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#include <linux/audit.h>
32#include <linux/khugepaged.h>
33
34#include <asm/uaccess.h>
35#include <asm/cacheflush.h>
36#include <asm/tlb.h>
37#include <asm/mmu_context.h>
38
39#include "internal.h"
40
41#ifndef arch_mmap_check
42#define arch_mmap_check(addr, len, flags) (0)
43#endif
44
45#ifndef arch_rebalance_pgtables
46#define arch_rebalance_pgtables(addr, len) (addr)
47#endif
48
49static void unmap_region(struct mm_struct *mm,
50        struct vm_area_struct *vma, struct vm_area_struct *prev,
51        unsigned long start, unsigned long end);
52
53/*
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
56 */
57#undef DEBUG_MM_RB
58
59/* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
62 *
63 * map_type prot
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
68 *
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 *
73 */
74pgprot_t protection_map[16] = {
75    __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
76    __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
77};
78
79pgprot_t vm_get_page_prot(unsigned long vm_flags)
80{
81    return __pgprot(pgprot_val(protection_map[vm_flags &
82                (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
83            pgprot_val(arch_vm_get_page_prot(vm_flags)));
84}
85EXPORT_SYMBOL(vm_get_page_prot);
86
87int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */
88int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */
89int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
90/*
91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
92 * other variables. It can be updated by several CPUs frequently.
93 */
94struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
95
96/*
97 * Check that a process has enough memory to allocate a new virtual
98 * mapping. 0 means there is enough memory for the allocation to
99 * succeed and -ENOMEM implies there is not.
100 *
101 * We currently support three overcommit policies, which are set via the
102 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
103 *
104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
105 * Additional code 2002 Jul 20 by Robert Love.
106 *
107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
108 *
109 * Note this is a helper function intended to be used by LSMs which
110 * wish to use this logic.
111 */
112int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
113{
114    unsigned long free, allowed;
115
116    vm_acct_memory(pages);
117
118    /*
119     * Sometimes we want to use more memory than we have
120     */
121    if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
122        return 0;
123
124    if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
125        unsigned long n;
126
127        free = global_page_state(NR_FILE_PAGES);
128        free += nr_swap_pages;
129
130        /*
131         * Any slabs which are created with the
132         * SLAB_RECLAIM_ACCOUNT flag claim to have contents
133         * which are reclaimable, under pressure. The dentry
134         * cache and most inode caches should fall into this
135         */
136        free += global_page_state(NR_SLAB_RECLAIMABLE);
137
138        /*
139         * Leave the last 3% for root
140         */
141        if (!cap_sys_admin)
142            free -= free / 32;
143
144        if (free > pages)
145            return 0;
146
147        /*
148         * nr_free_pages() is very expensive on large systems,
149         * only call if we're about to fail.
150         */
151        n = nr_free_pages();
152
153        /*
154         * Leave reserved pages. The pages are not for anonymous pages.
155         */
156        if (n <= totalreserve_pages)
157            goto error;
158        else
159            n -= totalreserve_pages;
160
161        /*
162         * Leave the last 3% for root
163         */
164        if (!cap_sys_admin)
165            n -= n / 32;
166        free += n;
167
168        if (free > pages)
169            return 0;
170
171        goto error;
172    }
173
174    allowed = (totalram_pages - hugetlb_total_pages())
175               * sysctl_overcommit_ratio / 100;
176    /*
177     * Leave the last 3% for root
178     */
179    if (!cap_sys_admin)
180        allowed -= allowed / 32;
181    allowed += total_swap_pages;
182
183    /* Don't let a single process grow too big:
184       leave 3% of the size of this process for other processes */
185    if (mm)
186        allowed -= mm->total_vm / 32;
187
188    if (percpu_counter_read_positive(&vm_committed_as) < allowed)
189        return 0;
190error:
191    vm_unacct_memory(pages);
192
193    return -ENOMEM;
194}
195
196/*
197 * Requires inode->i_mapping->i_mmap_mutex
198 */
199static void __remove_shared_vm_struct(struct vm_area_struct *vma,
200        struct file *file, struct address_space *mapping)
201{
202    if (vma->vm_flags & VM_DENYWRITE)
203        atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
204    if (vma->vm_flags & VM_SHARED)
205        mapping->i_mmap_writable--;
206
207    flush_dcache_mmap_lock(mapping);
208    if (unlikely(vma->vm_flags & VM_NONLINEAR))
209        list_del_init(&vma->shared.vm_set.list);
210    else
211        vma_prio_tree_remove(vma, &mapping->i_mmap);
212    flush_dcache_mmap_unlock(mapping);
213}
214
215/*
216 * Unlink a file-based vm structure from its prio_tree, to hide
217 * vma from rmap and vmtruncate before freeing its page tables.
218 */
219void unlink_file_vma(struct vm_area_struct *vma)
220{
221    struct file *file = vma->vm_file;
222
223    if (file) {
224        struct address_space *mapping = file->f_mapping;
225        mutex_lock(&mapping->i_mmap_mutex);
226        __remove_shared_vm_struct(vma, file, mapping);
227        mutex_unlock(&mapping->i_mmap_mutex);
228    }
229}
230
231/*
232 * Close a vm structure and free it, returning the next.
233 */
234static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
235{
236    struct vm_area_struct *next = vma->vm_next;
237
238    might_sleep();
239    if (vma->vm_ops && vma->vm_ops->close)
240        vma->vm_ops->close(vma);
241    if (vma->vm_file) {
242        fput(vma->vm_file);
243        if (vma->vm_flags & VM_EXECUTABLE)
244            removed_exe_file_vma(vma->vm_mm);
245    }
246    mpol_put(vma_policy(vma));
247    kmem_cache_free(vm_area_cachep, vma);
248    return next;
249}
250
251SYSCALL_DEFINE1(brk, unsigned long, brk)
252{
253    unsigned long rlim, retval;
254    unsigned long newbrk, oldbrk;
255    struct mm_struct *mm = current->mm;
256    unsigned long min_brk;
257
258    down_write(&mm->mmap_sem);
259
260#ifdef CONFIG_COMPAT_BRK
261    /*
262     * CONFIG_COMPAT_BRK can still be overridden by setting
263     * randomize_va_space to 2, which will still cause mm->start_brk
264     * to be arbitrarily shifted
265     */
266    if (current->brk_randomized)
267        min_brk = mm->start_brk;
268    else
269        min_brk = mm->end_data;
270#else
271    min_brk = mm->start_brk;
272#endif
273    if (brk < min_brk)
274        goto out;
275
276    /*
277     * Check against rlimit here. If this check is done later after the test
278     * of oldbrk with newbrk then it can escape the test and let the data
279     * segment grow beyond its set limit the in case where the limit is
280     * not page aligned -Ram Gupta
281     */
282    rlim = rlimit(RLIMIT_DATA);
283    if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
284            (mm->end_data - mm->start_data) > rlim)
285        goto out;
286
287    newbrk = PAGE_ALIGN(brk);
288    oldbrk = PAGE_ALIGN(mm->brk);
289    if (oldbrk == newbrk)
290        goto set_brk;
291
292    /* Always allow shrinking brk. */
293    if (brk <= mm->brk) {
294        if (!do_munmap(mm, newbrk, oldbrk-newbrk))
295            goto set_brk;
296        goto out;
297    }
298
299    /* Check against existing mmap mappings. */
300    if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
301        goto out;
302
303    /* Ok, looks good - let it rip. */
304    if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
305        goto out;
306set_brk:
307    mm->brk = brk;
308out:
309    retval = mm->brk;
310    up_write(&mm->mmap_sem);
311    return retval;
312}
313
314#ifdef DEBUG_MM_RB
315static int browse_rb(struct rb_root *root)
316{
317    int i = 0, j;
318    struct rb_node *nd, *pn = NULL;
319    unsigned long prev = 0, pend = 0;
320
321    for (nd = rb_first(root); nd; nd = rb_next(nd)) {
322        struct vm_area_struct *vma;
323        vma = rb_entry(nd, struct vm_area_struct, vm_rb);
324        if (vma->vm_start < prev)
325            printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
326        if (vma->vm_start < pend)
327            printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
328        if (vma->vm_start > vma->vm_end)
329            printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
330        i++;
331        pn = nd;
332        prev = vma->vm_start;
333        pend = vma->vm_end;
334    }
335    j = 0;
336    for (nd = pn; nd; nd = rb_prev(nd)) {
337        j++;
338    }
339    if (i != j)
340        printk("backwards %d, forwards %d\n", j, i), i = 0;
341    return i;
342}
343
344void validate_mm(struct mm_struct *mm)
345{
346    int bug = 0;
347    int i = 0;
348    struct vm_area_struct *tmp = mm->mmap;
349    while (tmp) {
350        tmp = tmp->vm_next;
351        i++;
352    }
353    if (i != mm->map_count)
354        printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
355    i = browse_rb(&mm->mm_rb);
356    if (i != mm->map_count)
357        printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
358    BUG_ON(bug);
359}
360#else
361#define validate_mm(mm) do { } while (0)
362#endif
363
364static struct vm_area_struct *
365find_vma_prepare(struct mm_struct *mm, unsigned long addr,
366        struct vm_area_struct **pprev, struct rb_node ***rb_link,
367        struct rb_node ** rb_parent)
368{
369    struct vm_area_struct * vma;
370    struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
371
372    __rb_link = &mm->mm_rb.rb_node;
373    rb_prev = __rb_parent = NULL;
374    vma = NULL;
375
376    while (*__rb_link) {
377        struct vm_area_struct *vma_tmp;
378
379        __rb_parent = *__rb_link;
380        vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
381
382        if (vma_tmp->vm_end > addr) {
383            vma = vma_tmp;
384            if (vma_tmp->vm_start <= addr)
385                break;
386            __rb_link = &__rb_parent->rb_left;
387        } else {
388            rb_prev = __rb_parent;
389            __rb_link = &__rb_parent->rb_right;
390        }
391    }
392
393    *pprev = NULL;
394    if (rb_prev)
395        *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
396    *rb_link = __rb_link;
397    *rb_parent = __rb_parent;
398    return vma;
399}
400
401void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
402        struct rb_node **rb_link, struct rb_node *rb_parent)
403{
404    rb_link_node(&vma->vm_rb, rb_parent, rb_link);
405    rb_insert_color(&vma->vm_rb, &mm->mm_rb);
406}
407
408static void __vma_link_file(struct vm_area_struct *vma)
409{
410    struct file *file;
411
412    file = vma->vm_file;
413    if (file) {
414        struct address_space *mapping = file->f_mapping;
415
416        if (vma->vm_flags & VM_DENYWRITE)
417            atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
418        if (vma->vm_flags & VM_SHARED)
419            mapping->i_mmap_writable++;
420
421        flush_dcache_mmap_lock(mapping);
422        if (unlikely(vma->vm_flags & VM_NONLINEAR))
423            vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
424        else
425            vma_prio_tree_insert(vma, &mapping->i_mmap);
426        flush_dcache_mmap_unlock(mapping);
427    }
428}
429
430static void
431__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
432    struct vm_area_struct *prev, struct rb_node **rb_link,
433    struct rb_node *rb_parent)
434{
435    __vma_link_list(mm, vma, prev, rb_parent);
436    __vma_link_rb(mm, vma, rb_link, rb_parent);
437}
438
439static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
440            struct vm_area_struct *prev, struct rb_node **rb_link,
441            struct rb_node *rb_parent)
442{
443    struct address_space *mapping = NULL;
444
445    if (vma->vm_file)
446        mapping = vma->vm_file->f_mapping;
447
448    if (mapping)
449        mutex_lock(&mapping->i_mmap_mutex);
450
451    __vma_link(mm, vma, prev, rb_link, rb_parent);
452    __vma_link_file(vma);
453
454    if (mapping)
455        mutex_unlock(&mapping->i_mmap_mutex);
456
457    mm->map_count++;
458    validate_mm(mm);
459}
460
461/*
462 * Helper for vma_adjust in the split_vma insert case:
463 * insert vm structure into list and rbtree and anon_vma,
464 * but it has already been inserted into prio_tree earlier.
465 */
466static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
467{
468    struct vm_area_struct *__vma, *prev;
469    struct rb_node **rb_link, *rb_parent;
470
471    __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
472    BUG_ON(__vma && __vma->vm_start < vma->vm_end);
473    __vma_link(mm, vma, prev, rb_link, rb_parent);
474    mm->map_count++;
475}
476
477static inline void
478__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
479        struct vm_area_struct *prev)
480{
481    struct vm_area_struct *next = vma->vm_next;
482
483    prev->vm_next = next;
484    if (next)
485        next->vm_prev = prev;
486    rb_erase(&vma->vm_rb, &mm->mm_rb);
487    if (mm->mmap_cache == vma)
488        mm->mmap_cache = prev;
489}
490
491/*
492 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
493 * is already present in an i_mmap tree without adjusting the tree.
494 * The following helper function should be used when such adjustments
495 * are necessary. The "insert" vma (if any) is to be inserted
496 * before we drop the necessary locks.
497 */
498int vma_adjust(struct vm_area_struct *vma, unsigned long start,
499    unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
500{
501    struct mm_struct *mm = vma->vm_mm;
502    struct vm_area_struct *next = vma->vm_next;
503    struct vm_area_struct *importer = NULL;
504    struct address_space *mapping = NULL;
505    struct prio_tree_root *root = NULL;
506    struct anon_vma *anon_vma = NULL;
507    struct file *file = vma->vm_file;
508    long adjust_next = 0;
509    int remove_next = 0;
510
511    if (next && !insert) {
512        struct vm_area_struct *exporter = NULL;
513
514        if (end >= next->vm_end) {
515            /*
516             * vma expands, overlapping all the next, and
517             * perhaps the one after too (mprotect case 6).
518             */
519again: remove_next = 1 + (end > next->vm_end);
520            end = next->vm_end;
521            exporter = next;
522            importer = vma;
523        } else if (end > next->vm_start) {
524            /*
525             * vma expands, overlapping part of the next:
526             * mprotect case 5 shifting the boundary up.
527             */
528            adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
529            exporter = next;
530            importer = vma;
531        } else if (end < vma->vm_end) {
532            /*
533             * vma shrinks, and !insert tells it's not
534             * split_vma inserting another: so it must be
535             * mprotect case 4 shifting the boundary down.
536             */
537            adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
538            exporter = vma;
539            importer = next;
540        }
541
542        /*
543         * Easily overlooked: when mprotect shifts the boundary,
544         * make sure the expanding vma has anon_vma set if the
545         * shrinking vma had, to cover any anon pages imported.
546         */
547        if (exporter && exporter->anon_vma && !importer->anon_vma) {
548            if (anon_vma_clone(importer, exporter))
549                return -ENOMEM;
550            importer->anon_vma = exporter->anon_vma;
551        }
552    }
553
554    if (file) {
555        mapping = file->f_mapping;
556        if (!(vma->vm_flags & VM_NONLINEAR))
557            root = &mapping->i_mmap;
558        mutex_lock(&mapping->i_mmap_mutex);
559        if (insert) {
560            /*
561             * Put into prio_tree now, so instantiated pages
562             * are visible to arm/parisc __flush_dcache_page
563             * throughout; but we cannot insert into address
564             * space until vma start or end is updated.
565             */
566            __vma_link_file(insert);
567        }
568    }
569
570    vma_adjust_trans_huge(vma, start, end, adjust_next);
571
572    /*
573     * When changing only vma->vm_end, we don't really need anon_vma
574     * lock. This is a fairly rare case by itself, but the anon_vma
575     * lock may be shared between many sibling processes. Skipping
576     * the lock for brk adjustments makes a difference sometimes.
577     */
578    if (vma->anon_vma && (importer || start != vma->vm_start)) {
579        anon_vma = vma->anon_vma;
580        anon_vma_lock(anon_vma);
581    }
582
583    if (root) {
584        flush_dcache_mmap_lock(mapping);
585        vma_prio_tree_remove(vma, root);
586        if (adjust_next)
587            vma_prio_tree_remove(next, root);
588    }
589
590    vma->vm_start = start;
591    vma->vm_end = end;
592    vma->vm_pgoff = pgoff;
593    if (adjust_next) {
594        next->vm_start += adjust_next << PAGE_SHIFT;
595        next->vm_pgoff += adjust_next;
596    }
597
598    if (root) {
599        if (adjust_next)
600            vma_prio_tree_insert(next, root);
601        vma_prio_tree_insert(vma, root);
602        flush_dcache_mmap_unlock(mapping);
603    }
604
605    if (remove_next) {
606        /*
607         * vma_merge has merged next into vma, and needs
608         * us to remove next before dropping the locks.
609         */
610        __vma_unlink(mm, next, vma);
611        if (file)
612            __remove_shared_vm_struct(next, file, mapping);
613    } else if (insert) {
614        /*
615         * split_vma has split insert from vma, and needs
616         * us to insert it before dropping the locks
617         * (it may either follow vma or precede it).
618         */
619        __insert_vm_struct(mm, insert);
620    }
621
622    if (anon_vma)
623        anon_vma_unlock(anon_vma);
624    if (mapping)
625        mutex_unlock(&mapping->i_mmap_mutex);
626
627    if (remove_next) {
628        if (file) {
629            fput(file);
630            if (next->vm_flags & VM_EXECUTABLE)
631                removed_exe_file_vma(mm);
632        }
633        if (next->anon_vma)
634            anon_vma_merge(vma, next);
635        mm->map_count--;
636        mpol_put(vma_policy(next));
637        kmem_cache_free(vm_area_cachep, next);
638        /*
639         * In mprotect's case 6 (see comments on vma_merge),
640         * we must remove another next too. It would clutter
641         * up the code too much to do both in one go.
642         */
643        if (remove_next == 2) {
644            next = vma->vm_next;
645            goto again;
646        }
647    }
648
649    validate_mm(mm);
650
651    return 0;
652}
653
654/*
655 * If the vma has a ->close operation then the driver probably needs to release
656 * per-vma resources, so we don't attempt to merge those.
657 */
658static inline int is_mergeable_vma(struct vm_area_struct *vma,
659            struct file *file, unsigned long vm_flags)
660{
661    /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
662    if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
663        return 0;
664    if (vma->vm_file != file)
665        return 0;
666    if (vma->vm_ops && vma->vm_ops->close)
667        return 0;
668    return 1;
669}
670
671static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
672                    struct anon_vma *anon_vma2,
673                    struct vm_area_struct *vma)
674{
675    /*
676     * The list_is_singular() test is to avoid merging VMA cloned from
677     * parents. This can improve scalability caused by anon_vma lock.
678     */
679    if ((!anon_vma1 || !anon_vma2) && (!vma ||
680        list_is_singular(&vma->anon_vma_chain)))
681        return 1;
682    return anon_vma1 == anon_vma2;
683}
684
685/*
686 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
687 * in front of (at a lower virtual address and file offset than) the vma.
688 *
689 * We cannot merge two vmas if they have differently assigned (non-NULL)
690 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
691 *
692 * We don't check here for the merged mmap wrapping around the end of pagecache
693 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
694 * wrap, nor mmaps which cover the final page at index -1UL.
695 */
696static int
697can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
698    struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
699{
700    if (is_mergeable_vma(vma, file, vm_flags) &&
701        is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
702        if (vma->vm_pgoff == vm_pgoff)
703            return 1;
704    }
705    return 0;
706}
707
708/*
709 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
710 * beyond (at a higher virtual address and file offset than) the vma.
711 *
712 * We cannot merge two vmas if they have differently assigned (non-NULL)
713 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
714 */
715static int
716can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
717    struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
718{
719    if (is_mergeable_vma(vma, file, vm_flags) &&
720        is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
721        pgoff_t vm_pglen;
722        vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
723        if (vma->vm_pgoff + vm_pglen == vm_pgoff)
724            return 1;
725    }
726    return 0;
727}
728
729/*
730 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
731 * whether that can be merged with its predecessor or its successor.
732 * Or both (it neatly fills a hole).
733 *
734 * In most cases - when called for mmap, brk or mremap - [addr,end) is
735 * certain not to be mapped by the time vma_merge is called; but when
736 * called for mprotect, it is certain to be already mapped (either at
737 * an offset within prev, or at the start of next), and the flags of
738 * this area are about to be changed to vm_flags - and the no-change
739 * case has already been eliminated.
740 *
741 * The following mprotect cases have to be considered, where AAAA is
742 * the area passed down from mprotect_fixup, never extending beyond one
743 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
744 *
745 * AAAA AAAA AAAA AAAA
746 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
747 * cannot merge might become might become might become
748 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
749 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
750 * mremap move: PPPPNNNNNNNN 8
751 * AAAA
752 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
753 * might become case 1 below case 2 below case 3 below
754 *
755 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
756 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
757 */
758struct vm_area_struct *vma_merge(struct mm_struct *mm,
759            struct vm_area_struct *prev, unsigned long addr,
760            unsigned long end, unsigned long vm_flags,
761                 struct anon_vma *anon_vma, struct file *file,
762            pgoff_t pgoff, struct mempolicy *policy)
763{
764    pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
765    struct vm_area_struct *area, *next;
766    int err;
767
768    /*
769     * We later require that vma->vm_flags == vm_flags,
770     * so this tests vma->vm_flags & VM_SPECIAL, too.
771     */
772    if (vm_flags & VM_SPECIAL)
773        return NULL;
774
775    if (prev)
776        next = prev->vm_next;
777    else
778        next = mm->mmap;
779    area = next;
780    if (next && next->vm_end == end) /* cases 6, 7, 8 */
781        next = next->vm_next;
782
783    /*
784     * Can it merge with the predecessor?
785     */
786    if (prev && prev->vm_end == addr &&
787              mpol_equal(vma_policy(prev), policy) &&
788            can_vma_merge_after(prev, vm_flags,
789                        anon_vma, file, pgoff)) {
790        /*
791         * OK, it can. Can we now merge in the successor as well?
792         */
793        if (next && end == next->vm_start &&
794                mpol_equal(policy, vma_policy(next)) &&
795                can_vma_merge_before(next, vm_flags,
796                    anon_vma, file, pgoff+pglen) &&
797                is_mergeable_anon_vma(prev->anon_vma,
798                              next->anon_vma, NULL)) {
799                            /* cases 1, 6 */
800            err = vma_adjust(prev, prev->vm_start,
801                next->vm_end, prev->vm_pgoff, NULL);
802        } else /* cases 2, 5, 7 */
803            err = vma_adjust(prev, prev->vm_start,
804                end, prev->vm_pgoff, NULL);
805        if (err)
806            return NULL;
807        khugepaged_enter_vma_merge(prev);
808        return prev;
809    }
810
811    /*
812     * Can this new request be merged in front of next?
813     */
814    if (next && end == next->vm_start &&
815             mpol_equal(policy, vma_policy(next)) &&
816            can_vma_merge_before(next, vm_flags,
817                    anon_vma, file, pgoff+pglen)) {
818        if (prev && addr < prev->vm_end) /* case 4 */
819            err = vma_adjust(prev, prev->vm_start,
820                addr, prev->vm_pgoff, NULL);
821        else /* cases 3, 8 */
822            err = vma_adjust(area, addr, next->vm_end,
823                next->vm_pgoff - pglen, NULL);
824        if (err)
825            return NULL;
826        khugepaged_enter_vma_merge(area);
827        return area;
828    }
829
830    return NULL;
831}
832
833/*
834 * Rough compatbility check to quickly see if it's even worth looking
835 * at sharing an anon_vma.
836 *
837 * They need to have the same vm_file, and the flags can only differ
838 * in things that mprotect may change.
839 *
840 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
841 * we can merge the two vma's. For example, we refuse to merge a vma if
842 * there is a vm_ops->close() function, because that indicates that the
843 * driver is doing some kind of reference counting. But that doesn't
844 * really matter for the anon_vma sharing case.
845 */
846static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
847{
848    return a->vm_end == b->vm_start &&
849        mpol_equal(vma_policy(a), vma_policy(b)) &&
850        a->vm_file == b->vm_file &&
851        !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
852        b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
853}
854
855/*
856 * Do some basic sanity checking to see if we can re-use the anon_vma
857 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
858 * the same as 'old', the other will be the new one that is trying
859 * to share the anon_vma.
860 *
861 * NOTE! This runs with mm_sem held for reading, so it is possible that
862 * the anon_vma of 'old' is concurrently in the process of being set up
863 * by another page fault trying to merge _that_. But that's ok: if it
864 * is being set up, that automatically means that it will be a singleton
865 * acceptable for merging, so we can do all of this optimistically. But
866 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
867 *
868 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
869 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
870 * is to return an anon_vma that is "complex" due to having gone through
871 * a fork).
872 *
873 * We also make sure that the two vma's are compatible (adjacent,
874 * and with the same memory policies). That's all stable, even with just
875 * a read lock on the mm_sem.
876 */
877static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
878{
879    if (anon_vma_compatible(a, b)) {
880        struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
881
882        if (anon_vma && list_is_singular(&old->anon_vma_chain))
883            return anon_vma;
884    }
885    return NULL;
886}
887
888/*
889 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
890 * neighbouring vmas for a suitable anon_vma, before it goes off
891 * to allocate a new anon_vma. It checks because a repetitive
892 * sequence of mprotects and faults may otherwise lead to distinct
893 * anon_vmas being allocated, preventing vma merge in subsequent
894 * mprotect.
895 */
896struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
897{
898    struct anon_vma *anon_vma;
899    struct vm_area_struct *near;
900
901    near = vma->vm_next;
902    if (!near)
903        goto try_prev;
904
905    anon_vma = reusable_anon_vma(near, vma, near);
906    if (anon_vma)
907        return anon_vma;
908try_prev:
909    near = vma->vm_prev;
910    if (!near)
911        goto none;
912
913    anon_vma = reusable_anon_vma(near, near, vma);
914    if (anon_vma)
915        return anon_vma;
916none:
917    /*
918     * There's no absolute need to look only at touching neighbours:
919     * we could search further afield for "compatible" anon_vmas.
920     * But it would probably just be a waste of time searching,
921     * or lead to too many vmas hanging off the same anon_vma.
922     * We're trying to allow mprotect remerging later on,
923     * not trying to minimize memory used for anon_vmas.
924     */
925    return NULL;
926}
927
928#ifdef CONFIG_PROC_FS
929void vm_stat_account(struct mm_struct *mm, unsigned long flags,
930                        struct file *file, long pages)
931{
932    const unsigned long stack_flags
933        = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
934
935    if (file) {
936        mm->shared_vm += pages;
937        if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
938            mm->exec_vm += pages;
939    } else if (flags & stack_flags)
940        mm->stack_vm += pages;
941    if (flags & (VM_RESERVED|VM_IO))
942        mm->reserved_vm += pages;
943}
944#endif /* CONFIG_PROC_FS */
945
946/*
947 * The caller must hold down_write(&current->mm->mmap_sem).
948 */
949
950unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
951            unsigned long len, unsigned long prot,
952            unsigned long flags, unsigned long pgoff)
953{
954    struct mm_struct * mm = current->mm;
955    struct inode *inode;
956    vm_flags_t vm_flags;
957    int error;
958    unsigned long reqprot = prot;
959
960    /*
961     * Does the application expect PROT_READ to imply PROT_EXEC?
962     *
963     * (the exception is when the underlying filesystem is noexec
964     * mounted, in which case we dont add PROT_EXEC.)
965     */
966    if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
967        if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
968            prot |= PROT_EXEC;
969
970    if (!len)
971        return -EINVAL;
972
973    if (!(flags & MAP_FIXED))
974        addr = round_hint_to_min(addr);
975
976    /* Careful about overflows.. */
977    len = PAGE_ALIGN(len);
978    if (!len)
979        return -ENOMEM;
980
981    /* offset overflow? */
982    if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
983               return -EOVERFLOW;
984
985    /* Too many mappings? */
986    if (mm->map_count > sysctl_max_map_count)
987        return -ENOMEM;
988
989    /* Obtain the address to map to. we verify (or select) it and ensure
990     * that it represents a valid section of the address space.
991     */
992    addr = get_unmapped_area(file, addr, len, pgoff, flags);
993    if (addr & ~PAGE_MASK)
994        return addr;
995
996    /* Do simple checking here so the lower-level routines won't have
997     * to. we assume access permissions have been handled by the open
998     * of the memory object, so we don't do any here.
999     */
1000    vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1001            mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1002
1003    if (flags & MAP_LOCKED)
1004        if (!can_do_mlock())
1005            return -EPERM;
1006
1007    /* mlock MCL_FUTURE? */
1008    if (vm_flags & VM_LOCKED) {
1009        unsigned long locked, lock_limit;
1010        locked = len >> PAGE_SHIFT;
1011        locked += mm->locked_vm;
1012        lock_limit = rlimit(RLIMIT_MEMLOCK);
1013        lock_limit >>= PAGE_SHIFT;
1014        if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1015            return -EAGAIN;
1016    }
1017
1018    inode = file ? file->f_path.dentry->d_inode : NULL;
1019
1020    if (file) {
1021        switch (flags & MAP_TYPE) {
1022        case MAP_SHARED:
1023            if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1024                return -EACCES;
1025
1026            /*
1027             * Make sure we don't allow writing to an append-only
1028             * file..
1029             */
1030            if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1031                return -EACCES;
1032
1033            /*
1034             * Make sure there are no mandatory locks on the file.
1035             */
1036            if (locks_verify_locked(inode))
1037                return -EAGAIN;
1038
1039            vm_flags |= VM_SHARED | VM_MAYSHARE;
1040            if (!(file->f_mode & FMODE_WRITE))
1041                vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1042
1043            /* fall through */
1044        case MAP_PRIVATE:
1045            if (!(file->f_mode & FMODE_READ))
1046                return -EACCES;
1047            if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1048                if (vm_flags & VM_EXEC)
1049                    return -EPERM;
1050                vm_flags &= ~VM_MAYEXEC;
1051            }
1052
1053            if (!file->f_op || !file->f_op->mmap)
1054                return -ENODEV;
1055            break;
1056
1057        default:
1058            return -EINVAL;
1059        }
1060    } else {
1061        switch (flags & MAP_TYPE) {
1062        case MAP_SHARED:
1063            /*
1064             * Ignore pgoff.
1065             */
1066            pgoff = 0;
1067            vm_flags |= VM_SHARED | VM_MAYSHARE;
1068            break;
1069        case MAP_PRIVATE:
1070            /*
1071             * Set pgoff according to addr for anon_vma.
1072             */
1073            pgoff = addr >> PAGE_SHIFT;
1074            break;
1075        default:
1076            return -EINVAL;
1077        }
1078    }
1079
1080    error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1081    if (error)
1082        return error;
1083
1084    return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1085}
1086EXPORT_SYMBOL(do_mmap_pgoff);
1087
1088SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1089        unsigned long, prot, unsigned long, flags,
1090        unsigned long, fd, unsigned long, pgoff)
1091{
1092    struct file *file = NULL;
1093    unsigned long retval = -EBADF;
1094
1095    if (!(flags & MAP_ANONYMOUS)) {
1096        audit_mmap_fd(fd, flags);
1097        if (unlikely(flags & MAP_HUGETLB))
1098            return -EINVAL;
1099        file = fget(fd);
1100        if (!file)
1101            goto out;
1102    } else if (flags & MAP_HUGETLB) {
1103        struct user_struct *user = NULL;
1104        /*
1105         * VM_NORESERVE is used because the reservations will be
1106         * taken when vm_ops->mmap() is called
1107         * A dummy user value is used because we are not locking
1108         * memory so no accounting is necessary
1109         */
1110        len = ALIGN(len, huge_page_size(&default_hstate));
1111        file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1112                        &user, HUGETLB_ANONHUGE_INODE);
1113        if (IS_ERR(file))
1114            return PTR_ERR(file);
1115    }
1116
1117    flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1118
1119    down_write(&current->mm->mmap_sem);
1120    retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1121    up_write(&current->mm->mmap_sem);
1122
1123    if (file)
1124        fput(file);
1125out:
1126    return retval;
1127}
1128
1129#ifdef __ARCH_WANT_SYS_OLD_MMAP
1130struct mmap_arg_struct {
1131    unsigned long addr;
1132    unsigned long len;
1133    unsigned long prot;
1134    unsigned long flags;
1135    unsigned long fd;
1136    unsigned long offset;
1137};
1138
1139SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1140{
1141    struct mmap_arg_struct a;
1142
1143    if (copy_from_user(&a, arg, sizeof(a)))
1144        return -EFAULT;
1145    if (a.offset & ~PAGE_MASK)
1146        return -EINVAL;
1147
1148    return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1149                  a.offset >> PAGE_SHIFT);
1150}
1151#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1152
1153/*
1154 * Some shared mappigns will want the pages marked read-only
1155 * to track write events. If so, we'll downgrade vm_page_prot
1156 * to the private version (using protection_map[] without the
1157 * VM_SHARED bit).
1158 */
1159int vma_wants_writenotify(struct vm_area_struct *vma)
1160{
1161    vm_flags_t vm_flags = vma->vm_flags;
1162
1163    /* If it was private or non-writable, the write bit is already clear */
1164    if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1165        return 0;
1166
1167    /* The backer wishes to know when pages are first written to? */
1168    if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1169        return 1;
1170
1171    /* The open routine did something to the protections already? */
1172    if (pgprot_val(vma->vm_page_prot) !=
1173        pgprot_val(vm_get_page_prot(vm_flags)))
1174        return 0;
1175
1176    /* Specialty mapping? */
1177    if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1178        return 0;
1179
1180    /* Can the mapping track the dirty pages? */
1181    return vma->vm_file && vma->vm_file->f_mapping &&
1182        mapping_cap_account_dirty(vma->vm_file->f_mapping);
1183}
1184
1185/*
1186 * We account for memory if it's a private writeable mapping,
1187 * not hugepages and VM_NORESERVE wasn't set.
1188 */
1189static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1190{
1191    /*
1192     * hugetlb has its own accounting separate from the core VM
1193     * VM_HUGETLB may not be set yet so we cannot check for that flag.
1194     */
1195    if (file && is_file_hugepages(file))
1196        return 0;
1197
1198    return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1199}
1200
1201unsigned long mmap_region(struct file *file, unsigned long addr,
1202              unsigned long len, unsigned long flags,
1203              vm_flags_t vm_flags, unsigned long pgoff)
1204{
1205    struct mm_struct *mm = current->mm;
1206    struct vm_area_struct *vma, *prev;
1207    int correct_wcount = 0;
1208    int error;
1209    struct rb_node **rb_link, *rb_parent;
1210    unsigned long charged = 0;
1211    struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1212
1213    /* Clear old maps */
1214    error = -ENOMEM;
1215munmap_back:
1216    vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1217    if (vma && vma->vm_start < addr + len) {
1218        if (do_munmap(mm, addr, len))
1219            return -ENOMEM;
1220        goto munmap_back;
1221    }
1222
1223    /* Check against address space limit. */
1224    if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1225        return -ENOMEM;
1226
1227    /*
1228     * Set 'VM_NORESERVE' if we should not account for the
1229     * memory use of this mapping.
1230     */
1231    if ((flags & MAP_NORESERVE)) {
1232        /* We honor MAP_NORESERVE if allowed to overcommit */
1233        if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1234            vm_flags |= VM_NORESERVE;
1235
1236        /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1237        if (file && is_file_hugepages(file))
1238            vm_flags |= VM_NORESERVE;
1239    }
1240
1241    /*
1242     * Private writable mapping: check memory availability
1243     */
1244    if (accountable_mapping(file, vm_flags)) {
1245        charged = len >> PAGE_SHIFT;
1246        if (security_vm_enough_memory(charged))
1247            return -ENOMEM;
1248        vm_flags |= VM_ACCOUNT;
1249    }
1250
1251    /*
1252     * Can we just expand an old mapping?
1253     */
1254    vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1255    if (vma)
1256        goto out;
1257
1258    /*
1259     * Determine the object being mapped and call the appropriate
1260     * specific mapper. the address has already been validated, but
1261     * not unmapped, but the maps are removed from the list.
1262     */
1263    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1264    if (!vma) {
1265        error = -ENOMEM;
1266        goto unacct_error;
1267    }
1268
1269    vma->vm_mm = mm;
1270    vma->vm_start = addr;
1271    vma->vm_end = addr + len;
1272    vma->vm_flags = vm_flags;
1273    vma->vm_page_prot = vm_get_page_prot(vm_flags);
1274    vma->vm_pgoff = pgoff;
1275    INIT_LIST_HEAD(&vma->anon_vma_chain);
1276
1277    if (file) {
1278        error = -EINVAL;
1279        if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1280            goto free_vma;
1281        if (vm_flags & VM_DENYWRITE) {
1282            error = deny_write_access(file);
1283            if (error)
1284                goto free_vma;
1285            correct_wcount = 1;
1286        }
1287        vma->vm_file = file;
1288        get_file(file);
1289        error = file->f_op->mmap(file, vma);
1290        if (error)
1291            goto unmap_and_free_vma;
1292        if (vm_flags & VM_EXECUTABLE)
1293            added_exe_file_vma(mm);
1294
1295        /* Can addr have changed??
1296         *
1297         * Answer: Yes, several device drivers can do it in their
1298         * f_op->mmap method. -DaveM
1299         */
1300        addr = vma->vm_start;
1301        pgoff = vma->vm_pgoff;
1302        vm_flags = vma->vm_flags;
1303    } else if (vm_flags & VM_SHARED) {
1304        error = shmem_zero_setup(vma);
1305        if (error)
1306            goto free_vma;
1307    }
1308
1309    if (vma_wants_writenotify(vma)) {
1310        pgprot_t pprot = vma->vm_page_prot;
1311
1312        /* Can vma->vm_page_prot have changed??
1313         *
1314         * Answer: Yes, drivers may have changed it in their
1315         * f_op->mmap method.
1316         *
1317         * Ensures that vmas marked as uncached stay that way.
1318         */
1319        vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1320        if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1321            vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1322    }
1323
1324    vma_link(mm, vma, prev, rb_link, rb_parent);
1325    file = vma->vm_file;
1326
1327    /* Once vma denies write, undo our temporary denial count */
1328    if (correct_wcount)
1329        atomic_inc(&inode->i_writecount);
1330out:
1331    perf_event_mmap(vma);
1332
1333    mm->total_vm += len >> PAGE_SHIFT;
1334    vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1335    if (vm_flags & VM_LOCKED) {
1336        if (!mlock_vma_pages_range(vma, addr, addr + len))
1337            mm->locked_vm += (len >> PAGE_SHIFT);
1338    } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1339        make_pages_present(addr, addr + len);
1340    return addr;
1341
1342unmap_and_free_vma:
1343    if (correct_wcount)
1344        atomic_inc(&inode->i_writecount);
1345    vma->vm_file = NULL;
1346    fput(file);
1347
1348    /* Undo any partial mapping done by a device driver. */
1349    unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1350    charged = 0;
1351free_vma:
1352    kmem_cache_free(vm_area_cachep, vma);
1353unacct_error:
1354    if (charged)
1355        vm_unacct_memory(charged);
1356    return error;
1357}
1358
1359/* Get an address range which is currently unmapped.
1360 * For shmat() with addr=0.
1361 *
1362 * Ugly calling convention alert:
1363 * Return value with the low bits set means error value,
1364 * ie
1365 * if (ret & ~PAGE_MASK)
1366 * error = ret;
1367 *
1368 * This function "knows" that -ENOMEM has the bits set.
1369 */
1370#ifndef HAVE_ARCH_UNMAPPED_AREA
1371unsigned long
1372arch_get_unmapped_area(struct file *filp, unsigned long addr,
1373        unsigned long len, unsigned long pgoff, unsigned long flags)
1374{
1375    struct mm_struct *mm = current->mm;
1376    struct vm_area_struct *vma;
1377    unsigned long start_addr;
1378
1379    if (len > TASK_SIZE)
1380        return -ENOMEM;
1381
1382    if (flags & MAP_FIXED)
1383        return addr;
1384
1385    if (addr) {
1386        addr = PAGE_ALIGN(addr);
1387        vma = find_vma(mm, addr);
1388        if (TASK_SIZE - len >= addr &&
1389            (!vma || addr + len <= vma->vm_start))
1390            return addr;
1391    }
1392    if (len > mm->cached_hole_size) {
1393            start_addr = addr = mm->free_area_cache;
1394    } else {
1395            start_addr = addr = TASK_UNMAPPED_BASE;
1396            mm->cached_hole_size = 0;
1397    }
1398
1399full_search:
1400    for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1401        /* At this point: (!vma || addr < vma->vm_end). */
1402        if (TASK_SIZE - len < addr) {
1403            /*
1404             * Start a new search - just in case we missed
1405             * some holes.
1406             */
1407            if (start_addr != TASK_UNMAPPED_BASE) {
1408                addr = TASK_UNMAPPED_BASE;
1409                    start_addr = addr;
1410                mm->cached_hole_size = 0;
1411                goto full_search;
1412            }
1413            return -ENOMEM;
1414        }
1415        if (!vma || addr + len <= vma->vm_start) {
1416            /*
1417             * Remember the place where we stopped the search:
1418             */
1419            mm->free_area_cache = addr + len;
1420            return addr;
1421        }
1422        if (addr + mm->cached_hole_size < vma->vm_start)
1423                mm->cached_hole_size = vma->vm_start - addr;
1424        addr = vma->vm_end;
1425    }
1426}
1427#endif
1428
1429void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1430{
1431    /*
1432     * Is this a new hole at the lowest possible address?
1433     */
1434    if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1435        mm->free_area_cache = addr;
1436        mm->cached_hole_size = ~0UL;
1437    }
1438}
1439
1440/*
1441 * This mmap-allocator allocates new areas top-down from below the
1442 * stack's low limit (the base):
1443 */
1444#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1445unsigned long
1446arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1447              const unsigned long len, const unsigned long pgoff,
1448              const unsigned long flags)
1449{
1450    struct vm_area_struct *vma;
1451    struct mm_struct *mm = current->mm;
1452    unsigned long addr = addr0;
1453
1454    /* requested length too big for entire address space */
1455    if (len > TASK_SIZE)
1456        return -ENOMEM;
1457
1458    if (flags & MAP_FIXED)
1459        return addr;
1460
1461    /* requesting a specific address */
1462    if (addr) {
1463        addr = PAGE_ALIGN(addr);
1464        vma = find_vma(mm, addr);
1465        if (TASK_SIZE - len >= addr &&
1466                (!vma || addr + len <= vma->vm_start))
1467            return addr;
1468    }
1469
1470    /* check if free_area_cache is useful for us */
1471    if (len <= mm->cached_hole_size) {
1472             mm->cached_hole_size = 0;
1473         mm->free_area_cache = mm->mmap_base;
1474     }
1475
1476    /* either no address requested or can't fit in requested address hole */
1477    addr = mm->free_area_cache;
1478
1479    /* make sure it can fit in the remaining address space */
1480    if (addr > len) {
1481        vma = find_vma(mm, addr-len);
1482        if (!vma || addr <= vma->vm_start)
1483            /* remember the address as a hint for next time */
1484            return (mm->free_area_cache = addr-len);
1485    }
1486
1487    if (mm->mmap_base < len)
1488        goto bottomup;
1489
1490    addr = mm->mmap_base-len;
1491
1492    do {
1493        /*
1494         * Lookup failure means no vma is above this address,
1495         * else if new region fits below vma->vm_start,
1496         * return with success:
1497         */
1498        vma = find_vma(mm, addr);
1499        if (!vma || addr+len <= vma->vm_start)
1500            /* remember the address as a hint for next time */
1501            return (mm->free_area_cache = addr);
1502
1503         /* remember the largest hole we saw so far */
1504         if (addr + mm->cached_hole_size < vma->vm_start)
1505                 mm->cached_hole_size = vma->vm_start - addr;
1506
1507        /* try just below the current vma->vm_start */
1508        addr = vma->vm_start-len;
1509    } while (len < vma->vm_start);
1510
1511bottomup:
1512    /*
1513     * A failed mmap() very likely causes application failure,
1514     * so fall back to the bottom-up function here. This scenario
1515     * can happen with large stack limits and large mmap()
1516     * allocations.
1517     */
1518    mm->cached_hole_size = ~0UL;
1519      mm->free_area_cache = TASK_UNMAPPED_BASE;
1520    addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1521    /*
1522     * Restore the topdown base:
1523     */
1524    mm->free_area_cache = mm->mmap_base;
1525    mm->cached_hole_size = ~0UL;
1526
1527    return addr;
1528}
1529#endif
1530
1531void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1532{
1533    /*
1534     * Is this a new hole at the highest possible address?
1535     */
1536    if (addr > mm->free_area_cache)
1537        mm->free_area_cache = addr;
1538
1539    /* dont allow allocations above current base */
1540    if (mm->free_area_cache > mm->mmap_base)
1541        mm->free_area_cache = mm->mmap_base;
1542}
1543
1544unsigned long
1545get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1546        unsigned long pgoff, unsigned long flags)
1547{
1548    unsigned long (*get_area)(struct file *, unsigned long,
1549                  unsigned long, unsigned long, unsigned long);
1550
1551    unsigned long error = arch_mmap_check(addr, len, flags);
1552    if (error)
1553        return error;
1554
1555    /* Careful about overflows.. */
1556    if (len > TASK_SIZE)
1557        return -ENOMEM;
1558
1559    get_area = current->mm->get_unmapped_area;
1560    if (file && file->f_op && file->f_op->get_unmapped_area)
1561        get_area = file->f_op->get_unmapped_area;
1562    addr = get_area(file, addr, len, pgoff, flags);
1563    if (IS_ERR_VALUE(addr))
1564        return addr;
1565
1566    if (addr > TASK_SIZE - len)
1567        return -ENOMEM;
1568    if (addr & ~PAGE_MASK)
1569        return -EINVAL;
1570
1571    return arch_rebalance_pgtables(addr, len);
1572}
1573
1574EXPORT_SYMBOL(get_unmapped_area);
1575
1576/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1577struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1578{
1579    struct vm_area_struct *vma = NULL;
1580
1581    if (mm) {
1582        /* Check the cache first. */
1583        /* (Cache hit rate is typically around 35%.) */
1584        vma = mm->mmap_cache;
1585        if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1586            struct rb_node * rb_node;
1587
1588            rb_node = mm->mm_rb.rb_node;
1589            vma = NULL;
1590
1591            while (rb_node) {
1592                struct vm_area_struct * vma_tmp;
1593
1594                vma_tmp = rb_entry(rb_node,
1595                        struct vm_area_struct, vm_rb);
1596
1597                if (vma_tmp->vm_end > addr) {
1598                    vma = vma_tmp;
1599                    if (vma_tmp->vm_start <= addr)
1600                        break;
1601                    rb_node = rb_node->rb_left;
1602                } else
1603                    rb_node = rb_node->rb_right;
1604            }
1605            if (vma)
1606                mm->mmap_cache = vma;
1607        }
1608    }
1609    return vma;
1610}
1611
1612EXPORT_SYMBOL(find_vma);
1613
1614/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1615struct vm_area_struct *
1616find_vma_prev(struct mm_struct *mm, unsigned long addr,
1617            struct vm_area_struct **pprev)
1618{
1619    struct vm_area_struct *vma = NULL, *prev = NULL;
1620    struct rb_node *rb_node;
1621    if (!mm)
1622        goto out;
1623
1624    /* Guard against addr being lower than the first VMA */
1625    vma = mm->mmap;
1626
1627    /* Go through the RB tree quickly. */
1628    rb_node = mm->mm_rb.rb_node;
1629
1630    while (rb_node) {
1631        struct vm_area_struct *vma_tmp;
1632        vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1633
1634        if (addr < vma_tmp->vm_end) {
1635            rb_node = rb_node->rb_left;
1636        } else {
1637            prev = vma_tmp;
1638            if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1639                break;
1640            rb_node = rb_node->rb_right;
1641        }
1642    }
1643
1644out:
1645    *pprev = prev;
1646    return prev ? prev->vm_next : vma;
1647}
1648
1649/*
1650 * Verify that the stack growth is acceptable and
1651 * update accounting. This is shared with both the
1652 * grow-up and grow-down cases.
1653 */
1654static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1655{
1656    struct mm_struct *mm = vma->vm_mm;
1657    struct rlimit *rlim = current->signal->rlim;
1658    unsigned long new_start;
1659
1660    /* address space limit tests */
1661    if (!may_expand_vm(mm, grow))
1662        return -ENOMEM;
1663
1664    /* Stack limit test */
1665    if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1666        return -ENOMEM;
1667
1668    /* mlock limit tests */
1669    if (vma->vm_flags & VM_LOCKED) {
1670        unsigned long locked;
1671        unsigned long limit;
1672        locked = mm->locked_vm + grow;
1673        limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1674        limit >>= PAGE_SHIFT;
1675        if (locked > limit && !capable(CAP_IPC_LOCK))
1676            return -ENOMEM;
1677    }
1678
1679    /* Check to ensure the stack will not grow into a hugetlb-only region */
1680    new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1681            vma->vm_end - size;
1682    if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1683        return -EFAULT;
1684
1685    /*
1686     * Overcommit.. This must be the final test, as it will
1687     * update security statistics.
1688     */
1689    if (security_vm_enough_memory_mm(mm, grow))
1690        return -ENOMEM;
1691
1692    /* Ok, everything looks good - let it rip */
1693    mm->total_vm += grow;
1694    if (vma->vm_flags & VM_LOCKED)
1695        mm->locked_vm += grow;
1696    vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1697    return 0;
1698}
1699
1700#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1701/*
1702 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1703 * vma is the last one with address > vma->vm_end. Have to extend vma.
1704 */
1705int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1706{
1707    int error;
1708
1709    if (!(vma->vm_flags & VM_GROWSUP))
1710        return -EFAULT;
1711
1712    /*
1713     * We must make sure the anon_vma is allocated
1714     * so that the anon_vma locking is not a noop.
1715     */
1716    if (unlikely(anon_vma_prepare(vma)))
1717        return -ENOMEM;
1718    vma_lock_anon_vma(vma);
1719
1720    /*
1721     * vma->vm_start/vm_end cannot change under us because the caller
1722     * is required to hold the mmap_sem in read mode. We need the
1723     * anon_vma lock to serialize against concurrent expand_stacks.
1724     * Also guard against wrapping around to address 0.
1725     */
1726    if (address < PAGE_ALIGN(address+4))
1727        address = PAGE_ALIGN(address+4);
1728    else {
1729        vma_unlock_anon_vma(vma);
1730        return -ENOMEM;
1731    }
1732    error = 0;
1733
1734    /* Somebody else might have raced and expanded it already */
1735    if (address > vma->vm_end) {
1736        unsigned long size, grow;
1737
1738        size = address - vma->vm_start;
1739        grow = (address - vma->vm_end) >> PAGE_SHIFT;
1740
1741        error = -ENOMEM;
1742        if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1743            error = acct_stack_growth(vma, size, grow);
1744            if (!error) {
1745                vma->vm_end = address;
1746                perf_event_mmap(vma);
1747            }
1748        }
1749    }
1750    vma_unlock_anon_vma(vma);
1751    khugepaged_enter_vma_merge(vma);
1752    return error;
1753}
1754#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1755
1756/*
1757 * vma is the first one with address < vma->vm_start. Have to extend vma.
1758 */
1759int expand_downwards(struct vm_area_struct *vma,
1760                   unsigned long address)
1761{
1762    int error;
1763
1764    /*
1765     * We must make sure the anon_vma is allocated
1766     * so that the anon_vma locking is not a noop.
1767     */
1768    if (unlikely(anon_vma_prepare(vma)))
1769        return -ENOMEM;
1770
1771    address &= PAGE_MASK;
1772    error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1773    if (error)
1774        return error;
1775
1776    vma_lock_anon_vma(vma);
1777
1778    /*
1779     * vma->vm_start/vm_end cannot change under us because the caller
1780     * is required to hold the mmap_sem in read mode. We need the
1781     * anon_vma lock to serialize against concurrent expand_stacks.
1782     */
1783
1784    /* Somebody else might have raced and expanded it already */
1785    if (address < vma->vm_start) {
1786        unsigned long size, grow;
1787
1788        size = vma->vm_end - address;
1789        grow = (vma->vm_start - address) >> PAGE_SHIFT;
1790
1791        error = -ENOMEM;
1792        if (grow <= vma->vm_pgoff) {
1793            error = acct_stack_growth(vma, size, grow);
1794            if (!error) {
1795                vma->vm_start = address;
1796                vma->vm_pgoff -= grow;
1797                perf_event_mmap(vma);
1798            }
1799        }
1800    }
1801    vma_unlock_anon_vma(vma);
1802    khugepaged_enter_vma_merge(vma);
1803    return error;
1804}
1805
1806#ifdef CONFIG_STACK_GROWSUP
1807int expand_stack(struct vm_area_struct *vma, unsigned long address)
1808{
1809    return expand_upwards(vma, address);
1810}
1811
1812struct vm_area_struct *
1813find_extend_vma(struct mm_struct *mm, unsigned long addr)
1814{
1815    struct vm_area_struct *vma, *prev;
1816
1817    addr &= PAGE_MASK;
1818    vma = find_vma_prev(mm, addr, &prev);
1819    if (vma && (vma->vm_start <= addr))
1820        return vma;
1821    if (!prev || expand_stack(prev, addr))
1822        return NULL;
1823    if (prev->vm_flags & VM_LOCKED) {
1824        mlock_vma_pages_range(prev, addr, prev->vm_end);
1825    }
1826    return prev;
1827}
1828#else
1829int expand_stack(struct vm_area_struct *vma, unsigned long address)
1830{
1831    return expand_downwards(vma, address);
1832}
1833
1834struct vm_area_struct *
1835find_extend_vma(struct mm_struct * mm, unsigned long addr)
1836{
1837    struct vm_area_struct * vma;
1838    unsigned long start;
1839
1840    addr &= PAGE_MASK;
1841    vma = find_vma(mm,addr);
1842    if (!vma)
1843        return NULL;
1844    if (vma->vm_start <= addr)
1845        return vma;
1846    if (!(vma->vm_flags & VM_GROWSDOWN))
1847        return NULL;
1848    start = vma->vm_start;
1849    if (expand_stack(vma, addr))
1850        return NULL;
1851    if (vma->vm_flags & VM_LOCKED) {
1852        mlock_vma_pages_range(vma, addr, start);
1853    }
1854    return vma;
1855}
1856#endif
1857
1858/*
1859 * Ok - we have the memory areas we should free on the vma list,
1860 * so release them, and do the vma updates.
1861 *
1862 * Called with the mm semaphore held.
1863 */
1864static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1865{
1866    /* Update high watermark before we lower total_vm */
1867    update_hiwater_vm(mm);
1868    do {
1869        long nrpages = vma_pages(vma);
1870
1871        mm->total_vm -= nrpages;
1872        vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1873        vma = remove_vma(vma);
1874    } while (vma);
1875    validate_mm(mm);
1876}
1877
1878/*
1879 * Get rid of page table information in the indicated region.
1880 *
1881 * Called with the mm semaphore held.
1882 */
1883static void unmap_region(struct mm_struct *mm,
1884        struct vm_area_struct *vma, struct vm_area_struct *prev,
1885        unsigned long start, unsigned long end)
1886{
1887    struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1888    struct mmu_gather tlb;
1889    unsigned long nr_accounted = 0;
1890
1891    lru_add_drain();
1892    tlb_gather_mmu(&tlb, mm, 0);
1893    update_hiwater_rss(mm);
1894    unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1895    vm_unacct_memory(nr_accounted);
1896    free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1897                 next ? next->vm_start : 0);
1898    tlb_finish_mmu(&tlb, start, end);
1899}
1900
1901/*
1902 * Create a list of vma's touched by the unmap, removing them from the mm's
1903 * vma list as we go..
1904 */
1905static void
1906detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1907    struct vm_area_struct *prev, unsigned long end)
1908{
1909    struct vm_area_struct **insertion_point;
1910    struct vm_area_struct *tail_vma = NULL;
1911    unsigned long addr;
1912
1913    insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1914    vma->vm_prev = NULL;
1915    do {
1916        rb_erase(&vma->vm_rb, &mm->mm_rb);
1917        mm->map_count--;
1918        tail_vma = vma;
1919        vma = vma->vm_next;
1920    } while (vma && vma->vm_start < end);
1921    *insertion_point = vma;
1922    if (vma)
1923        vma->vm_prev = prev;
1924    tail_vma->vm_next = NULL;
1925    if (mm->unmap_area == arch_unmap_area)
1926        addr = prev ? prev->vm_end : mm->mmap_base;
1927    else
1928        addr = vma ? vma->vm_start : mm->mmap_base;
1929    mm->unmap_area(mm, addr);
1930    mm->mmap_cache = NULL; /* Kill the cache. */
1931}
1932
1933/*
1934 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1935 * munmap path where it doesn't make sense to fail.
1936 */
1937static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1938          unsigned long addr, int new_below)
1939{
1940    struct mempolicy *pol;
1941    struct vm_area_struct *new;
1942    int err = -ENOMEM;
1943
1944    if (is_vm_hugetlb_page(vma) && (addr &
1945                    ~(huge_page_mask(hstate_vma(vma)))))
1946        return -EINVAL;
1947
1948    new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1949    if (!new)
1950        goto out_err;
1951
1952    /* most fields are the same, copy all, and then fixup */
1953    *new = *vma;
1954
1955    INIT_LIST_HEAD(&new->anon_vma_chain);
1956
1957    if (new_below)
1958        new->vm_end = addr;
1959    else {
1960        new->vm_start = addr;
1961        new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1962    }
1963
1964    pol = mpol_dup(vma_policy(vma));
1965    if (IS_ERR(pol)) {
1966        err = PTR_ERR(pol);
1967        goto out_free_vma;
1968    }
1969    vma_set_policy(new, pol);
1970
1971    if (anon_vma_clone(new, vma))
1972        goto out_free_mpol;
1973
1974    if (new->vm_file) {
1975        get_file(new->vm_file);
1976        if (vma->vm_flags & VM_EXECUTABLE)
1977            added_exe_file_vma(mm);
1978    }
1979
1980    if (new->vm_ops && new->vm_ops->open)
1981        new->vm_ops->open(new);
1982
1983    if (new_below)
1984        err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1985            ((addr - new->vm_start) >> PAGE_SHIFT), new);
1986    else
1987        err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1988
1989    /* Success. */
1990    if (!err)
1991        return 0;
1992
1993    /* Clean everything up if vma_adjust failed. */
1994    if (new->vm_ops && new->vm_ops->close)
1995        new->vm_ops->close(new);
1996    if (new->vm_file) {
1997        if (vma->vm_flags & VM_EXECUTABLE)
1998            removed_exe_file_vma(mm);
1999        fput(new->vm_file);
2000    }
2001    unlink_anon_vmas(new);
2002 out_free_mpol:
2003    mpol_put(pol);
2004 out_free_vma:
2005    kmem_cache_free(vm_area_cachep, new);
2006 out_err:
2007    return err;
2008}
2009
2010/*
2011 * Split a vma into two pieces at address 'addr', a new vma is allocated
2012 * either for the first part or the tail.
2013 */
2014int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2015          unsigned long addr, int new_below)
2016{
2017    if (mm->map_count >= sysctl_max_map_count)
2018        return -ENOMEM;
2019
2020    return __split_vma(mm, vma, addr, new_below);
2021}
2022
2023/* Munmap is split into 2 main parts -- this part which finds
2024 * what needs doing, and the areas themselves, which do the
2025 * work. This now handles partial unmappings.
2026 * Jeremy Fitzhardinge <jeremy@goop.org>
2027 */
2028int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2029{
2030    unsigned long end;
2031    struct vm_area_struct *vma, *prev, *last;
2032
2033    if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2034        return -EINVAL;
2035
2036    if ((len = PAGE_ALIGN(len)) == 0)
2037        return -EINVAL;
2038
2039    /* Find the first overlapping VMA */
2040    vma = find_vma(mm, start);
2041    if (!vma)
2042        return 0;
2043    prev = vma->vm_prev;
2044    /* we have start < vma->vm_end */
2045
2046    /* if it doesn't overlap, we have nothing.. */
2047    end = start + len;
2048    if (vma->vm_start >= end)
2049        return 0;
2050
2051    /*
2052     * If we need to split any vma, do it now to save pain later.
2053     *
2054     * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2055     * unmapped vm_area_struct will remain in use: so lower split_vma
2056     * places tmp vma above, and higher split_vma places tmp vma below.
2057     */
2058    if (start > vma->vm_start) {
2059        int error;
2060
2061        /*
2062         * Make sure that map_count on return from munmap() will
2063         * not exceed its limit; but let map_count go just above
2064         * its limit temporarily, to help free resources as expected.
2065         */
2066        if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2067            return -ENOMEM;
2068
2069        error = __split_vma(mm, vma, start, 0);
2070        if (error)
2071            return error;
2072        prev = vma;
2073    }
2074
2075    /* Does it split the last one? */
2076    last = find_vma(mm, end);
2077    if (last && end > last->vm_start) {
2078        int error = __split_vma(mm, last, end, 1);
2079        if (error)
2080            return error;
2081    }
2082    vma = prev? prev->vm_next: mm->mmap;
2083
2084    /*
2085     * unlock any mlock()ed ranges before detaching vmas
2086     */
2087    if (mm->locked_vm) {
2088        struct vm_area_struct *tmp = vma;
2089        while (tmp && tmp->vm_start < end) {
2090            if (tmp->vm_flags & VM_LOCKED) {
2091                mm->locked_vm -= vma_pages(tmp);
2092                munlock_vma_pages_all(tmp);
2093            }
2094            tmp = tmp->vm_next;
2095        }
2096    }
2097
2098    /*
2099     * Remove the vma's, and unmap the actual pages
2100     */
2101    detach_vmas_to_be_unmapped(mm, vma, prev, end);
2102    unmap_region(mm, vma, prev, start, end);
2103
2104    /* Fix up all other VM information */
2105    remove_vma_list(mm, vma);
2106
2107    return 0;
2108}
2109
2110EXPORT_SYMBOL(do_munmap);
2111
2112SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2113{
2114    int ret;
2115    struct mm_struct *mm = current->mm;
2116
2117    profile_munmap(addr);
2118
2119    down_write(&mm->mmap_sem);
2120    ret = do_munmap(mm, addr, len);
2121    up_write(&mm->mmap_sem);
2122    return ret;
2123}
2124
2125static inline void verify_mm_writelocked(struct mm_struct *mm)
2126{
2127#ifdef CONFIG_DEBUG_VM
2128    if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2129        WARN_ON(1);
2130        up_read(&mm->mmap_sem);
2131    }
2132#endif
2133}
2134
2135/*
2136 * this is really a simplified "do_mmap". it only handles
2137 * anonymous maps. eventually we may be able to do some
2138 * brk-specific accounting here.
2139 */
2140unsigned long do_brk(unsigned long addr, unsigned long len)
2141{
2142    struct mm_struct * mm = current->mm;
2143    struct vm_area_struct * vma, * prev;
2144    unsigned long flags;
2145    struct rb_node ** rb_link, * rb_parent;
2146    pgoff_t pgoff = addr >> PAGE_SHIFT;
2147    int error;
2148
2149    len = PAGE_ALIGN(len);
2150    if (!len)
2151        return addr;
2152
2153    error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2154    if (error)
2155        return error;
2156
2157    flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2158
2159    error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2160    if (error & ~PAGE_MASK)
2161        return error;
2162
2163    /*
2164     * mlock MCL_FUTURE?
2165     */
2166    if (mm->def_flags & VM_LOCKED) {
2167        unsigned long locked, lock_limit;
2168        locked = len >> PAGE_SHIFT;
2169        locked += mm->locked_vm;
2170        lock_limit = rlimit(RLIMIT_MEMLOCK);
2171        lock_limit >>= PAGE_SHIFT;
2172        if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2173            return -EAGAIN;
2174    }
2175
2176    /*
2177     * mm->mmap_sem is required to protect against another thread
2178     * changing the mappings in case we sleep.
2179     */
2180    verify_mm_writelocked(mm);
2181
2182    /*
2183     * Clear old maps. this also does some error checking for us
2184     */
2185 munmap_back:
2186    vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2187    if (vma && vma->vm_start < addr + len) {
2188        if (do_munmap(mm, addr, len))
2189            return -ENOMEM;
2190        goto munmap_back;
2191    }
2192
2193    /* Check against address space limits *after* clearing old maps... */
2194    if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2195        return -ENOMEM;
2196
2197    if (mm->map_count > sysctl_max_map_count)
2198        return -ENOMEM;
2199
2200    if (security_vm_enough_memory(len >> PAGE_SHIFT))
2201        return -ENOMEM;
2202
2203    /* Can we just expand an old private anonymous mapping? */
2204    vma = vma_merge(mm, prev, addr, addr + len, flags,
2205                    NULL, NULL, pgoff, NULL);
2206    if (vma)
2207        goto out;
2208
2209    /*
2210     * create a vma struct for an anonymous mapping
2211     */
2212    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2213    if (!vma) {
2214        vm_unacct_memory(len >> PAGE_SHIFT);
2215        return -ENOMEM;
2216    }
2217
2218    INIT_LIST_HEAD(&vma->anon_vma_chain);
2219    vma->vm_mm = mm;
2220    vma->vm_start = addr;
2221    vma->vm_end = addr + len;
2222    vma->vm_pgoff = pgoff;
2223    vma->vm_flags = flags;
2224    vma->vm_page_prot = vm_get_page_prot(flags);
2225    vma_link(mm, vma, prev, rb_link, rb_parent);
2226out:
2227    perf_event_mmap(vma);
2228    mm->total_vm += len >> PAGE_SHIFT;
2229    if (flags & VM_LOCKED) {
2230        if (!mlock_vma_pages_range(vma, addr, addr + len))
2231            mm->locked_vm += (len >> PAGE_SHIFT);
2232    }
2233    return addr;
2234}
2235
2236EXPORT_SYMBOL(do_brk);
2237
2238/* Release all mmaps. */
2239void exit_mmap(struct mm_struct *mm)
2240{
2241    struct mmu_gather tlb;
2242    struct vm_area_struct *vma;
2243    unsigned long nr_accounted = 0;
2244    unsigned long end;
2245
2246    /* mm's last user has gone, and its about to be pulled down */
2247    mmu_notifier_release(mm);
2248
2249    if (mm->locked_vm) {
2250        vma = mm->mmap;
2251        while (vma) {
2252            if (vma->vm_flags & VM_LOCKED)
2253                munlock_vma_pages_all(vma);
2254            vma = vma->vm_next;
2255        }
2256    }
2257
2258    arch_exit_mmap(mm);
2259
2260    vma = mm->mmap;
2261    if (!vma) /* Can happen if dup_mmap() received an OOM */
2262        return;
2263
2264    lru_add_drain();
2265    flush_cache_mm(mm);
2266    tlb_gather_mmu(&tlb, mm, 1);
2267    /* update_hiwater_rss(mm) here? but nobody should be looking */
2268    /* Use -1 here to ensure all VMAs in the mm are unmapped */
2269    end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2270    vm_unacct_memory(nr_accounted);
2271
2272    free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2273    tlb_finish_mmu(&tlb, 0, end);
2274
2275    /*
2276     * Walk the list again, actually closing and freeing it,
2277     * with preemption enabled, without holding any MM locks.
2278     */
2279    while (vma)
2280        vma = remove_vma(vma);
2281
2282    BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2283}
2284
2285/* Insert vm structure into process list sorted by address
2286 * and into the inode's i_mmap tree. If vm_file is non-NULL
2287 * then i_mmap_mutex is taken here.
2288 */
2289int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2290{
2291    struct vm_area_struct * __vma, * prev;
2292    struct rb_node ** rb_link, * rb_parent;
2293
2294    /*
2295     * The vm_pgoff of a purely anonymous vma should be irrelevant
2296     * until its first write fault, when page's anon_vma and index
2297     * are set. But now set the vm_pgoff it will almost certainly
2298     * end up with (unless mremap moves it elsewhere before that
2299     * first wfault), so /proc/pid/maps tells a consistent story.
2300     *
2301     * By setting it to reflect the virtual start address of the
2302     * vma, merges and splits can happen in a seamless way, just
2303     * using the existing file pgoff checks and manipulations.
2304     * Similarly in do_mmap_pgoff and in do_brk.
2305     */
2306    if (!vma->vm_file) {
2307        BUG_ON(vma->anon_vma);
2308        vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2309    }
2310    __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2311    if (__vma && __vma->vm_start < vma->vm_end)
2312        return -ENOMEM;
2313    if ((vma->vm_flags & VM_ACCOUNT) &&
2314         security_vm_enough_memory_mm(mm, vma_pages(vma)))
2315        return -ENOMEM;
2316    vma_link(mm, vma, prev, rb_link, rb_parent);
2317    return 0;
2318}
2319
2320/*
2321 * Copy the vma structure to a new location in the same mm,
2322 * prior to moving page table entries, to effect an mremap move.
2323 */
2324struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2325    unsigned long addr, unsigned long len, pgoff_t pgoff)
2326{
2327    struct vm_area_struct *vma = *vmap;
2328    unsigned long vma_start = vma->vm_start;
2329    struct mm_struct *mm = vma->vm_mm;
2330    struct vm_area_struct *new_vma, *prev;
2331    struct rb_node **rb_link, *rb_parent;
2332    struct mempolicy *pol;
2333
2334    /*
2335     * If anonymous vma has not yet been faulted, update new pgoff
2336     * to match new location, to increase its chance of merging.
2337     */
2338    if (!vma->vm_file && !vma->anon_vma)
2339        pgoff = addr >> PAGE_SHIFT;
2340
2341    find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2342    new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2343            vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2344    if (new_vma) {
2345        /*
2346         * Source vma may have been merged into new_vma
2347         */
2348        if (vma_start >= new_vma->vm_start &&
2349            vma_start < new_vma->vm_end)
2350            *vmap = new_vma;
2351    } else {
2352        new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2353        if (new_vma) {
2354            *new_vma = *vma;
2355            pol = mpol_dup(vma_policy(vma));
2356            if (IS_ERR(pol))
2357                goto out_free_vma;
2358            INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2359            if (anon_vma_clone(new_vma, vma))
2360                goto out_free_mempol;
2361            vma_set_policy(new_vma, pol);
2362            new_vma->vm_start = addr;
2363            new_vma->vm_end = addr + len;
2364            new_vma->vm_pgoff = pgoff;
2365            if (new_vma->vm_file) {
2366                get_file(new_vma->vm_file);
2367                if (vma->vm_flags & VM_EXECUTABLE)
2368                    added_exe_file_vma(mm);
2369            }
2370            if (new_vma->vm_ops && new_vma->vm_ops->open)
2371                new_vma->vm_ops->open(new_vma);
2372            vma_link(mm, new_vma, prev, rb_link, rb_parent);
2373        }
2374    }
2375    return new_vma;
2376
2377 out_free_mempol:
2378    mpol_put(pol);
2379 out_free_vma:
2380    kmem_cache_free(vm_area_cachep, new_vma);
2381    return NULL;
2382}
2383
2384/*
2385 * Return true if the calling process may expand its vm space by the passed
2386 * number of pages
2387 */
2388int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2389{
2390    unsigned long cur = mm->total_vm; /* pages */
2391    unsigned long lim;
2392
2393    lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2394
2395    if (cur + npages > lim)
2396        return 0;
2397    return 1;
2398}
2399
2400
2401static int special_mapping_fault(struct vm_area_struct *vma,
2402                struct vm_fault *vmf)
2403{
2404    pgoff_t pgoff;
2405    struct page **pages;
2406
2407    /*
2408     * special mappings have no vm_file, and in that case, the mm
2409     * uses vm_pgoff internally. So we have to subtract it from here.
2410     * We are allowed to do this because we are the mm; do not copy
2411     * this code into drivers!
2412     */
2413    pgoff = vmf->pgoff - vma->vm_pgoff;
2414
2415    for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2416        pgoff--;
2417
2418    if (*pages) {
2419        struct page *page = *pages;
2420        get_page(page);
2421        vmf->page = page;
2422        return 0;
2423    }
2424
2425    return VM_FAULT_SIGBUS;
2426}
2427
2428/*
2429 * Having a close hook prevents vma merging regardless of flags.
2430 */
2431static void special_mapping_close(struct vm_area_struct *vma)
2432{
2433}
2434
2435static const struct vm_operations_struct special_mapping_vmops = {
2436    .close = special_mapping_close,
2437    .fault = special_mapping_fault,
2438};
2439
2440/*
2441 * Called with mm->mmap_sem held for writing.
2442 * Insert a new vma covering the given region, with the given flags.
2443 * Its pages are supplied by the given array of struct page *.
2444 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2445 * The region past the last page supplied will always produce SIGBUS.
2446 * The array pointer and the pages it points to are assumed to stay alive
2447 * for as long as this mapping might exist.
2448 */
2449int install_special_mapping(struct mm_struct *mm,
2450                unsigned long addr, unsigned long len,
2451                unsigned long vm_flags, struct page **pages)
2452{
2453    int ret;
2454    struct vm_area_struct *vma;
2455
2456    vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2457    if (unlikely(vma == NULL))
2458        return -ENOMEM;
2459
2460    INIT_LIST_HEAD(&vma->anon_vma_chain);
2461    vma->vm_mm = mm;
2462    vma->vm_start = addr;
2463    vma->vm_end = addr + len;
2464
2465    vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2466    vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2467
2468    vma->vm_ops = &special_mapping_vmops;
2469    vma->vm_private_data = pages;
2470
2471    ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2472    if (ret)
2473        goto out;
2474
2475    ret = insert_vm_struct(mm, vma);
2476    if (ret)
2477        goto out;
2478
2479    mm->total_vm += len >> PAGE_SHIFT;
2480
2481    perf_event_mmap(vma);
2482
2483    return 0;
2484
2485out:
2486    kmem_cache_free(vm_area_cachep, vma);
2487    return ret;
2488}
2489
2490static DEFINE_MUTEX(mm_all_locks_mutex);
2491
2492static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2493{
2494    if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2495        /*
2496         * The LSB of head.next can't change from under us
2497         * because we hold the mm_all_locks_mutex.
2498         */
2499        mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2500        /*
2501         * We can safely modify head.next after taking the
2502         * anon_vma->root->mutex. If some other vma in this mm shares
2503         * the same anon_vma we won't take it again.
2504         *
2505         * No need of atomic instructions here, head.next
2506         * can't change from under us thanks to the
2507         * anon_vma->root->mutex.
2508         */
2509        if (__test_and_set_bit(0, (unsigned long *)
2510                       &anon_vma->root->head.next))
2511            BUG();
2512    }
2513}
2514
2515static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2516{
2517    if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2518        /*
2519         * AS_MM_ALL_LOCKS can't change from under us because
2520         * we hold the mm_all_locks_mutex.
2521         *
2522         * Operations on ->flags have to be atomic because
2523         * even if AS_MM_ALL_LOCKS is stable thanks to the
2524         * mm_all_locks_mutex, there may be other cpus
2525         * changing other bitflags in parallel to us.
2526         */
2527        if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2528            BUG();
2529        mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2530    }
2531}
2532
2533/*
2534 * This operation locks against the VM for all pte/vma/mm related
2535 * operations that could ever happen on a certain mm. This includes
2536 * vmtruncate, try_to_unmap, and all page faults.
2537 *
2538 * The caller must take the mmap_sem in write mode before calling
2539 * mm_take_all_locks(). The caller isn't allowed to release the
2540 * mmap_sem until mm_drop_all_locks() returns.
2541 *
2542 * mmap_sem in write mode is required in order to block all operations
2543 * that could modify pagetables and free pages without need of
2544 * altering the vma layout (for example populate_range() with
2545 * nonlinear vmas). It's also needed in write mode to avoid new
2546 * anon_vmas to be associated with existing vmas.
2547 *
2548 * A single task can't take more than one mm_take_all_locks() in a row
2549 * or it would deadlock.
2550 *
2551 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2552 * mapping->flags avoid to take the same lock twice, if more than one
2553 * vma in this mm is backed by the same anon_vma or address_space.
2554 *
2555 * We can take all the locks in random order because the VM code
2556 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2557 * takes more than one of them in a row. Secondly we're protected
2558 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2559 *
2560 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2561 * that may have to take thousand of locks.
2562 *
2563 * mm_take_all_locks() can fail if it's interrupted by signals.
2564 */
2565int mm_take_all_locks(struct mm_struct *mm)
2566{
2567    struct vm_area_struct *vma;
2568    struct anon_vma_chain *avc;
2569    int ret = -EINTR;
2570
2571    BUG_ON(down_read_trylock(&mm->mmap_sem));
2572
2573    mutex_lock(&mm_all_locks_mutex);
2574
2575    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2576        if (signal_pending(current))
2577            goto out_unlock;
2578        if (vma->vm_file && vma->vm_file->f_mapping)
2579            vm_lock_mapping(mm, vma->vm_file->f_mapping);
2580    }
2581
2582    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2583        if (signal_pending(current))
2584            goto out_unlock;
2585        if (vma->anon_vma)
2586            list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2587                vm_lock_anon_vma(mm, avc->anon_vma);
2588    }
2589
2590    ret = 0;
2591
2592out_unlock:
2593    if (ret)
2594        mm_drop_all_locks(mm);
2595
2596    return ret;
2597}
2598
2599static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2600{
2601    if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2602        /*
2603         * The LSB of head.next can't change to 0 from under
2604         * us because we hold the mm_all_locks_mutex.
2605         *
2606         * We must however clear the bitflag before unlocking
2607         * the vma so the users using the anon_vma->head will
2608         * never see our bitflag.
2609         *
2610         * No need of atomic instructions here, head.next
2611         * can't change from under us until we release the
2612         * anon_vma->root->mutex.
2613         */
2614        if (!__test_and_clear_bit(0, (unsigned long *)
2615                      &anon_vma->root->head.next))
2616            BUG();
2617        anon_vma_unlock(anon_vma);
2618    }
2619}
2620
2621static void vm_unlock_mapping(struct address_space *mapping)
2622{
2623    if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2624        /*
2625         * AS_MM_ALL_LOCKS can't change to 0 from under us
2626         * because we hold the mm_all_locks_mutex.
2627         */
2628        mutex_unlock(&mapping->i_mmap_mutex);
2629        if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2630                    &mapping->flags))
2631            BUG();
2632    }
2633}
2634
2635/*
2636 * The mmap_sem cannot be released by the caller until
2637 * mm_drop_all_locks() returns.
2638 */
2639void mm_drop_all_locks(struct mm_struct *mm)
2640{
2641    struct vm_area_struct *vma;
2642    struct anon_vma_chain *avc;
2643
2644    BUG_ON(down_read_trylock(&mm->mmap_sem));
2645    BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2646
2647    for (vma = mm->mmap; vma; vma = vma->vm_next) {
2648        if (vma->anon_vma)
2649            list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2650                vm_unlock_anon_vma(avc->anon_vma);
2651        if (vma->vm_file && vma->vm_file->f_mapping)
2652            vm_unlock_mapping(vma->vm_file->f_mapping);
2653    }
2654
2655    mutex_unlock(&mm_all_locks_mutex);
2656}
2657
2658/*
2659 * initialise the VMA slab
2660 */
2661void __init mmap_init(void)
2662{
2663    int ret;
2664
2665    ret = percpu_counter_init(&vm_committed_as, 0);
2666    VM_BUG_ON(ret);
2667}
2668

Archive Download this file



interactive