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    struct vm_area_struct *next;
392
393    vma->vm_prev = prev;
394    if (prev) {
395        next = prev->vm_next;
396        prev->vm_next = vma;
397    } else {
398        mm->mmap = vma;
399        if (rb_parent)
400            next = rb_entry(rb_parent,
401                    struct vm_area_struct, vm_rb);
402        else
403            next = NULL;
404    }
405    vma->vm_next = next;
406    if (next)
407        next->vm_prev = vma;
408}
409
410void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
411        struct rb_node **rb_link, struct rb_node *rb_parent)
412{
413    rb_link_node(&vma->vm_rb, rb_parent, rb_link);
414    rb_insert_color(&vma->vm_rb, &mm->mm_rb);
415}
416
417static void __vma_link_file(struct vm_area_struct *vma)
418{
419    struct file *file;
420
421    file = vma->vm_file;
422    if (file) {
423        struct address_space *mapping = file->f_mapping;
424
425        if (vma->vm_flags & VM_DENYWRITE)
426            atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
427        if (vma->vm_flags & VM_SHARED)
428            mapping->i_mmap_writable++;
429
430        flush_dcache_mmap_lock(mapping);
431        if (unlikely(vma->vm_flags & VM_NONLINEAR))
432            vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
433        else
434            vma_prio_tree_insert(vma, &mapping->i_mmap);
435        flush_dcache_mmap_unlock(mapping);
436    }
437}
438
439static void
440__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
441    struct vm_area_struct *prev, struct rb_node **rb_link,
442    struct rb_node *rb_parent)
443{
444    __vma_link_list(mm, vma, prev, rb_parent);
445    __vma_link_rb(mm, vma, rb_link, rb_parent);
446}
447
448static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
449            struct vm_area_struct *prev, struct rb_node **rb_link,
450            struct rb_node *rb_parent)
451{
452    struct address_space *mapping = NULL;
453
454    if (vma->vm_file)
455        mapping = vma->vm_file->f_mapping;
456
457    if (mapping) {
458        spin_lock(&mapping->i_mmap_lock);
459        vma->vm_truncate_count = mapping->truncate_count;
460    }
461
462    __vma_link(mm, vma, prev, rb_link, rb_parent);
463    __vma_link_file(vma);
464
465    if (mapping)
466        spin_unlock(&mapping->i_mmap_lock);
467
468    mm->map_count++;
469    validate_mm(mm);
470}
471
472/*
473 * Helper for vma_adjust in the split_vma insert case:
474 * insert vm structure into list and rbtree and anon_vma,
475 * but it has already been inserted into prio_tree earlier.
476 */
477static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
478{
479    struct vm_area_struct *__vma, *prev;
480    struct rb_node **rb_link, *rb_parent;
481
482    __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
483    BUG_ON(__vma && __vma->vm_start < vma->vm_end);
484    __vma_link(mm, vma, prev, rb_link, rb_parent);
485    mm->map_count++;
486}
487
488static inline void
489__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
490        struct vm_area_struct *prev)
491{
492    struct vm_area_struct *next = vma->vm_next;
493
494    prev->vm_next = next;
495    if (next)
496        next->vm_prev = prev;
497    rb_erase(&vma->vm_rb, &mm->mm_rb);
498    if (mm->mmap_cache == vma)
499        mm->mmap_cache = prev;
500}
501
502/*
503 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
504 * is already present in an i_mmap tree without adjusting the tree.
505 * The following helper function should be used when such adjustments
506 * are necessary. The "insert" vma (if any) is to be inserted
507 * before we drop the necessary locks.
508 */
509int vma_adjust(struct vm_area_struct *vma, unsigned long start,
510    unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
511{
512    struct mm_struct *mm = vma->vm_mm;
513    struct vm_area_struct *next = vma->vm_next;
514    struct vm_area_struct *importer = NULL;
515    struct address_space *mapping = NULL;
516    struct prio_tree_root *root = NULL;
517    struct anon_vma *anon_vma = NULL;
518    struct file *file = vma->vm_file;
519    long adjust_next = 0;
520    int remove_next = 0;
521
522    if (next && !insert) {
523        struct vm_area_struct *exporter = NULL;
524
525        if (end >= next->vm_end) {
526            /*
527             * vma expands, overlapping all the next, and
528             * perhaps the one after too (mprotect case 6).
529             */
530again: remove_next = 1 + (end > next->vm_end);
531            end = next->vm_end;
532            exporter = next;
533            importer = vma;
534        } else if (end > next->vm_start) {
535            /*
536             * vma expands, overlapping part of the next:
537             * mprotect case 5 shifting the boundary up.
538             */
539            adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
540            exporter = next;
541            importer = vma;
542        } else if (end < vma->vm_end) {
543            /*
544             * vma shrinks, and !insert tells it's not
545             * split_vma inserting another: so it must be
546             * mprotect case 4 shifting the boundary down.
547             */
548            adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
549            exporter = vma;
550            importer = next;
551        }
552
553        /*
554         * Easily overlooked: when mprotect shifts the boundary,
555         * make sure the expanding vma has anon_vma set if the
556         * shrinking vma had, to cover any anon pages imported.
557         */
558        if (exporter && exporter->anon_vma && !importer->anon_vma) {
559            if (anon_vma_clone(importer, exporter))
560                return -ENOMEM;
561            importer->anon_vma = exporter->anon_vma;
562        }
563    }
564
565    if (file) {
566        mapping = file->f_mapping;
567        if (!(vma->vm_flags & VM_NONLINEAR))
568            root = &mapping->i_mmap;
569        spin_lock(&mapping->i_mmap_lock);
570        if (importer &&
571            vma->vm_truncate_count != next->vm_truncate_count) {
572            /*
573             * unmap_mapping_range might be in progress:
574             * ensure that the expanding vma is rescanned.
575             */
576            importer->vm_truncate_count = 0;
577        }
578        if (insert) {
579            insert->vm_truncate_count = vma->vm_truncate_count;
580            /*
581             * Put into prio_tree now, so instantiated pages
582             * are visible to arm/parisc __flush_dcache_page
583             * throughout; but we cannot insert into address
584             * space until vma start or end is updated.
585             */
586            __vma_link_file(insert);
587        }
588    }
589
590    /*
591     * When changing only vma->vm_end, we don't really need anon_vma
592     * lock. This is a fairly rare case by itself, but the anon_vma
593     * lock may be shared between many sibling processes. Skipping
594     * the lock for brk adjustments makes a difference sometimes.
595     */
596    if (vma->anon_vma && (insert || importer || start != vma->vm_start)) {
597        anon_vma = vma->anon_vma;
598        anon_vma_lock(anon_vma);
599    }
600
601    if (root) {
602        flush_dcache_mmap_lock(mapping);
603        vma_prio_tree_remove(vma, root);
604        if (adjust_next)
605            vma_prio_tree_remove(next, root);
606    }
607
608    vma->vm_start = start;
609    vma->vm_end = end;
610    vma->vm_pgoff = pgoff;
611    if (adjust_next) {
612        next->vm_start += adjust_next << PAGE_SHIFT;
613        next->vm_pgoff += adjust_next;
614    }
615
616    if (root) {
617        if (adjust_next)
618            vma_prio_tree_insert(next, root);
619        vma_prio_tree_insert(vma, root);
620        flush_dcache_mmap_unlock(mapping);
621    }
622
623    if (remove_next) {
624        /*
625         * vma_merge has merged next into vma, and needs
626         * us to remove next before dropping the locks.
627         */
628        __vma_unlink(mm, next, vma);
629        if (file)
630            __remove_shared_vm_struct(next, file, mapping);
631    } else if (insert) {
632        /*
633         * split_vma has split insert from vma, and needs
634         * us to insert it before dropping the locks
635         * (it may either follow vma or precede it).
636         */
637        __insert_vm_struct(mm, insert);
638    }
639
640    if (anon_vma)
641        anon_vma_unlock(anon_vma);
642    if (mapping)
643        spin_unlock(&mapping->i_mmap_lock);
644
645    if (remove_next) {
646        if (file) {
647            fput(file);
648            if (next->vm_flags & VM_EXECUTABLE)
649                removed_exe_file_vma(mm);
650        }
651        if (next->anon_vma)
652            anon_vma_merge(vma, next);
653        mm->map_count--;
654        mpol_put(vma_policy(next));
655        kmem_cache_free(vm_area_cachep, next);
656        /*
657         * In mprotect's case 6 (see comments on vma_merge),
658         * we must remove another next too. It would clutter
659         * up the code too much to do both in one go.
660         */
661        if (remove_next == 2) {
662            next = vma->vm_next;
663            goto again;
664        }
665    }
666
667    validate_mm(mm);
668
669    return 0;
670}
671
672/*
673 * If the vma has a ->close operation then the driver probably needs to release
674 * per-vma resources, so we don't attempt to merge those.
675 */
676static inline int is_mergeable_vma(struct vm_area_struct *vma,
677            struct file *file, unsigned long vm_flags)
678{
679    /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
680    if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
681        return 0;
682    if (vma->vm_file != file)
683        return 0;
684    if (vma->vm_ops && vma->vm_ops->close)
685        return 0;
686    return 1;
687}
688
689static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
690                    struct anon_vma *anon_vma2)
691{
692    return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
693}
694
695/*
696 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
697 * in front of (at a lower virtual address and file offset than) the vma.
698 *
699 * We cannot merge two vmas if they have differently assigned (non-NULL)
700 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
701 *
702 * We don't check here for the merged mmap wrapping around the end of pagecache
703 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
704 * wrap, nor mmaps which cover the final page at index -1UL.
705 */
706static int
707can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
708    struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
709{
710    if (is_mergeable_vma(vma, file, vm_flags) &&
711        is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
712        if (vma->vm_pgoff == vm_pgoff)
713            return 1;
714    }
715    return 0;
716}
717
718/*
719 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
720 * beyond (at a higher virtual address and file offset than) the vma.
721 *
722 * We cannot merge two vmas if they have differently assigned (non-NULL)
723 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
724 */
725static int
726can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
727    struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
728{
729    if (is_mergeable_vma(vma, file, vm_flags) &&
730        is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
731        pgoff_t vm_pglen;
732        vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
733        if (vma->vm_pgoff + vm_pglen == vm_pgoff)
734            return 1;
735    }
736    return 0;
737}
738
739/*
740 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
741 * whether that can be merged with its predecessor or its successor.
742 * Or both (it neatly fills a hole).
743 *
744 * In most cases - when called for mmap, brk or mremap - [addr,end) is
745 * certain not to be mapped by the time vma_merge is called; but when
746 * called for mprotect, it is certain to be already mapped (either at
747 * an offset within prev, or at the start of next), and the flags of
748 * this area are about to be changed to vm_flags - and the no-change
749 * case has already been eliminated.
750 *
751 * The following mprotect cases have to be considered, where AAAA is
752 * the area passed down from mprotect_fixup, never extending beyond one
753 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
754 *
755 * AAAA AAAA AAAA AAAA
756 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
757 * cannot merge might become might become might become
758 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
759 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
760 * mremap move: PPPPNNNNNNNN 8
761 * AAAA
762 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
763 * might become case 1 below case 2 below case 3 below
764 *
765 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
766 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
767 */
768struct vm_area_struct *vma_merge(struct mm_struct *mm,
769            struct vm_area_struct *prev, unsigned long addr,
770            unsigned long end, unsigned long vm_flags,
771                 struct anon_vma *anon_vma, struct file *file,
772            pgoff_t pgoff, struct mempolicy *policy)
773{
774    pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
775    struct vm_area_struct *area, *next;
776    int err;
777
778    /*
779     * We later require that vma->vm_flags == vm_flags,
780     * so this tests vma->vm_flags & VM_SPECIAL, too.
781     */
782    if (vm_flags & VM_SPECIAL)
783        return NULL;
784
785    if (prev)
786        next = prev->vm_next;
787    else
788        next = mm->mmap;
789    area = next;
790    if (next && next->vm_end == end) /* cases 6, 7, 8 */
791        next = next->vm_next;
792
793    /*
794     * Can it merge with the predecessor?
795     */
796    if (prev && prev->vm_end == addr &&
797              mpol_equal(vma_policy(prev), policy) &&
798            can_vma_merge_after(prev, vm_flags,
799                        anon_vma, file, pgoff)) {
800        /*
801         * OK, it can. Can we now merge in the successor as well?
802         */
803        if (next && end == next->vm_start &&
804                mpol_equal(policy, vma_policy(next)) &&
805                can_vma_merge_before(next, vm_flags,
806                    anon_vma, file, pgoff+pglen) &&
807                is_mergeable_anon_vma(prev->anon_vma,
808                              next->anon_vma)) {
809                            /* cases 1, 6 */
810            err = vma_adjust(prev, prev->vm_start,
811                next->vm_end, prev->vm_pgoff, NULL);
812        } else /* cases 2, 5, 7 */
813            err = vma_adjust(prev, prev->vm_start,
814                end, prev->vm_pgoff, NULL);
815        if (err)
816            return NULL;
817        return prev;
818    }
819
820    /*
821     * Can this new request be merged in front of next?
822     */
823    if (next && end == next->vm_start &&
824             mpol_equal(policy, vma_policy(next)) &&
825            can_vma_merge_before(next, vm_flags,
826                    anon_vma, file, pgoff+pglen)) {
827        if (prev && addr < prev->vm_end) /* case 4 */
828            err = vma_adjust(prev, prev->vm_start,
829                addr, prev->vm_pgoff, NULL);
830        else /* cases 3, 8 */
831            err = vma_adjust(area, addr, next->vm_end,
832                next->vm_pgoff - pglen, NULL);
833        if (err)
834            return NULL;
835        return area;
836    }
837
838    return NULL;
839}
840
841/*
842 * Rough compatbility check to quickly see if it's even worth looking
843 * at sharing an anon_vma.
844 *
845 * They need to have the same vm_file, and the flags can only differ
846 * in things that mprotect may change.
847 *
848 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
849 * we can merge the two vma's. For example, we refuse to merge a vma if
850 * there is a vm_ops->close() function, because that indicates that the
851 * driver is doing some kind of reference counting. But that doesn't
852 * really matter for the anon_vma sharing case.
853 */
854static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
855{
856    return a->vm_end == b->vm_start &&
857        mpol_equal(vma_policy(a), vma_policy(b)) &&
858        a->vm_file == b->vm_file &&
859        !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
860        b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
861}
862
863/*
864 * Do some basic sanity checking to see if we can re-use the anon_vma
865 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
866 * the same as 'old', the other will be the new one that is trying
867 * to share the anon_vma.
868 *
869 * NOTE! This runs with mm_sem held for reading, so it is possible that
870 * the anon_vma of 'old' is concurrently in the process of being set up
871 * by another page fault trying to merge _that_. But that's ok: if it
872 * is being set up, that automatically means that it will be a singleton
873 * acceptable for merging, so we can do all of this optimistically. But
874 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
875 *
876 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
877 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
878 * is to return an anon_vma that is "complex" due to having gone through
879 * a fork).
880 *
881 * We also make sure that the two vma's are compatible (adjacent,
882 * and with the same memory policies). That's all stable, even with just
883 * a read lock on the mm_sem.
884 */
885static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
886{
887    if (anon_vma_compatible(a, b)) {
888        struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
889
890        if (anon_vma && list_is_singular(&old->anon_vma_chain))
891            return anon_vma;
892    }
893    return NULL;
894}
895
896/*
897 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
898 * neighbouring vmas for a suitable anon_vma, before it goes off
899 * to allocate a new anon_vma. It checks because a repetitive
900 * sequence of mprotects and faults may otherwise lead to distinct
901 * anon_vmas being allocated, preventing vma merge in subsequent
902 * mprotect.
903 */
904struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
905{
906    struct anon_vma *anon_vma;
907    struct vm_area_struct *near;
908
909    near = vma->vm_next;
910    if (!near)
911        goto try_prev;
912
913    anon_vma = reusable_anon_vma(near, vma, near);
914    if (anon_vma)
915        return anon_vma;
916try_prev:
917    /*
918     * It is potentially slow to have to call find_vma_prev here.
919     * But it's only on the first write fault on the vma, not
920     * every time, and we could devise a way to avoid it later
921     * (e.g. stash info in next's anon_vma_node when assigning
922     * an anon_vma, or when trying vma_merge). Another time.
923     */
924    BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
925    if (!near)
926        goto none;
927
928    anon_vma = reusable_anon_vma(near, near, vma);
929    if (anon_vma)
930        return anon_vma;
931none:
932    /*
933     * There's no absolute need to look only at touching neighbours:
934     * we could search further afield for "compatible" anon_vmas.
935     * But it would probably just be a waste of time searching,
936     * or lead to too many vmas hanging off the same anon_vma.
937     * We're trying to allow mprotect remerging later on,
938     * not trying to minimize memory used for anon_vmas.
939     */
940    return NULL;
941}
942
943#ifdef CONFIG_PROC_FS
944void vm_stat_account(struct mm_struct *mm, unsigned long flags,
945                        struct file *file, long pages)
946{
947    const unsigned long stack_flags
948        = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
949
950    if (file) {
951        mm->shared_vm += pages;
952        if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
953            mm->exec_vm += pages;
954    } else if (flags & stack_flags)
955        mm->stack_vm += pages;
956    if (flags & (VM_RESERVED|VM_IO))
957        mm->reserved_vm += pages;
958}
959#endif /* CONFIG_PROC_FS */
960
961/*
962 * The caller must hold down_write(&current->mm->mmap_sem).
963 */
964
965unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
966            unsigned long len, unsigned long prot,
967            unsigned long flags, unsigned long pgoff)
968{
969    struct mm_struct * mm = current->mm;
970    struct inode *inode;
971    unsigned int vm_flags;
972    int error;
973    unsigned long reqprot = prot;
974
975    /*
976     * Does the application expect PROT_READ to imply PROT_EXEC?
977     *
978     * (the exception is when the underlying filesystem is noexec
979     * mounted, in which case we dont add PROT_EXEC.)
980     */
981    if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
982        if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
983            prot |= PROT_EXEC;
984
985    if (!len)
986        return -EINVAL;
987
988    if (!(flags & MAP_FIXED))
989        addr = round_hint_to_min(addr);
990
991    /* Careful about overflows.. */
992    len = PAGE_ALIGN(len);
993    if (!len)
994        return -ENOMEM;
995
996    /* offset overflow? */
997    if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
998               return -EOVERFLOW;
999
1000    /* Too many mappings? */
1001    if (mm->map_count > sysctl_max_map_count)
1002        return -ENOMEM;
1003
1004    /* Obtain the address to map to. we verify (or select) it and ensure
1005     * that it represents a valid section of the address space.
1006     */
1007    addr = get_unmapped_area(file, addr, len, pgoff, flags);
1008    if (addr & ~PAGE_MASK)
1009        return addr;
1010
1011    /* Do simple checking here so the lower-level routines won't have
1012     * to. we assume access permissions have been handled by the open
1013     * of the memory object, so we don't do any here.
1014     */
1015    vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1016            mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1017
1018    if (flags & MAP_LOCKED)
1019        if (!can_do_mlock())
1020            return -EPERM;
1021
1022    /* mlock MCL_FUTURE? */
1023    if (vm_flags & VM_LOCKED) {
1024        unsigned long locked, lock_limit;
1025        locked = len >> PAGE_SHIFT;
1026        locked += mm->locked_vm;
1027        lock_limit = rlimit(RLIMIT_MEMLOCK);
1028        lock_limit >>= PAGE_SHIFT;
1029        if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1030            return -EAGAIN;
1031    }
1032
1033    inode = file ? file->f_path.dentry->d_inode : NULL;
1034
1035    if (file) {
1036        switch (flags & MAP_TYPE) {
1037        case MAP_SHARED:
1038            if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1039                return -EACCES;
1040
1041            /*
1042             * Make sure we don't allow writing to an append-only
1043             * file..
1044             */
1045            if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1046                return -EACCES;
1047
1048            /*
1049             * Make sure there are no mandatory locks on the file.
1050             */
1051            if (locks_verify_locked(inode))
1052                return -EAGAIN;
1053
1054            vm_flags |= VM_SHARED | VM_MAYSHARE;
1055            if (!(file->f_mode & FMODE_WRITE))
1056                vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1057
1058            /* fall through */
1059        case MAP_PRIVATE:
1060            if (!(file->f_mode & FMODE_READ))
1061                return -EACCES;
1062            if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1063                if (vm_flags & VM_EXEC)
1064                    return -EPERM;
1065                vm_flags &= ~VM_MAYEXEC;
1066            }
1067
1068            if (!file->f_op || !file->f_op->mmap)
1069                return -ENODEV;
1070            break;
1071
1072        default:
1073            return -EINVAL;
1074        }
1075    } else {
1076        switch (flags & MAP_TYPE) {
1077        case MAP_SHARED:
1078            /*
1079             * Ignore pgoff.
1080             */
1081            pgoff = 0;
1082            vm_flags |= VM_SHARED | VM_MAYSHARE;
1083            break;
1084        case MAP_PRIVATE:
1085            /*
1086             * Set pgoff according to addr for anon_vma.
1087             */
1088            pgoff = addr >> PAGE_SHIFT;
1089            break;
1090        default:
1091            return -EINVAL;
1092        }
1093    }
1094
1095    error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1096    if (error)
1097        return error;
1098
1099    return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1100}
1101EXPORT_SYMBOL(do_mmap_pgoff);
1102
1103SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1104        unsigned long, prot, unsigned long, flags,
1105        unsigned long, fd, unsigned long, pgoff)
1106{
1107    struct file *file = NULL;
1108    unsigned long retval = -EBADF;
1109
1110    if (!(flags & MAP_ANONYMOUS)) {
1111        if (unlikely(flags & MAP_HUGETLB))
1112            return -EINVAL;
1113        file = fget(fd);
1114        if (!file)
1115            goto out;
1116    } else if (flags & MAP_HUGETLB) {
1117        struct user_struct *user = NULL;
1118        /*
1119         * VM_NORESERVE is used because the reservations will be
1120         * taken when vm_ops->mmap() is called
1121         * A dummy user value is used because we are not locking
1122         * memory so no accounting is necessary
1123         */
1124        len = ALIGN(len, huge_page_size(&default_hstate));
1125        file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1126                        &user, HUGETLB_ANONHUGE_INODE);
1127        if (IS_ERR(file))
1128            return PTR_ERR(file);
1129    }
1130
1131    flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1132
1133    down_write(&current->mm->mmap_sem);
1134    retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1135    up_write(&current->mm->mmap_sem);
1136
1137    if (file)
1138        fput(file);
1139out:
1140    return retval;
1141}
1142
1143#ifdef __ARCH_WANT_SYS_OLD_MMAP
1144struct mmap_arg_struct {
1145    unsigned long addr;
1146    unsigned long len;
1147    unsigned long prot;
1148    unsigned long flags;
1149    unsigned long fd;
1150    unsigned long offset;
1151};
1152
1153SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1154{
1155    struct mmap_arg_struct a;
1156
1157    if (copy_from_user(&a, arg, sizeof(a)))
1158        return -EFAULT;
1159    if (a.offset & ~PAGE_MASK)
1160        return -EINVAL;
1161
1162    return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1163                  a.offset >> PAGE_SHIFT);
1164}
1165#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1166
1167/*
1168 * Some shared mappigns will want the pages marked read-only
1169 * to track write events. If so, we'll downgrade vm_page_prot
1170 * to the private version (using protection_map[] without the
1171 * VM_SHARED bit).
1172 */
1173int vma_wants_writenotify(struct vm_area_struct *vma)
1174{
1175    unsigned int vm_flags = vma->vm_flags;
1176
1177    /* If it was private or non-writable, the write bit is already clear */
1178    if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1179        return 0;
1180
1181    /* The backer wishes to know when pages are first written to? */
1182    if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1183        return 1;
1184
1185    /* The open routine did something to the protections already? */
1186    if (pgprot_val(vma->vm_page_prot) !=
1187        pgprot_val(vm_get_page_prot(vm_flags)))
1188        return 0;
1189
1190    /* Specialty mapping? */
1191    if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1192        return 0;
1193
1194    /* Can the mapping track the dirty pages? */
1195    return vma->vm_file && vma->vm_file->f_mapping &&
1196        mapping_cap_account_dirty(vma->vm_file->f_mapping);
1197}
1198
1199/*
1200 * We account for memory if it's a private writeable mapping,
1201 * not hugepages and VM_NORESERVE wasn't set.
1202 */
1203static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
1204{
1205    /*
1206     * hugetlb has its own accounting separate from the core VM
1207     * VM_HUGETLB may not be set yet so we cannot check for that flag.
1208     */
1209    if (file && is_file_hugepages(file))
1210        return 0;
1211
1212    return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1213}
1214
1215unsigned long mmap_region(struct file *file, unsigned long addr,
1216              unsigned long len, unsigned long flags,
1217              unsigned int vm_flags, unsigned long pgoff)
1218{
1219    struct mm_struct *mm = current->mm;
1220    struct vm_area_struct *vma, *prev;
1221    int correct_wcount = 0;
1222    int error;
1223    struct rb_node **rb_link, *rb_parent;
1224    unsigned long charged = 0;
1225    struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1226
1227    /* Clear old maps */
1228    error = -ENOMEM;
1229munmap_back:
1230    vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1231    if (vma && vma->vm_start < addr + len) {
1232        if (do_munmap(mm, addr, len))
1233            return -ENOMEM;
1234        goto munmap_back;
1235    }
1236
1237    /* Check against address space limit. */
1238    if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1239        return -ENOMEM;
1240
1241    /*
1242     * Set 'VM_NORESERVE' if we should not account for the
1243     * memory use of this mapping.
1244     */
1245    if ((flags & MAP_NORESERVE)) {
1246        /* We honor MAP_NORESERVE if allowed to overcommit */
1247        if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1248            vm_flags |= VM_NORESERVE;
1249
1250        /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1251        if (file && is_file_hugepages(file))
1252            vm_flags |= VM_NORESERVE;
1253    }
1254
1255    /*
1256     * Private writable mapping: check memory availability
1257     */
1258    if (accountable_mapping(file, vm_flags)) {
1259        charged = len >> PAGE_SHIFT;
1260        if (security_vm_enough_memory(charged))
1261            return -ENOMEM;
1262        vm_flags |= VM_ACCOUNT;
1263    }
1264
1265    /*
1266     * Can we just expand an old mapping?
1267     */
1268    vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1269    if (vma)
1270        goto out;
1271
1272    /*
1273     * Determine the object being mapped and call the appropriate
1274     * specific mapper. the address has already been validated, but
1275     * not unmapped, but the maps are removed from the list.
1276     */
1277    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1278    if (!vma) {
1279        error = -ENOMEM;
1280        goto unacct_error;
1281    }
1282
1283    vma->vm_mm = mm;
1284    vma->vm_start = addr;
1285    vma->vm_end = addr + len;
1286    vma->vm_flags = vm_flags;
1287    vma->vm_page_prot = vm_get_page_prot(vm_flags);
1288    vma->vm_pgoff = pgoff;
1289    INIT_LIST_HEAD(&vma->anon_vma_chain);
1290
1291    if (file) {
1292        error = -EINVAL;
1293        if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1294            goto free_vma;
1295        if (vm_flags & VM_DENYWRITE) {
1296            error = deny_write_access(file);
1297            if (error)
1298                goto free_vma;
1299            correct_wcount = 1;
1300        }
1301        vma->vm_file = file;
1302        get_file(file);
1303        error = file->f_op->mmap(file, vma);
1304        if (error)
1305            goto unmap_and_free_vma;
1306        if (vm_flags & VM_EXECUTABLE)
1307            added_exe_file_vma(mm);
1308
1309        /* Can addr have changed??
1310         *
1311         * Answer: Yes, several device drivers can do it in their
1312         * f_op->mmap method. -DaveM
1313         */
1314        addr = vma->vm_start;
1315        pgoff = vma->vm_pgoff;
1316        vm_flags = vma->vm_flags;
1317    } else if (vm_flags & VM_SHARED) {
1318        error = shmem_zero_setup(vma);
1319        if (error)
1320            goto free_vma;
1321    }
1322
1323    if (vma_wants_writenotify(vma)) {
1324        pgprot_t pprot = vma->vm_page_prot;
1325
1326        /* Can vma->vm_page_prot have changed??
1327         *
1328         * Answer: Yes, drivers may have changed it in their
1329         * f_op->mmap method.
1330         *
1331         * Ensures that vmas marked as uncached stay that way.
1332         */
1333        vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1334        if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1335            vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1336    }
1337
1338    vma_link(mm, vma, prev, rb_link, rb_parent);
1339    file = vma->vm_file;
1340
1341    /* Once vma denies write, undo our temporary denial count */
1342    if (correct_wcount)
1343        atomic_inc(&inode->i_writecount);
1344out:
1345    perf_event_mmap(vma);
1346
1347    mm->total_vm += len >> PAGE_SHIFT;
1348    vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1349    if (vm_flags & VM_LOCKED) {
1350        if (!mlock_vma_pages_range(vma, addr, addr + len))
1351            mm->locked_vm += (len >> PAGE_SHIFT);
1352    } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1353        make_pages_present(addr, addr + len);
1354    return addr;
1355
1356unmap_and_free_vma:
1357    if (correct_wcount)
1358        atomic_inc(&inode->i_writecount);
1359    vma->vm_file = NULL;
1360    fput(file);
1361
1362    /* Undo any partial mapping done by a device driver. */
1363    unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1364    charged = 0;
1365free_vma:
1366    kmem_cache_free(vm_area_cachep, vma);
1367unacct_error:
1368    if (charged)
1369        vm_unacct_memory(charged);
1370    return error;
1371}
1372
1373/* Get an address range which is currently unmapped.
1374 * For shmat() with addr=0.
1375 *
1376 * Ugly calling convention alert:
1377 * Return value with the low bits set means error value,
1378 * ie
1379 * if (ret & ~PAGE_MASK)
1380 * error = ret;
1381 *
1382 * This function "knows" that -ENOMEM has the bits set.
1383 */
1384#ifndef HAVE_ARCH_UNMAPPED_AREA
1385unsigned long
1386arch_get_unmapped_area(struct file *filp, unsigned long addr,
1387        unsigned long len, unsigned long pgoff, unsigned long flags)
1388{
1389    struct mm_struct *mm = current->mm;
1390    struct vm_area_struct *vma;
1391    unsigned long start_addr;
1392
1393    if (len > TASK_SIZE)
1394        return -ENOMEM;
1395
1396    if (flags & MAP_FIXED)
1397        return addr;
1398
1399    if (addr) {
1400        addr = PAGE_ALIGN(addr);
1401        vma = find_vma(mm, addr);
1402        if (TASK_SIZE - len >= addr &&
1403            (!vma || addr + len <= vma->vm_start))
1404            return addr;
1405    }
1406    if (len > mm->cached_hole_size) {
1407            start_addr = addr = mm->free_area_cache;
1408    } else {
1409            start_addr = addr = TASK_UNMAPPED_BASE;
1410            mm->cached_hole_size = 0;
1411    }
1412
1413full_search:
1414    for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1415        /* At this point: (!vma || addr < vma->vm_end). */
1416        if (TASK_SIZE - len < addr) {
1417            /*
1418             * Start a new search - just in case we missed
1419             * some holes.
1420             */
1421            if (start_addr != TASK_UNMAPPED_BASE) {
1422                addr = TASK_UNMAPPED_BASE;
1423                    start_addr = addr;
1424                mm->cached_hole_size = 0;
1425                goto full_search;
1426            }
1427            return -ENOMEM;
1428        }
1429        if (!vma || addr + len <= vma->vm_start) {
1430            /*
1431             * Remember the place where we stopped the search:
1432             */
1433            mm->free_area_cache = addr + len;
1434            return addr;
1435        }
1436        if (addr + mm->cached_hole_size < vma->vm_start)
1437                mm->cached_hole_size = vma->vm_start - addr;
1438        addr = vma->vm_end;
1439    }
1440}
1441#endif
1442
1443void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1444{
1445    /*
1446     * Is this a new hole at the lowest possible address?
1447     */
1448    if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1449        mm->free_area_cache = addr;
1450        mm->cached_hole_size = ~0UL;
1451    }
1452}
1453
1454/*
1455 * This mmap-allocator allocates new areas top-down from below the
1456 * stack's low limit (the base):
1457 */
1458#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1459unsigned long
1460arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1461              const unsigned long len, const unsigned long pgoff,
1462              const unsigned long flags)
1463{
1464    struct vm_area_struct *vma;
1465    struct mm_struct *mm = current->mm;
1466    unsigned long addr = addr0;
1467
1468    /* requested length too big for entire address space */
1469    if (len > TASK_SIZE)
1470        return -ENOMEM;
1471
1472    if (flags & MAP_FIXED)
1473        return addr;
1474
1475    /* requesting a specific address */
1476    if (addr) {
1477        addr = PAGE_ALIGN(addr);
1478        vma = find_vma(mm, addr);
1479        if (TASK_SIZE - len >= addr &&
1480                (!vma || addr + len <= vma->vm_start))
1481            return addr;
1482    }
1483
1484    /* check if free_area_cache is useful for us */
1485    if (len <= mm->cached_hole_size) {
1486             mm->cached_hole_size = 0;
1487         mm->free_area_cache = mm->mmap_base;
1488     }
1489
1490    /* either no address requested or can't fit in requested address hole */
1491    addr = mm->free_area_cache;
1492
1493    /* make sure it can fit in the remaining address space */
1494    if (addr > len) {
1495        vma = find_vma(mm, addr-len);
1496        if (!vma || addr <= vma->vm_start)
1497            /* remember the address as a hint for next time */
1498            return (mm->free_area_cache = addr-len);
1499    }
1500
1501    if (mm->mmap_base < len)
1502        goto bottomup;
1503
1504    addr = mm->mmap_base-len;
1505
1506    do {
1507        /*
1508         * Lookup failure means no vma is above this address,
1509         * else if new region fits below vma->vm_start,
1510         * return with success:
1511         */
1512        vma = find_vma(mm, addr);
1513        if (!vma || addr+len <= vma->vm_start)
1514            /* remember the address as a hint for next time */
1515            return (mm->free_area_cache = addr);
1516
1517         /* remember the largest hole we saw so far */
1518         if (addr + mm->cached_hole_size < vma->vm_start)
1519                 mm->cached_hole_size = vma->vm_start - addr;
1520
1521        /* try just below the current vma->vm_start */
1522        addr = vma->vm_start-len;
1523    } while (len < vma->vm_start);
1524
1525bottomup:
1526    /*
1527     * A failed mmap() very likely causes application failure,
1528     * so fall back to the bottom-up function here. This scenario
1529     * can happen with large stack limits and large mmap()
1530     * allocations.
1531     */
1532    mm->cached_hole_size = ~0UL;
1533      mm->free_area_cache = TASK_UNMAPPED_BASE;
1534    addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1535    /*
1536     * Restore the topdown base:
1537     */
1538    mm->free_area_cache = mm->mmap_base;
1539    mm->cached_hole_size = ~0UL;
1540
1541    return addr;
1542}
1543#endif
1544
1545void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1546{
1547    /*
1548     * Is this a new hole at the highest possible address?
1549     */
1550    if (addr > mm->free_area_cache)
1551        mm->free_area_cache = addr;
1552
1553    /* dont allow allocations above current base */
1554    if (mm->free_area_cache > mm->mmap_base)
1555        mm->free_area_cache = mm->mmap_base;
1556}
1557
1558unsigned long
1559get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1560        unsigned long pgoff, unsigned long flags)
1561{
1562    unsigned long (*get_area)(struct file *, unsigned long,
1563                  unsigned long, unsigned long, unsigned long);
1564
1565    unsigned long error = arch_mmap_check(addr, len, flags);
1566    if (error)
1567        return error;
1568
1569    /* Careful about overflows.. */
1570    if (len > TASK_SIZE)
1571        return -ENOMEM;
1572
1573    get_area = current->mm->get_unmapped_area;
1574    if (file && file->f_op && file->f_op->get_unmapped_area)
1575        get_area = file->f_op->get_unmapped_area;
1576    addr = get_area(file, addr, len, pgoff, flags);
1577    if (IS_ERR_VALUE(addr))
1578        return addr;
1579
1580    if (addr > TASK_SIZE - len)
1581        return -ENOMEM;
1582    if (addr & ~PAGE_MASK)
1583        return -EINVAL;
1584
1585    return arch_rebalance_pgtables(addr, len);
1586}
1587
1588EXPORT_SYMBOL(get_unmapped_area);
1589
1590/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1591struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1592{
1593    struct vm_area_struct *vma = NULL;
1594
1595    if (mm) {
1596        /* Check the cache first. */
1597        /* (Cache hit rate is typically around 35%.) */
1598        vma = mm->mmap_cache;
1599        if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1600            struct rb_node * rb_node;
1601
1602            rb_node = mm->mm_rb.rb_node;
1603            vma = NULL;
1604
1605            while (rb_node) {
1606                struct vm_area_struct * vma_tmp;
1607
1608                vma_tmp = rb_entry(rb_node,
1609                        struct vm_area_struct, vm_rb);
1610
1611                if (vma_tmp->vm_end > addr) {
1612                    vma = vma_tmp;
1613                    if (vma_tmp->vm_start <= addr)
1614                        break;
1615                    rb_node = rb_node->rb_left;
1616                } else
1617                    rb_node = rb_node->rb_right;
1618            }
1619            if (vma)
1620                mm->mmap_cache = vma;
1621        }
1622    }
1623    return vma;
1624}
1625
1626EXPORT_SYMBOL(find_vma);
1627
1628/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1629struct vm_area_struct *
1630find_vma_prev(struct mm_struct *mm, unsigned long addr,
1631            struct vm_area_struct **pprev)
1632{
1633    struct vm_area_struct *vma = NULL, *prev = NULL;
1634    struct rb_node *rb_node;
1635    if (!mm)
1636        goto out;
1637
1638    /* Guard against addr being lower than the first VMA */
1639    vma = mm->mmap;
1640
1641    /* Go through the RB tree quickly. */
1642    rb_node = mm->mm_rb.rb_node;
1643
1644    while (rb_node) {
1645        struct vm_area_struct *vma_tmp;
1646        vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1647
1648        if (addr < vma_tmp->vm_end) {
1649            rb_node = rb_node->rb_left;
1650        } else {
1651            prev = vma_tmp;
1652            if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1653                break;
1654            rb_node = rb_node->rb_right;
1655        }
1656    }
1657
1658out:
1659    *pprev = prev;
1660    return prev ? prev->vm_next : vma;
1661}
1662
1663/*
1664 * Verify that the stack growth is acceptable and
1665 * update accounting. This is shared with both the
1666 * grow-up and grow-down cases.
1667 */
1668static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1669{
1670    struct mm_struct *mm = vma->vm_mm;
1671    struct rlimit *rlim = current->signal->rlim;
1672    unsigned long new_start;
1673
1674    /* address space limit tests */
1675    if (!may_expand_vm(mm, grow))
1676        return -ENOMEM;
1677
1678    /* Stack limit test */
1679    if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1680        return -ENOMEM;
1681
1682    /* mlock limit tests */
1683    if (vma->vm_flags & VM_LOCKED) {
1684        unsigned long locked;
1685        unsigned long limit;
1686        locked = mm->locked_vm + grow;
1687        limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1688        limit >>= PAGE_SHIFT;
1689        if (locked > limit && !capable(CAP_IPC_LOCK))
1690            return -ENOMEM;
1691    }
1692
1693    /* Check to ensure the stack will not grow into a hugetlb-only region */
1694    new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1695            vma->vm_end - size;
1696    if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1697        return -EFAULT;
1698
1699    /*
1700     * Overcommit.. This must be the final test, as it will
1701     * update security statistics.
1702     */
1703    if (security_vm_enough_memory_mm(mm, grow))
1704        return -ENOMEM;
1705
1706    /* Ok, everything looks good - let it rip */
1707    mm->total_vm += grow;
1708    if (vma->vm_flags & VM_LOCKED)
1709        mm->locked_vm += grow;
1710    vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1711    return 0;
1712}
1713
1714#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1715/*
1716 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1717 * vma is the last one with address > vma->vm_end. Have to extend vma.
1718 */
1719int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1720{
1721    int error;
1722
1723    if (!(vma->vm_flags & VM_GROWSUP))
1724        return -EFAULT;
1725
1726    /*
1727     * We must make sure the anon_vma is allocated
1728     * so that the anon_vma locking is not a noop.
1729     */
1730    if (unlikely(anon_vma_prepare(vma)))
1731        return -ENOMEM;
1732    vma_lock_anon_vma(vma);
1733
1734    /*
1735     * vma->vm_start/vm_end cannot change under us because the caller
1736     * is required to hold the mmap_sem in read mode. We need the
1737     * anon_vma lock to serialize against concurrent expand_stacks.
1738     * Also guard against wrapping around to address 0.
1739     */
1740    if (address < PAGE_ALIGN(address+4))
1741        address = PAGE_ALIGN(address+4);
1742    else {
1743        vma_unlock_anon_vma(vma);
1744        return -ENOMEM;
1745    }
1746    error = 0;
1747
1748    /* Somebody else might have raced and expanded it already */
1749    if (address > vma->vm_end) {
1750        unsigned long size, grow;
1751
1752        size = address - vma->vm_start;
1753        grow = (address - vma->vm_end) >> PAGE_SHIFT;
1754
1755        error = acct_stack_growth(vma, size, grow);
1756        if (!error) {
1757            vma->vm_end = address;
1758            perf_event_mmap(vma);
1759        }
1760    }
1761    vma_unlock_anon_vma(vma);
1762    return error;
1763}
1764#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1765
1766/*
1767 * vma is the first one with address < vma->vm_start. Have to extend vma.
1768 */
1769static int expand_downwards(struct vm_area_struct *vma,
1770                   unsigned long address)
1771{
1772    int error;
1773
1774    /*
1775     * We must make sure the anon_vma is allocated
1776     * so that the anon_vma locking is not a noop.
1777     */
1778    if (unlikely(anon_vma_prepare(vma)))
1779        return -ENOMEM;
1780
1781    address &= PAGE_MASK;
1782    error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1783    if (error)
1784        return error;
1785
1786    vma_lock_anon_vma(vma);
1787
1788    /*
1789     * vma->vm_start/vm_end cannot change under us because the caller
1790     * is required to hold the mmap_sem in read mode. We need the
1791     * anon_vma lock to serialize against concurrent expand_stacks.
1792     */
1793
1794    /* Somebody else might have raced and expanded it already */
1795    if (address < vma->vm_start) {
1796        unsigned long size, grow;
1797
1798        size = vma->vm_end - address;
1799        grow = (vma->vm_start - address) >> PAGE_SHIFT;
1800
1801        error = acct_stack_growth(vma, size, grow);
1802        if (!error) {
1803            vma->vm_start = address;
1804            vma->vm_pgoff -= grow;
1805            perf_event_mmap(vma);
1806        }
1807    }
1808    vma_unlock_anon_vma(vma);
1809    return error;
1810}
1811
1812int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1813{
1814    return expand_downwards(vma, address);
1815}
1816
1817#ifdef CONFIG_STACK_GROWSUP
1818int expand_stack(struct vm_area_struct *vma, unsigned long address)
1819{
1820    return expand_upwards(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, *prev;
1827
1828    addr &= PAGE_MASK;
1829    vma = find_vma_prev(mm, addr, &prev);
1830    if (vma && (vma->vm_start <= addr))
1831        return vma;
1832    if (!prev || expand_stack(prev, addr))
1833        return NULL;
1834    if (prev->vm_flags & VM_LOCKED) {
1835        mlock_vma_pages_range(prev, addr, prev->vm_end);
1836    }
1837    return prev;
1838}
1839#else
1840int expand_stack(struct vm_area_struct *vma, unsigned long address)
1841{
1842    return expand_downwards(vma, address);
1843}
1844
1845struct vm_area_struct *
1846find_extend_vma(struct mm_struct * mm, unsigned long addr)
1847{
1848    struct vm_area_struct * vma;
1849    unsigned long start;
1850
1851    addr &= PAGE_MASK;
1852    vma = find_vma(mm,addr);
1853    if (!vma)
1854        return NULL;
1855    if (vma->vm_start <= addr)
1856        return vma;
1857    if (!(vma->vm_flags & VM_GROWSDOWN))
1858        return NULL;
1859    start = vma->vm_start;
1860    if (expand_stack(vma, addr))
1861        return NULL;
1862    if (vma->vm_flags & VM_LOCKED) {
1863        mlock_vma_pages_range(vma, addr, start);
1864    }
1865    return vma;
1866}
1867#endif
1868
1869/*
1870 * Ok - we have the memory areas we should free on the vma list,
1871 * so release them, and do the vma updates.
1872 *
1873 * Called with the mm semaphore held.
1874 */
1875static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1876{
1877    /* Update high watermark before we lower total_vm */
1878    update_hiwater_vm(mm);
1879    do {
1880        long nrpages = vma_pages(vma);
1881
1882        mm->total_vm -= nrpages;
1883        vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1884        vma = remove_vma(vma);
1885    } while (vma);
1886    validate_mm(mm);
1887}
1888
1889/*
1890 * Get rid of page table information in the indicated region.
1891 *
1892 * Called with the mm semaphore held.
1893 */
1894static void unmap_region(struct mm_struct *mm,
1895        struct vm_area_struct *vma, struct vm_area_struct *prev,
1896        unsigned long start, unsigned long end)
1897{
1898    struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1899    struct mmu_gather *tlb;
1900    unsigned long nr_accounted = 0;
1901
1902    lru_add_drain();
1903    tlb = tlb_gather_mmu(mm, 0);
1904    update_hiwater_rss(mm);
1905    unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1906    vm_unacct_memory(nr_accounted);
1907    free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1908                 next? next->vm_start: 0);
1909    tlb_finish_mmu(tlb, start, end);
1910}
1911
1912/*
1913 * Create a list of vma's touched by the unmap, removing them from the mm's
1914 * vma list as we go..
1915 */
1916static void
1917detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1918    struct vm_area_struct *prev, unsigned long end)
1919{
1920    struct vm_area_struct **insertion_point;
1921    struct vm_area_struct *tail_vma = NULL;
1922    unsigned long addr;
1923
1924    insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1925    vma->vm_prev = NULL;
1926    do {
1927        rb_erase(&vma->vm_rb, &mm->mm_rb);
1928        mm->map_count--;
1929        tail_vma = vma;
1930        vma = vma->vm_next;
1931    } while (vma && vma->vm_start < end);
1932    *insertion_point = vma;
1933    if (vma)
1934        vma->vm_prev = prev;
1935    tail_vma->vm_next = NULL;
1936    if (mm->unmap_area == arch_unmap_area)
1937        addr = prev ? prev->vm_end : mm->mmap_base;
1938    else
1939        addr = vma ? vma->vm_start : mm->mmap_base;
1940    mm->unmap_area(mm, addr);
1941    mm->mmap_cache = NULL; /* Kill the cache. */
1942}
1943
1944/*
1945 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1946 * munmap path where it doesn't make sense to fail.
1947 */
1948static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1949          unsigned long addr, int new_below)
1950{
1951    struct mempolicy *pol;
1952    struct vm_area_struct *new;
1953    int err = -ENOMEM;
1954
1955    if (is_vm_hugetlb_page(vma) && (addr &
1956                    ~(huge_page_mask(hstate_vma(vma)))))
1957        return -EINVAL;
1958
1959    new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1960    if (!new)
1961        goto out_err;
1962
1963    /* most fields are the same, copy all, and then fixup */
1964    *new = *vma;
1965
1966    INIT_LIST_HEAD(&new->anon_vma_chain);
1967
1968    if (new_below)
1969        new->vm_end = addr;
1970    else {
1971        new->vm_start = addr;
1972        new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1973    }
1974
1975    pol = mpol_dup(vma_policy(vma));
1976    if (IS_ERR(pol)) {
1977        err = PTR_ERR(pol);
1978        goto out_free_vma;
1979    }
1980    vma_set_policy(new, pol);
1981
1982    if (anon_vma_clone(new, vma))
1983        goto out_free_mpol;
1984
1985    if (new->vm_file) {
1986        get_file(new->vm_file);
1987        if (vma->vm_flags & VM_EXECUTABLE)
1988            added_exe_file_vma(mm);
1989    }
1990
1991    if (new->vm_ops && new->vm_ops->open)
1992        new->vm_ops->open(new);
1993
1994    if (new_below)
1995        err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1996            ((addr - new->vm_start) >> PAGE_SHIFT), new);
1997    else
1998        err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1999
2000    /* Success. */
2001    if (!err)
2002        return 0;
2003
2004    /* Clean everything up if vma_adjust failed. */
2005    if (new->vm_ops && new->vm_ops->close)
2006        new->vm_ops->close(new);
2007    if (new->vm_file) {
2008        if (vma->vm_flags & VM_EXECUTABLE)
2009            removed_exe_file_vma(mm);
2010        fput(new->vm_file);
2011    }
2012    unlink_anon_vmas(new);
2013 out_free_mpol:
2014    mpol_put(pol);
2015 out_free_vma:
2016    kmem_cache_free(vm_area_cachep, new);
2017 out_err:
2018    return err;
2019}
2020
2021/*
2022 * Split a vma into two pieces at address 'addr', a new vma is allocated
2023 * either for the first part or the tail.
2024 */
2025int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2026          unsigned long addr, int new_below)
2027{
2028    if (mm->map_count >= sysctl_max_map_count)
2029        return -ENOMEM;
2030
2031    return __split_vma(mm, vma, addr, new_below);
2032}
2033
2034/* Munmap is split into 2 main parts -- this part which finds
2035 * what needs doing, and the areas themselves, which do the
2036 * work. This now handles partial unmappings.
2037 * Jeremy Fitzhardinge <jeremy@goop.org>
2038 */
2039int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2040{
2041    unsigned long end;
2042    struct vm_area_struct *vma, *prev, *last;
2043
2044    if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2045        return -EINVAL;
2046
2047    if ((len = PAGE_ALIGN(len)) == 0)
2048        return -EINVAL;
2049
2050    /* Find the first overlapping VMA */
2051    vma = find_vma_prev(mm, start, &prev);
2052    if (!vma)
2053        return 0;
2054    /* we have start < vma->vm_end */
2055
2056    /* if it doesn't overlap, we have nothing.. */
2057    end = start + len;
2058    if (vma->vm_start >= end)
2059        return 0;
2060
2061    /*
2062     * If we need to split any vma, do it now to save pain later.
2063     *
2064     * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2065     * unmapped vm_area_struct will remain in use: so lower split_vma
2066     * places tmp vma above, and higher split_vma places tmp vma below.
2067     */
2068    if (start > vma->vm_start) {
2069        int error;
2070
2071        /*
2072         * Make sure that map_count on return from munmap() will
2073         * not exceed its limit; but let map_count go just above
2074         * its limit temporarily, to help free resources as expected.
2075         */
2076        if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2077            return -ENOMEM;
2078
2079        error = __split_vma(mm, vma, start, 0);
2080        if (error)
2081            return error;
2082        prev = vma;
2083    }
2084
2085    /* Does it split the last one? */
2086    last = find_vma(mm, end);
2087    if (last && end > last->vm_start) {
2088        int error = __split_vma(mm, last, end, 1);
2089        if (error)
2090            return error;
2091    }
2092    vma = prev? prev->vm_next: mm->mmap;
2093
2094    /*
2095     * unlock any mlock()ed ranges before detaching vmas
2096     */
2097    if (mm->locked_vm) {
2098        struct vm_area_struct *tmp = vma;
2099        while (tmp && tmp->vm_start < end) {
2100            if (tmp->vm_flags & VM_LOCKED) {
2101                mm->locked_vm -= vma_pages(tmp);
2102                munlock_vma_pages_all(tmp);
2103            }
2104            tmp = tmp->vm_next;
2105        }
2106    }
2107
2108    /*
2109     * Remove the vma's, and unmap the actual pages
2110     */
2111    detach_vmas_to_be_unmapped(mm, vma, prev, end);
2112    unmap_region(mm, vma, prev, start, end);
2113
2114    /* Fix up all other VM information */
2115    remove_vma_list(mm, vma);
2116
2117    return 0;
2118}
2119
2120EXPORT_SYMBOL(do_munmap);
2121
2122SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2123{
2124    int ret;
2125    struct mm_struct *mm = current->mm;
2126
2127    profile_munmap(addr);
2128
2129    down_write(&mm->mmap_sem);
2130    ret = do_munmap(mm, addr, len);
2131    up_write(&mm->mmap_sem);
2132    return ret;
2133}
2134
2135static inline void verify_mm_writelocked(struct mm_struct *mm)
2136{
2137#ifdef CONFIG_DEBUG_VM
2138    if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2139        WARN_ON(1);
2140        up_read(&mm->mmap_sem);
2141    }
2142#endif
2143}
2144
2145/*
2146 * this is really a simplified "do_mmap". it only handles
2147 * anonymous maps. eventually we may be able to do some
2148 * brk-specific accounting here.
2149 */
2150unsigned long do_brk(unsigned long addr, unsigned long len)
2151{
2152    struct mm_struct * mm = current->mm;
2153    struct vm_area_struct * vma, * prev;
2154    unsigned long flags;
2155    struct rb_node ** rb_link, * rb_parent;
2156    pgoff_t pgoff = addr >> PAGE_SHIFT;
2157    int error;
2158
2159    len = PAGE_ALIGN(len);
2160    if (!len)
2161        return addr;
2162
2163    error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2164    if (error)
2165        return error;
2166
2167    flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2168
2169    error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2170    if (error & ~PAGE_MASK)
2171        return error;
2172
2173    /*
2174     * mlock MCL_FUTURE?
2175     */
2176    if (mm->def_flags & VM_LOCKED) {
2177        unsigned long locked, lock_limit;
2178        locked = len >> PAGE_SHIFT;
2179        locked += mm->locked_vm;
2180        lock_limit = rlimit(RLIMIT_MEMLOCK);
2181        lock_limit >>= PAGE_SHIFT;
2182        if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2183            return -EAGAIN;
2184    }
2185
2186    /*
2187     * mm->mmap_sem is required to protect against another thread
2188     * changing the mappings in case we sleep.
2189     */
2190    verify_mm_writelocked(mm);
2191
2192    /*
2193     * Clear old maps. this also does some error checking for us
2194     */
2195 munmap_back:
2196    vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2197    if (vma && vma->vm_start < addr + len) {
2198        if (do_munmap(mm, addr, len))
2199            return -ENOMEM;
2200        goto munmap_back;
2201    }
2202
2203    /* Check against address space limits *after* clearing old maps... */
2204    if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2205        return -ENOMEM;
2206
2207    if (mm->map_count > sysctl_max_map_count)
2208        return -ENOMEM;
2209
2210    if (security_vm_enough_memory(len >> PAGE_SHIFT))
2211        return -ENOMEM;
2212
2213    /* Can we just expand an old private anonymous mapping? */
2214    vma = vma_merge(mm, prev, addr, addr + len, flags,
2215                    NULL, NULL, pgoff, NULL);
2216    if (vma)
2217        goto out;
2218
2219    /*
2220     * create a vma struct for an anonymous mapping
2221     */
2222    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2223    if (!vma) {
2224        vm_unacct_memory(len >> PAGE_SHIFT);
2225        return -ENOMEM;
2226    }
2227
2228    INIT_LIST_HEAD(&vma->anon_vma_chain);
2229    vma->vm_mm = mm;
2230    vma->vm_start = addr;
2231    vma->vm_end = addr + len;
2232    vma->vm_pgoff = pgoff;
2233    vma->vm_flags = flags;
2234    vma->vm_page_prot = vm_get_page_prot(flags);
2235    vma_link(mm, vma, prev, rb_link, rb_parent);
2236out:
2237    perf_event_mmap(vma);
2238    mm->total_vm += len >> PAGE_SHIFT;
2239    if (flags & VM_LOCKED) {
2240        if (!mlock_vma_pages_range(vma, addr, addr + len))
2241            mm->locked_vm += (len >> PAGE_SHIFT);
2242    }
2243    return addr;
2244}
2245
2246EXPORT_SYMBOL(do_brk);
2247
2248/* Release all mmaps. */
2249void exit_mmap(struct mm_struct *mm)
2250{
2251    struct mmu_gather *tlb;
2252    struct vm_area_struct *vma;
2253    unsigned long nr_accounted = 0;
2254    unsigned long end;
2255
2256    /* mm's last user has gone, and its about to be pulled down */
2257    mmu_notifier_release(mm);
2258
2259    if (mm->locked_vm) {
2260        vma = mm->mmap;
2261        while (vma) {
2262            if (vma->vm_flags & VM_LOCKED)
2263                munlock_vma_pages_all(vma);
2264            vma = vma->vm_next;
2265        }
2266    }
2267
2268    arch_exit_mmap(mm);
2269
2270    vma = mm->mmap;
2271    if (!vma) /* Can happen if dup_mmap() received an OOM */
2272        return;
2273
2274    lru_add_drain();
2275    flush_cache_mm(mm);
2276    tlb = tlb_gather_mmu(mm, 1);
2277    /* update_hiwater_rss(mm) here? but nobody should be looking */
2278    /* Use -1 here to ensure all VMAs in the mm are unmapped */
2279    end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2280    vm_unacct_memory(nr_accounted);
2281
2282    free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2283    tlb_finish_mmu(tlb, 0, end);
2284
2285    /*
2286     * Walk the list again, actually closing and freeing it,
2287     * with preemption enabled, without holding any MM locks.
2288     */
2289    while (vma)
2290        vma = remove_vma(vma);
2291
2292    BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2293}
2294
2295/* Insert vm structure into process list sorted by address
2296 * and into the inode's i_mmap tree. If vm_file is non-NULL
2297 * then i_mmap_lock is taken here.
2298 */
2299int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2300{
2301    struct vm_area_struct * __vma, * prev;
2302    struct rb_node ** rb_link, * rb_parent;
2303
2304    /*
2305     * The vm_pgoff of a purely anonymous vma should be irrelevant
2306     * until its first write fault, when page's anon_vma and index
2307     * are set. But now set the vm_pgoff it will almost certainly
2308     * end up with (unless mremap moves it elsewhere before that
2309     * first wfault), so /proc/pid/maps tells a consistent story.
2310     *
2311     * By setting it to reflect the virtual start address of the
2312     * vma, merges and splits can happen in a seamless way, just
2313     * using the existing file pgoff checks and manipulations.
2314     * Similarly in do_mmap_pgoff and in do_brk.
2315     */
2316    if (!vma->vm_file) {
2317        BUG_ON(vma->anon_vma);
2318        vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2319    }
2320    __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2321    if (__vma && __vma->vm_start < vma->vm_end)
2322        return -ENOMEM;
2323    if ((vma->vm_flags & VM_ACCOUNT) &&
2324         security_vm_enough_memory_mm(mm, vma_pages(vma)))
2325        return -ENOMEM;
2326    vma_link(mm, vma, prev, rb_link, rb_parent);
2327    return 0;
2328}
2329
2330/*
2331 * Copy the vma structure to a new location in the same mm,
2332 * prior to moving page table entries, to effect an mremap move.
2333 */
2334struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2335    unsigned long addr, unsigned long len, pgoff_t pgoff)
2336{
2337    struct vm_area_struct *vma = *vmap;
2338    unsigned long vma_start = vma->vm_start;
2339    struct mm_struct *mm = vma->vm_mm;
2340    struct vm_area_struct *new_vma, *prev;
2341    struct rb_node **rb_link, *rb_parent;
2342    struct mempolicy *pol;
2343
2344    /*
2345     * If anonymous vma has not yet been faulted, update new pgoff
2346     * to match new location, to increase its chance of merging.
2347     */
2348    if (!vma->vm_file && !vma->anon_vma)
2349        pgoff = addr >> PAGE_SHIFT;
2350
2351    find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2352    new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2353            vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2354    if (new_vma) {
2355        /*
2356         * Source vma may have been merged into new_vma
2357         */
2358        if (vma_start >= new_vma->vm_start &&
2359            vma_start < new_vma->vm_end)
2360            *vmap = new_vma;
2361    } else {
2362        new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2363        if (new_vma) {
2364            *new_vma = *vma;
2365            pol = mpol_dup(vma_policy(vma));
2366            if (IS_ERR(pol))
2367                goto out_free_vma;
2368            INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2369            if (anon_vma_clone(new_vma, vma))
2370                goto out_free_mempol;
2371            vma_set_policy(new_vma, pol);
2372            new_vma->vm_start = addr;
2373            new_vma->vm_end = addr + len;
2374            new_vma->vm_pgoff = pgoff;
2375            if (new_vma->vm_file) {
2376                get_file(new_vma->vm_file);
2377                if (vma->vm_flags & VM_EXECUTABLE)
2378                    added_exe_file_vma(mm);
2379            }
2380            if (new_vma->vm_ops && new_vma->vm_ops->open)
2381                new_vma->vm_ops->open(new_vma);
2382            vma_link(mm, new_vma, prev, rb_link, rb_parent);
2383        }
2384    }
2385    return new_vma;
2386
2387 out_free_mempol:
2388    mpol_put(pol);
2389 out_free_vma:
2390    kmem_cache_free(vm_area_cachep, new_vma);
2391    return NULL;
2392}
2393
2394/*
2395 * Return true if the calling process may expand its vm space by the passed
2396 * number of pages
2397 */
2398int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2399{
2400    unsigned long cur = mm->total_vm; /* pages */
2401    unsigned long lim;
2402
2403    lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2404
2405    if (cur + npages > lim)
2406        return 0;
2407    return 1;
2408}
2409
2410
2411static int special_mapping_fault(struct vm_area_struct *vma,
2412                struct vm_fault *vmf)
2413{
2414    pgoff_t pgoff;
2415    struct page **pages;
2416
2417    /*
2418     * special mappings have no vm_file, and in that case, the mm
2419     * uses vm_pgoff internally. So we have to subtract it from here.
2420     * We are allowed to do this because we are the mm; do not copy
2421     * this code into drivers!
2422     */
2423    pgoff = vmf->pgoff - vma->vm_pgoff;
2424
2425    for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2426        pgoff--;
2427
2428    if (*pages) {
2429        struct page *page = *pages;
2430        get_page(page);
2431        vmf->page = page;
2432        return 0;
2433    }
2434
2435    return VM_FAULT_SIGBUS;
2436}
2437
2438/*
2439 * Having a close hook prevents vma merging regardless of flags.
2440 */
2441static void special_mapping_close(struct vm_area_struct *vma)
2442{
2443}
2444
2445static const struct vm_operations_struct special_mapping_vmops = {
2446    .close = special_mapping_close,
2447    .fault = special_mapping_fault,
2448};
2449
2450/*
2451 * Called with mm->mmap_sem held for writing.
2452 * Insert a new vma covering the given region, with the given flags.
2453 * Its pages are supplied by the given array of struct page *.
2454 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2455 * The region past the last page supplied will always produce SIGBUS.
2456 * The array pointer and the pages it points to are assumed to stay alive
2457 * for as long as this mapping might exist.
2458 */
2459int install_special_mapping(struct mm_struct *mm,
2460                unsigned long addr, unsigned long len,
2461                unsigned long vm_flags, struct page **pages)
2462{
2463    struct vm_area_struct *vma;
2464
2465    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2466    if (unlikely(vma == NULL))
2467        return -ENOMEM;
2468
2469    INIT_LIST_HEAD(&vma->anon_vma_chain);
2470    vma->vm_mm = mm;
2471    vma->vm_start = addr;
2472    vma->vm_end = addr + len;
2473
2474    vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2475    vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2476
2477    vma->vm_ops = &special_mapping_vmops;
2478    vma->vm_private_data = pages;
2479
2480    if (unlikely(insert_vm_struct(mm, vma))) {
2481        kmem_cache_free(vm_area_cachep, vma);
2482        return -ENOMEM;
2483    }
2484
2485    mm->total_vm += len >> PAGE_SHIFT;
2486
2487    perf_event_mmap(vma);
2488
2489    return 0;
2490}
2491
2492static DEFINE_MUTEX(mm_all_locks_mutex);
2493
2494static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2495{
2496    if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2497        /*
2498         * The LSB of head.next can't change from under us
2499         * because we hold the mm_all_locks_mutex.
2500         */
2501        spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem);
2502        /*
2503         * We can safely modify head.next after taking the
2504         * anon_vma->root->lock. If some other vma in this mm shares
2505         * the same anon_vma we won't take it again.
2506         *
2507         * No need of atomic instructions here, head.next
2508         * can't change from under us thanks to the
2509         * anon_vma->root->lock.
2510         */
2511        if (__test_and_set_bit(0, (unsigned long *)
2512                       &anon_vma->root->head.next))
2513            BUG();
2514    }
2515}
2516
2517static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2518{
2519    if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2520        /*
2521         * AS_MM_ALL_LOCKS can't change from under us because
2522         * we hold the mm_all_locks_mutex.
2523         *
2524         * Operations on ->flags have to be atomic because
2525         * even if AS_MM_ALL_LOCKS is stable thanks to the
2526         * mm_all_locks_mutex, there may be other cpus
2527         * changing other bitflags in parallel to us.
2528         */
2529        if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2530            BUG();
2531        spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2532    }
2533}
2534
2535/*
2536 * This operation locks against the VM for all pte/vma/mm related
2537 * operations that could ever happen on a certain mm. This includes
2538 * vmtruncate, try_to_unmap, and all page faults.
2539 *
2540 * The caller must take the mmap_sem in write mode before calling
2541 * mm_take_all_locks(). The caller isn't allowed to release the
2542 * mmap_sem until mm_drop_all_locks() returns.
2543 *
2544 * mmap_sem in write mode is required in order to block all operations
2545 * that could modify pagetables and free pages without need of
2546 * altering the vma layout (for example populate_range() with
2547 * nonlinear vmas). It's also needed in write mode to avoid new
2548 * anon_vmas to be associated with existing vmas.
2549 *
2550 * A single task can't take more than one mm_take_all_locks() in a row
2551 * or it would deadlock.
2552 *
2553 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2554 * mapping->flags avoid to take the same lock twice, if more than one
2555 * vma in this mm is backed by the same anon_vma or address_space.
2556 *
2557 * We can take all the locks in random order because the VM code
2558 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2559 * takes more than one of them in a row. Secondly we're protected
2560 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2561 *
2562 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2563 * that may have to take thousand of locks.
2564 *
2565 * mm_take_all_locks() can fail if it's interrupted by signals.
2566 */
2567int mm_take_all_locks(struct mm_struct *mm)
2568{
2569    struct vm_area_struct *vma;
2570    struct anon_vma_chain *avc;
2571    int ret = -EINTR;
2572
2573    BUG_ON(down_read_trylock(&mm->mmap_sem));
2574
2575    mutex_lock(&mm_all_locks_mutex);
2576
2577    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2578        if (signal_pending(current))
2579            goto out_unlock;
2580        if (vma->vm_file && vma->vm_file->f_mapping)
2581            vm_lock_mapping(mm, vma->vm_file->f_mapping);
2582    }
2583
2584    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2585        if (signal_pending(current))
2586            goto out_unlock;
2587        if (vma->anon_vma)
2588            list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2589                vm_lock_anon_vma(mm, avc->anon_vma);
2590    }
2591
2592    ret = 0;
2593
2594out_unlock:
2595    if (ret)
2596        mm_drop_all_locks(mm);
2597
2598    return ret;
2599}
2600
2601static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2602{
2603    if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2604        /*
2605         * The LSB of head.next can't change to 0 from under
2606         * us because we hold the mm_all_locks_mutex.
2607         *
2608         * We must however clear the bitflag before unlocking
2609         * the vma so the users using the anon_vma->head will
2610         * never see our bitflag.
2611         *
2612         * No need of atomic instructions here, head.next
2613         * can't change from under us until we release the
2614         * anon_vma->root->lock.
2615         */
2616        if (!__test_and_clear_bit(0, (unsigned long *)
2617                      &anon_vma->root->head.next))
2618            BUG();
2619        anon_vma_unlock(anon_vma);
2620    }
2621}
2622
2623static void vm_unlock_mapping(struct address_space *mapping)
2624{
2625    if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2626        /*
2627         * AS_MM_ALL_LOCKS can't change to 0 from under us
2628         * because we hold the mm_all_locks_mutex.
2629         */
2630        spin_unlock(&mapping->i_mmap_lock);
2631        if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2632                    &mapping->flags))
2633            BUG();
2634    }
2635}
2636
2637/*
2638 * The mmap_sem cannot be released by the caller until
2639 * mm_drop_all_locks() returns.
2640 */
2641void mm_drop_all_locks(struct mm_struct *mm)
2642{
2643    struct vm_area_struct *vma;
2644    struct anon_vma_chain *avc;
2645
2646    BUG_ON(down_read_trylock(&mm->mmap_sem));
2647    BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2648
2649    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2650        if (vma->anon_vma)
2651            list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2652                vm_unlock_anon_vma(avc->anon_vma);
2653        if (vma->vm_file && vma->vm_file->f_mapping)
2654            vm_unlock_mapping(vma->vm_file->f_mapping);
2655    }
2656
2657    mutex_unlock(&mm_all_locks_mutex);
2658}
2659
2660/*
2661 * initialise the VMA slab
2662 */
2663void __init mmap_init(void)
2664{
2665    int ret;
2666
2667    ret = percpu_counter_init(&vm_committed_as, 0);
2668    VM_BUG_ON(ret);
2669}
2670

Archive Download this file



interactive