    /*
1678     * Overcommit.. This must be the final test, as it will
1679     * update security statistics.
1680     */
1681    if (security_vm_enough_memory_mm(mm, grow))
1682        return -ENOMEM;
1683
1684    /* Ok, everything looks good - let it rip */
1685    mm->total_vm += grow;
1686    if (vma->vm_flags & VM_LOCKED)
1687        mm->locked_vm += grow;
1688    vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1689    return 0;
1690}
1691
1692#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1693/*
1694 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1695 * vma is the last one with address > vma->vm_end. Have to extend vma.
1696 */
1697#ifndef CONFIG_IA64
1698static
1699#endif
1700int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1701{
1702    int error;
1703
1704    if (!(vma->vm_flags & VM_GROWSUP))
1705        return -EFAULT;
1706
1707    /*
1708     * We must make sure the anon_vma is allocated
1709     * so that the anon_vma locking is not a noop.
1710     */
1711    if (unlikely(anon_vma_prepare(vma)))
1712        return -ENOMEM;
1713    anon_vma_lock(vma);
1714
1715    /*
1716     * vma->vm_start/vm_end cannot change under us because the caller
1717     * is required to hold the mmap_sem in read mode. We need the
1718     * anon_vma lock to serialize against concurrent expand_stacks.
1719     * Also guard against wrapping around to address 0.
1720     */
1721    if (address < PAGE_ALIGN(address+4))
1722        address = PAGE_ALIGN(address+4);
1723    else {
1724        anon_vma_unlock(vma);
1725        return -ENOMEM;
1726    }
1727    error = 0;
1728
1729    /* Somebody else might have raced and expanded it already */
1730    if (address > vma->vm_end) {
1731        unsigned long size, grow;
1732
1733        size = address - vma->vm_start;
1734        grow = (address - vma->vm_end) >> PAGE_SHIFT;
1735
1736        error = acct_stack_growth(vma, size, grow);
1737        if (!error)
1738            vma->vm_end = address;
1739    }
1740    anon_vma_unlock(vma);
1741    return error;
1742}
1743#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1744
1745/*
1746 * vma is the first one with address < vma->vm_start. Have to extend vma.
1747 */
1748static int expand_downwards(struct vm_area_struct *vma,
1749                   unsigned long address)
1750{
1751    int error;
1752
1753    /*
1754     * We must make sure the anon_vma is allocated
1755     * so that the anon_vma locking is not a noop.
1756     */
1757    if (unlikely(anon_vma_prepare(vma)))
1758        return -ENOMEM;
1759
1760    address &= PAGE_MASK;
1761    error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1762    if (error)
1763        return error;
1764
1765    anon_vma_lock(vma);
1766
1767    /*
1768     * vma->vm_start/vm_end cannot change under us because the caller
1769     * is required to hold the mmap_sem in read mode. We need the
1770     * anon_vma lock to serialize against concurrent expand_stacks.
1771     */
1772
1773    /* Somebody else might have raced and expanded it already */
1774    if (address < vma->vm_start) {
1775        unsigned long size, grow;
1776
1777        size = vma->vm_end - address;
1778        grow = (vma->vm_start - address) >> PAGE_SHIFT;
1779
1780        error = acct_stack_growth(vma, size, grow);
1781        if (!error) {
1782            vma->vm_start = address;
1783            vma->vm_pgoff -= grow;
1784        }
1785    }
1786    anon_vma_unlock(vma);
1787    return error;
1788}
1789
1790int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1791{
1792    return expand_downwards(vma, address);
1793}
1794
1795#ifdef CONFIG_STACK_GROWSUP
1796int expand_stack(struct vm_area_struct *vma, unsigned long address)
1797{
1798    return expand_upwards(vma, address);
1799}
1800
1801struct vm_area_struct *
1802find_extend_vma(struct mm_struct *mm, unsigned long addr)
1803{
1804    struct vm_area_struct *vma, *prev;
1805
1806    addr &= PAGE_MASK;
1807    vma = find_vma_prev(mm, addr, &prev);
1808    if (vma && (vma->vm_start <= addr))
1809        return vma;
1810    if (!prev || expand_stack(prev, addr))
1811        return NULL;
1812    if (prev->vm_flags & VM_LOCKED) {
1813        mlock_vma_pages_range(prev, addr, prev->vm_end);
1814    }
1815    return prev;
1816}
1817#else
1818int expand_stack(struct vm_area_struct *vma, unsigned long address)
1819{
1820    return expand_downwards(vma, address);
1821}
1822
1823struct vm_area_struct *
1824find_extend_vma(struct mm_struct * mm, unsigned long addr)
1825{
1826    struct vm_area_struct * vma;
1827    unsigned long start;
1828
1829    addr &= PAGE_MASK;
1830    vma = find_vma(mm,addr);
1831    if (!vma)
1832        return NULL;
1833    if (vma->vm_start <= addr)
1834        return vma;
1835    if (!(vma->vm_flags & VM_GROWSDOWN))
1836        return NULL;
1837    start = vma->vm_start;
1838    if (expand_stack(vma, addr))
1839        return NULL;
1840    if (vma->vm_flags & VM_LOCKED) {
1841        mlock_vma_pages_range(vma, addr, start);
1842    }
1843    return vma;
1844}
1845#endif
1846
1847/*
1848 * Ok - we have the memory areas we should free on the vma list,
1849 * so release them, and do the vma updates.
1850 *
1851 * Called with the mm semaphore held.
1852 */
1853static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1854{
1855    /* Update high watermark before we lower total_vm */
1856    update_hiwater_vm(mm);
1857    do {
1858        long nrpages = vma_pages(vma);
1859
1860        mm->total_vm -= nrpages;
1861        vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1862        vma = remove_vma(vma);
1863    } while (vma);
1864    validate_mm(mm);
1865}
1866
1867/*
1868 * Get rid of page table information in the indicated region.
1869 *
1870 * Called with the mm semaphore held.
1871 */
1872static void unmap_region(struct mm_struct *mm,
1873        struct vm_area_struct *vma, struct vm_area_struct *prev,
1874        unsigned long start, unsigned long end)
1875{
1876    struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1877    struct mmu_gather *tlb;
1878    unsigned long nr_accounted = 0;
1879
1880    lru_add_drain();
1881    tlb = tlb_gather_mmu(mm, 0);
1882    update_hiwater_rss(mm);
1883    unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1884    vm_unacct_memory(nr_accounted);
1885    free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1886                 next? next->vm_start: 0);
1887    tlb_finish_mmu(tlb, start, end);
1888}
1889
1890/*
1891 * Create a list of vma's touched by the unmap, removing them from the mm's
1892 * vma list as we go..
1893 */
1894static void
1895detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1896    struct vm_area_struct *prev, unsigned long end)
1897{
1898    struct vm_area_struct **insertion_point;
1899    struct vm_area_struct *tail_vma = NULL;
1900    unsigned long addr;
1901
1902    insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1903    do {
1904        rb_erase(&vma->vm_rb, &mm->mm_rb);
1905        mm->map_count--;
1906        tail_vma = vma;
1907        vma = vma->vm_next;
1908    } while (vma && vma->vm_start < end);
1909    *insertion_point = vma;
1910    tail_vma->vm_next = NULL;
1911    if (mm->unmap_area == arch_unmap_area)
1912        addr = prev ? prev->vm_end : mm->mmap_base;
1913    else
1914        addr = vma ? vma->vm_start : mm->mmap_base;
1915    mm->unmap_area(mm, addr);
1916    mm->mmap_cache = NULL; /* Kill the cache. */
1917}
1918
1919/*
1920 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1921 * munmap path where it doesn't make sense to fail.
1922 */
1923static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1924          unsigned long addr, int new_below)
1925{
1926    struct mempolicy *pol;
1927    struct vm_area_struct *new;
1928    int err = -ENOMEM;
1929
1930    if (is_vm_hugetlb_page(vma) && (addr &
1931                    ~(huge_page_mask(hstate_vma(vma)))))
1932        return -EINVAL;
1933
1934    new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1935    if (!new)
1936        goto out_err;
1937
1938    /* most fields are the same, copy all, and then fixup */
1939    *new = *vma;
1940
1941    INIT_LIST_HEAD(&new->anon_vma_chain);
1942
1943    if (new_below)
1944        new->vm_end = addr;
1945    else {
1946        new->vm_start = addr;
1947        new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1948    }
1949
1950    pol = mpol_dup(vma_policy(vma));
1951    if (IS_ERR(pol)) {
1952        err = PTR_ERR(pol);
1953        goto out_free_vma;
1954    }
1955    vma_set_policy(new, pol);
1956
1957    if (anon_vma_clone(new, vma))
1958        goto out_free_mpol;
1959
1960    if (new->vm_file) {
1961        get_file(new->vm_file);
1962        if (vma->vm_flags & VM_EXECUTABLE)
1963            added_exe_file_vma(mm);
1964    }
1965
1966    if (new->vm_ops && new->vm_ops->open)
1967        new->vm_ops->open(new);
1968
1969    if (new_below)
1970        err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1971            ((addr - new->vm_start) >> PAGE_SHIFT), new);
1972    else
1973        err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1974
1975    /* Success. */
1976    if (!err)
1977        return 0;
1978
1979    /* Clean everything up if vma_adjust failed. */
1980    if (new->vm_ops && new->vm_ops->close)
1981        new->vm_ops->close(new);
1982    if (new->vm_file) {
1983        if (vma->vm_flags & VM_EXECUTABLE)
1984            removed_exe_file_vma(mm);
1985        fput(new->vm_file);
1986    }
1987 out_free_mpol:
1988    mpol_put(pol);
1989 out_free_vma:
1990    kmem_cache_free(vm_area_cachep, new);
1991 out_err:
1992    return err;
1993}
1994
1995/*
1996 * Split a vma into two pieces at address 'addr', a new vma is allocated
1997 * either for the first part or the tail.
1998 */
1999int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2000          unsigned long addr, int new_below)
2001{
2002    if (mm->map_count >= sysctl_max_map_count)
2003        return -ENOMEM;
2004
2005    return __split_vma(mm, vma, addr, new_below);
2006}
2007
2008/* Munmap is split into 2 main parts -- this part which finds
2009 * what needs doing, and the areas themselves, which do the
2010 * work. This now handles partial unmappings.
2011 * Jeremy Fitzhardinge <jeremy@goop.org>
2012 */
2013int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2014{
2015    unsigned long end;
2016    struct vm_area_struct *vma, *prev, *last;
2017
2018    if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2019        return -EINVAL;
2020
2021    if ((len = PAGE_ALIGN(len)) == 0)
2022        return -EINVAL;
2023
2024    /* Find the first overlapping VMA */
2025    vma = find_vma_prev(mm, start, &prev);
2026    if (!vma)
2027        return 0;
2028    /* we have start < vma->vm_end */
2029
2030    /* if it doesn't overlap, we have nothing.. */
2031    end = start + len;
2032    if (vma->vm_start >= end)
2033        return 0;
2034
2035    /*
2036     * If we need to split any vma, do it now to save pain later.
2037     *
2038     * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2039     * unmapped vm_area_struct will remain in use: so lower split_vma
2040     * places tmp vma above, and higher split_vma places tmp vma below.
2041     */
2042    if (start > vma->vm_start) {
2043        int error;
2044
2045        /*
2046         * Make sure that map_count on return from munmap() will
2047         * not exceed its limit; but let map_count go just above
2048         * its limit temporarily, to help free resources as expected.
2049         */
2050        if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2051            return -ENOMEM;
2052
2053        error = __split_vma(mm, vma, start, 0);
2054        if (error)
2055            return error;
2056        prev = vma;
2057    }
2058
2059    /* Does it split the last one? */
2060    last = find_vma(mm, end);
2061    if (last && end > last->vm_start) {
2062        int error = __split_vma(mm, last, end, 1);
2063        if (error)
2064            return error;
2065    }
2066    vma = prev? prev->vm_next: mm->mmap;
2067
2068    /*
2069     * unlock any mlock()ed ranges before detaching vmas
2070     */
2071    if (mm->locked_vm) {
2072        struct vm_area_struct *tmp = vma;
2073        while (tmp && tmp->vm_start < end) {
2074            if (tmp->vm_flags & VM_LOCKED) {
2075                mm->locked_vm -= vma_pages(tmp);
2076                munlock_vma_pages_all(tmp);
2077            }
2078            tmp = tmp->vm_next;
2079        }
2080    }
2081
2082    /*
2083     * Remove the vma's, and unmap the actual pages
2084     */
2085    detach_vmas_to_be_unmapped(mm, vma, prev, end);
2086    unmap_region(mm, vma, prev, start, end);
2087
2088    /* Fix up all other VM information */
2089    remove_vma_list(mm, vma);
2090
2091    return 0;
2092}
2093
2094EXPORT_SYMBOL(do_munmap);
2095
2096SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2097{
2098    int ret;
2099    struct mm_struct *mm = current->mm;
2100
2101    profile_munmap(addr);
2102
2103    down_write(&mm->mmap_sem);
2104    ret = do_munmap(mm, addr, len);
2105    up_write(&mm->mmap_sem);
2106    return ret;
2107}
2108
2109static inline void verify_mm_writelocked(struct mm_struct *mm)
2110{
2111#ifdef CONFIG_DEBUG_VM
2112    if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2113        WARN_ON(1);
2114        up_read(&mm->mmap_sem);
2115    }
2116#endif
2117}
2118
2119/*
2120 * this is really a simplified "do_mmap". it only handles
2121 * anonymous maps. eventually we may be able to do some
2122 * brk-specific accounting here.
2123 */
2124unsigned long do_brk(unsigned long addr, unsigned long len)
2125{
2126    struct mm_struct * mm = current->mm;
2127    struct vm_area_struct * vma, * prev;
2128    unsigned long flags;
2129    struct rb_node ** rb_link, * rb_parent;
2130    pgoff_t pgoff = addr >> PAGE_SHIFT;
2131    int error;
2132
2133    len = PAGE_ALIGN(len);
2134    if (!len)
2135        return addr;
2136
2137    error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2138    if (error)
2139        return error;
2140
2141    flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2142
2143    error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2144    if (error & ~PAGE_MASK)
2145        return error;
2146
2147    /*
2148     * mlock MCL_FUTURE?
2149     */
2150    if (mm->def_flags & VM_LOCKED) {
2151        unsigned long locked, lock_limit;
2152        locked = len >> PAGE_SHIFT;
2153        locked += mm->locked_vm;
2154        lock_limit = rlimit(RLIMIT_MEMLOCK);
2155        lock_limit >>= PAGE_SHIFT;
2156        if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2157            return -EAGAIN;
2158    }
2159
2160    /*
2161     * mm->mmap_sem is required to protect against another thread
2162     * changing the mappings in case we sleep.
2163     */
2164    verify_mm_writelocked(mm);
2165
2166    /*
2167     * Clear old maps. this also does some error checking for us
2168     */
2169 munmap_back:
2170    vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2171    if (vma && vma->vm_start < addr + len) {
2172        if (do_munmap(mm, addr, len))
2173            return -ENOMEM;
2174        goto munmap_back;
2175    }
2176
2177    /* Check against address space limits *after* clearing old maps... */
2178    if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2179        return -ENOMEM;
2180
2181    if (mm->map_count > sysctl_max_map_count)
2182        return -ENOMEM;
2183
2184    if (security_vm_enough_memory(len >> PAGE_SHIFT))
2185        return -ENOMEM;
2186
2187    /* Can we just expand an old private anonymous mapping? */
2188    vma = vma_merge(mm, prev, addr, addr + len, flags,
2189                    NULL, NULL, pgoff, NULL);
2190    if (vma)
2191        goto out;
2192
2193    /*
2194     * create a vma struct for an anonymous mapping
2195     */
2196    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2197    if (!vma) {
2198        vm_unacct_memory(len >> PAGE_SHIFT);
2199        return -ENOMEM;
2200    }
2201
2202    INIT_LIST_HEAD(&vma->anon_vma_chain);
2203    vma->vm_mm = mm;
2204    vma->vm_start = addr;
2205    vma->vm_end = addr + len;
2206    vma->vm_pgoff = pgoff;
2207    vma->vm_flags = flags;
2208    vma->vm_page_prot = vm_get_page_prot(flags);
2209    vma_link(mm, vma, prev, rb_link, rb_parent);
2210out:
2211    mm->total_vm += len >> PAGE_SHIFT;
2212    if (flags & VM_LOCKED) {
2213        if (!mlock_vma_pages_range(vma, addr, addr + len))
2214            mm->locked_vm += (len >> PAGE_SHIFT);
2215    }
2216    return addr;
2217}
2218
2219EXPORT_SYMBOL(do_brk);
2220
2221/* Release all mmaps. */
2222void exit_mmap(struct mm_struct *mm)
2223{
2224    struct mmu_gather *tlb;
2225    struct vm_area_struct *vma;
2226    unsigned long nr_accounted = 0;
2227    unsigned long end;
2228
2229    /* mm's last user has gone, and its about to be pulled down */
2230    mmu_notifier_release(mm);
2231
2232    if (mm->locked_vm) {
2233        vma = mm->mmap;
2234        while (vma) {
2235            if (vma->vm_flags & VM_LOCKED)
2236                munlock_vma_pages_all(vma);
2237            vma = vma->vm_next;
2238        }
2239    }
2240
2241    arch_exit_mmap(mm);
2242
2243    vma = mm->mmap;
2244    if (!vma) /* Can happen if dup_mmap() received an OOM */
2245        return;
2246
2247    lru_add_drain();
2248    flush_cache_mm(mm);
2249    tlb = tlb_gather_mmu(mm, 1);
2250    /* update_hiwater_rss(mm) here? but nobody should be looking */
2251    /* Use -1 here to ensure all VMAs in the mm are unmapped */
2252    end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2253    vm_unacct_memory(nr_accounted);
2254
2255    free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2256    tlb_finish_mmu(tlb, 0, end);
2257
2258    /*
2259     * Walk the list again, actually closing and freeing it,
2260     * with preemption enabled, without holding any MM locks.
2261     */
2262    while (vma)
2263        vma = remove_vma(vma);
2264
2265    BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2266}
2267
2268/* Insert vm structure into process list sorted by address
2269 * and into the inode's i_mmap tree. If vm_file is non-NULL
2270 * then i_mmap_lock is taken here.
2271 */
2272int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2273{
2274    struct vm_area_struct * __vma, * prev;
2275    struct rb_node ** rb_link, * rb_parent;
2276
2277    /*
2278     * The vm_pgoff of a purely anonymous vma should be irrelevant
2279     * until its first write fault, when page's anon_vma and index
2280     * are set. But now set the vm_pgoff it will almost certainly
2281     * end up with (unless mremap moves it elsewhere before that
2282     * first wfault), so /proc/pid/maps tells a consistent story.
2283     *
2284     * By setting it to reflect the virtual start address of the
2285     * vma, merges and splits can happen in a seamless way, just
2286     * using the existing file pgoff checks and manipulations.
2287     * Similarly in do_mmap_pgoff and in do_brk.
2288     */
2289    if (!vma->vm_file) {
2290        BUG_ON(vma->anon_vma);
2291        vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2292    }
2293    __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2294    if (__vma && __vma->vm_start < vma->vm_end)
2295        return -ENOMEM;
2296    if ((vma->vm_flags & VM_ACCOUNT) &&
2297         security_vm_enough_memory_mm(mm, vma_pages(vma)))
2298        return -ENOMEM;
2299    vma_link(mm, vma, prev, rb_link, rb_parent);
2300    return 0;
2301}
2302
2303/*
2304 * Copy the vma structure to a new location in the same mm,
2305 * prior to moving page table entries, to effect an mremap move.
2306 */
2307struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2308    unsigned long addr, unsigned long len, pgoff_t pgoff)
2309{
2310    struct vm_area_struct *vma = *vmap;
2311    unsigned long vma_start = vma->vm_start;
2312    struct mm_struct *mm = vma->vm_mm;
2313    struct vm_area_struct *new_vma, *prev;
2314    struct rb_node **rb_link, *rb_parent;
2315    struct mempolicy *pol;
2316
2317    /*
2318     * If anonymous vma has not yet been faulted, update new pgoff
2319     * to match new location, to increase its chance of merging.
2320     */
2321    if (!vma->vm_file && !vma->anon_vma)
2322        pgoff = addr >> PAGE_SHIFT;
2323
2324    find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2325    new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2326            vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2327    if (new_vma) {
2328        /*
2329         * Source vma may have been merged into new_vma
2330         */
2331        if (vma_start >= new_vma->vm_start &&
2332            vma_start < new_vma->vm_end)
2333            *vmap = new_vma;
2334    } else {
2335        new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2336        if (new_vma) {
2337            *new_vma = *vma;
2338            pol = mpol_dup(vma_policy(vma));
2339            if (IS_ERR(pol))
2340                goto out_free_vma;
2341            INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2342            if (anon_vma_clone(new_vma, vma))
2343                goto out_free_mempol;
2344            vma_set_policy(new_vma, pol);
2345            new_vma->vm_start = addr;
2346            new_vma->vm_end = addr + len;
2347            new_vma->vm_pgoff = pgoff;
2348            if (new_vma->vm_file) {
2349                get_file(new_vma->vm_file);
2350                if (vma->vm_flags & VM_EXECUTABLE)
2351                    added_exe_file_vma(mm);
2352            }
2353            if (new_vma->vm_ops && new_vma->vm_ops->open)
2354                new_vma->vm_ops->open(new_vma);
2355            vma_link(mm, new_vma, prev, rb_link, rb_parent);
2356        }
2357    }
2358    return new_vma;
2359
2360 out_free_mempol:
2361    mpol_put(pol);
2362 out_free_vma:
2363    kmem_cache_free(vm_area_cachep, new_vma);
2364    return NULL;
2365}
2366
2367/*
2368 * Return true if the calling process may expand its vm space by the passed
2369 * number of pages
2370 */
2371int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2372{
2373    unsigned long cur = mm->total_vm; /* pages */
2374    unsigned long lim;
2375
2376    lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2377
2378    if (cur + npages > lim)
2379        return 0;
2380    return 1;
2381}
2382
2383
2384static int special_mapping_fault(struct vm_area_struct *vma,
2385                struct vm_fault *vmf)
2386{
2387    pgoff_t pgoff;
2388    struct page **pages;
2389
2390    /*
2391     * special mappings have no vm_file, and in that case, the mm
2392     * uses vm_pgoff internally. So we have to subtract it from here.
2393     * We are allowed to do this because we are the mm; do not copy
2394     * this code into drivers!
2395     */
2396    pgoff = vmf->pgoff - vma->vm_pgoff;
2397
2398    for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2399        pgoff--;
2400
2401    if (*pages) {
2402        struct page *page = *pages;
2403        get_page(page);
2404        vmf->page = page;
2405        return 0;
2406    }
2407
2408    return VM_FAULT_SIGBUS;
2409}
2410
2411/*
2412 * Having a close hook prevents vma merging regardless of flags.
2413 */
2414static void special_mapping_close(struct vm_area_struct *vma)
2415{
2416}
2417
2418static const struct vm_operations_struct special_mapping_vmops = {
2419    .close = special_mapping_close,
2420    .fault = special_mapping_fault,
2421};
2422
2423/*
2424 * Called with mm->mmap_sem held for writing.
2425 * Insert a new vma covering the given region, with the given flags.
2426 * Its pages are supplied by the given array of struct page *.
2427 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2428 * The region past the last page supplied will always produce SIGBUS.
2429 * The array pointer and the pages it points to are assumed to stay alive
2430 * for as long as this mapping might exist.
2431 */
2432int install_special_mapping(struct mm_struct *mm,
2433                unsigned long addr, unsigned long len,
2434                unsigned long vm_flags, struct page **pages)
2435{
2436    struct vm_area_struct *vma;
2437
2438    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2439    if (unlikely(vma == NULL))
2440        return -ENOMEM;
2441
2442    INIT_LIST_HEAD(&vma->anon_vma_chain);
2443    vma->vm_mm = mm;
2444    vma->vm_start = addr;
2445    vma->vm_end = addr + len;
2446
2447    vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2448    vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2449
2450    vma->vm_ops = &special_mapping_vmops;
2451    vma->vm_private_data = pages;
2452
2453    if (unlikely(insert_vm_struct(mm, vma))) {
2454        kmem_cache_free(vm_area_cachep, vma);
2455        return -ENOMEM;
2456    }
2457
2458    mm->total_vm += len >> PAGE_SHIFT;
2459
2460    perf_event_mmap(vma);
2461
2462    return 0;
2463}
2464
2465static DEFINE_MUTEX(mm_all_locks_mutex);
2466
2467static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2468{
2469    if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2470        /*
2471         * The LSB of head.next can't change from under us
2472         * because we hold the mm_all_locks_mutex.
2473         */
2474        spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
2475        /*
2476         * We can safely modify head.next after taking the
2477         * anon_vma->lock. If some other vma in this mm shares
2478         * the same anon_vma we won't take it again.
2479         *
2480         * No need of atomic instructions here, head.next
2481         * can't change from under us thanks to the
2482         * anon_vma->lock.
2483         */
2484        if (__test_and_set_bit(0, (unsigned long *)
2485                       &anon_vma->head.next))
2486            BUG();
2487    }
2488}
2489
2490static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2491{
2492    if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2493        /*
2494         * AS_MM_ALL_LOCKS can't change from under us because
2495         * we hold the mm_all_locks_mutex.
2496         *
2497         * Operations on ->flags have to be atomic because
2498         * even if AS_MM_ALL_LOCKS is stable thanks to the
2499         * mm_all_locks_mutex, there may be other cpus
2500         * changing other bitflags in parallel to us.
2501         */
2502        if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2503            BUG();
2504        spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2505    }
2506}
2507
2508/*
2509 * This operation locks against the VM for all pte/vma/mm related
2510 * operations that could ever happen on a certain mm. This includes
2511 * vmtruncate, try_to_unmap, and all page faults.
2512 *
2513 * The caller must take the mmap_sem in write mode before calling
2514 * mm_take_all_locks(). The caller isn't allowed to release the
2515 * mmap_sem until mm_drop_all_locks() returns.
2516 *
2517 * mmap_sem in write mode is required in order to block all operations
2518 * that could modify pagetables and free pages without need of
2519 * altering the vma layout (for example populate_range() with
2520 * nonlinear vmas). It's also needed in write mode to avoid new
2521 * anon_vmas to be associated with existing vmas.
2522 *
2523 * A single task can't take more than one mm_take_all_locks() in a row
2524 * or it would deadlock.
2525 *
2526 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2527 * mapping->flags avoid to take the same lock twice, if more than one
2528 * vma in this mm is backed by the same anon_vma or address_space.
2529 *
2530 * We can take all the locks in random order because the VM code
2531 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2532 * takes more than one of them in a row. Secondly we're protected
2533 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2534 *
2535 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2536 * that may have to take thousand of locks.
2537 *
2538 * mm_take_all_locks() can fail if it's interrupted by signals.
2539 */
2540int mm_take_all_locks(struct mm_struct *mm)
2541{
2542    struct vm_area_struct *vma;
2543    struct anon_vma_chain *avc;
2544    int ret = -EINTR;
2545
2546    BUG_ON(down_read_trylock(&mm->mmap_sem));
2547
2548    mutex_lock(&mm_all_locks_mutex);
2549
2550    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2551        if (signal_pending(current))
2552            goto out_unlock;
2553        if (vma->vm_file && vma->vm_file->f_mapping)
2554            vm_lock_mapping(mm, vma->vm_file->f_mapping);
2555    }
2556
2557    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2558        if (signal_pending(current))
2559            goto out_unlock;
2560        if (vma->anon_vma)
2561            list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2562                vm_lock_anon_vma(mm, avc->anon_vma);
2563    }
2564
2565    ret = 0;
2566
2567out_unlock:
2568    if (ret)
2569        mm_drop_all_locks(mm);
2570
2571    return ret;
2572}
2573
2574static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2575{
2576    if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2577        /*
2578         * The LSB of head.next can't change to 0 from under
2579         * us because we hold the mm_all_locks_mutex.
2580         *
2581         * We must however clear the bitflag before unlocking
2582         * the vma so the users using the anon_vma->head will
2583         * never see our bitflag.
2584         *
2585         * No need of atomic instructions here, head.next
2586         * can't change from under us until we release the
2587         * anon_vma->lock.
2588         */
2589        if (!__test_and_clear_bit(0, (unsigned long *)
2590                      &anon_vma->head.next))
2591            BUG();
2592        spin_unlock(&anon_vma->lock);
2593    }
2594}
2595
2596static void vm_unlock_mapping(struct address_space *mapping)
2597{
2598    if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2599        /*
2600         * AS_MM_ALL_LOCKS can't change to 0 from under us
2601         * because we hold the mm_all_locks_mutex.
2602         */
2603        spin_unlock(&mapping->i_mmap_lock);
2604        if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2605                    &mapping->flags))
2606            BUG();
2607    }
2608}
2609
2610/*
2611 * The mmap_sem cannot be released by the caller until
2612 * mm_drop_all_locks() returns.
2613 */
2614void mm_drop_all_locks(struct mm_struct *mm)
2615{
2616    struct vm_area_struct *vma;
2617    struct anon_vma_chain *avc;
2618
2619    BUG_ON(down_read_trylock(&mm->mmap_sem));
2620    BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2621
2622    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2623        if (vma->anon_vma)
2624            list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2625                vm_unlock_anon_vma(avc->anon_vma);
2626        if (vma->vm_file && vma->vm_file->f_mapping)
2627            vm_unlock_mapping(vma->vm_file->f_mapping);
2628    }
2629
2630    mutex_unlock(&mm_all_locks_mutex);
2631}
2632
2633/*
2634 * initialise the VMA slab
2635 */
2636void __init mmap_init(void)
2637{
2638    int ret;
2639
2640    ret = percpu_counter_init(&vm_committed_as, 0);
2641    VM_BUG_ON(ret);
2642}
2643

Archive Download this file



interactive