diff options
Diffstat (limited to 'mm/nommu.c')
-rw-r--r-- | mm/nommu.c | 1027 |
1 files changed, 701 insertions, 326 deletions
diff --git a/mm/nommu.c b/mm/nommu.c index 1c28ea3a4e9c..60ed8375c986 100644 --- a/mm/nommu.c +++ b/mm/nommu.c | |||
@@ -6,11 +6,11 @@ | |||
6 | * | 6 | * |
7 | * See Documentation/nommu-mmap.txt | 7 | * See Documentation/nommu-mmap.txt |
8 | * | 8 | * |
9 | * Copyright (c) 2004-2005 David Howells <dhowells@redhat.com> | 9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
10 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> | 10 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> |
11 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> | 11 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> |
12 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> | 12 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> |
13 | * Copyright (c) 2007 Paul Mundt <lethal@linux-sh.org> | 13 | * Copyright (c) 2007-2008 Paul Mundt <lethal@linux-sh.org> |
14 | */ | 14 | */ |
15 | 15 | ||
16 | #include <linux/module.h> | 16 | #include <linux/module.h> |
@@ -33,6 +33,28 @@ | |||
33 | #include <asm/uaccess.h> | 33 | #include <asm/uaccess.h> |
34 | #include <asm/tlb.h> | 34 | #include <asm/tlb.h> |
35 | #include <asm/tlbflush.h> | 35 | #include <asm/tlbflush.h> |
36 | #include "internal.h" | ||
37 | |||
38 | static inline __attribute__((format(printf, 1, 2))) | ||
39 | void no_printk(const char *fmt, ...) | ||
40 | { | ||
41 | } | ||
42 | |||
43 | #if 0 | ||
44 | #define kenter(FMT, ...) \ | ||
45 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | ||
46 | #define kleave(FMT, ...) \ | ||
47 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | ||
48 | #define kdebug(FMT, ...) \ | ||
49 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) | ||
50 | #else | ||
51 | #define kenter(FMT, ...) \ | ||
52 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | ||
53 | #define kleave(FMT, ...) \ | ||
54 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | ||
55 | #define kdebug(FMT, ...) \ | ||
56 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) | ||
57 | #endif | ||
36 | 58 | ||
37 | #include "internal.h" | 59 | #include "internal.h" |
38 | 60 | ||
@@ -40,19 +62,22 @@ void *high_memory; | |||
40 | struct page *mem_map; | 62 | struct page *mem_map; |
41 | unsigned long max_mapnr; | 63 | unsigned long max_mapnr; |
42 | unsigned long num_physpages; | 64 | unsigned long num_physpages; |
43 | unsigned long askedalloc, realalloc; | ||
44 | atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0); | 65 | atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0); |
45 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ | 66 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
46 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | 67 | int sysctl_overcommit_ratio = 50; /* default is 50% */ |
47 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; | 68 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; |
69 | int sysctl_nr_trim_pages = 1; /* page trimming behaviour */ | ||
48 | int heap_stack_gap = 0; | 70 | int heap_stack_gap = 0; |
49 | 71 | ||
72 | atomic_t mmap_pages_allocated; | ||
73 | |||
50 | EXPORT_SYMBOL(mem_map); | 74 | EXPORT_SYMBOL(mem_map); |
51 | EXPORT_SYMBOL(num_physpages); | 75 | EXPORT_SYMBOL(num_physpages); |
52 | 76 | ||
53 | /* list of shareable VMAs */ | 77 | /* list of mapped, potentially shareable regions */ |
54 | struct rb_root nommu_vma_tree = RB_ROOT; | 78 | static struct kmem_cache *vm_region_jar; |
55 | DECLARE_RWSEM(nommu_vma_sem); | 79 | struct rb_root nommu_region_tree = RB_ROOT; |
80 | DECLARE_RWSEM(nommu_region_sem); | ||
56 | 81 | ||
57 | struct vm_operations_struct generic_file_vm_ops = { | 82 | struct vm_operations_struct generic_file_vm_ops = { |
58 | }; | 83 | }; |
@@ -124,6 +149,20 @@ unsigned int kobjsize(const void *objp) | |||
124 | return ksize(objp); | 149 | return ksize(objp); |
125 | 150 | ||
126 | /* | 151 | /* |
152 | * If it's not a compound page, see if we have a matching VMA | ||
153 | * region. This test is intentionally done in reverse order, | ||
154 | * so if there's no VMA, we still fall through and hand back | ||
155 | * PAGE_SIZE for 0-order pages. | ||
156 | */ | ||
157 | if (!PageCompound(page)) { | ||
158 | struct vm_area_struct *vma; | ||
159 | |||
160 | vma = find_vma(current->mm, (unsigned long)objp); | ||
161 | if (vma) | ||
162 | return vma->vm_end - vma->vm_start; | ||
163 | } | ||
164 | |||
165 | /* | ||
127 | * The ksize() function is only guaranteed to work for pointers | 166 | * The ksize() function is only guaranteed to work for pointers |
128 | * returned by kmalloc(). So handle arbitrary pointers here. | 167 | * returned by kmalloc(). So handle arbitrary pointers here. |
129 | */ | 168 | */ |
@@ -401,129 +440,178 @@ asmlinkage unsigned long sys_brk(unsigned long brk) | |||
401 | return mm->brk = brk; | 440 | return mm->brk = brk; |
402 | } | 441 | } |
403 | 442 | ||
404 | #ifdef DEBUG | 443 | /* |
405 | static void show_process_blocks(void) | 444 | * initialise the VMA and region record slabs |
445 | */ | ||
446 | void __init mmap_init(void) | ||
406 | { | 447 | { |
407 | struct vm_list_struct *vml; | 448 | vm_region_jar = kmem_cache_create("vm_region_jar", |
408 | 449 | sizeof(struct vm_region), 0, | |
409 | printk("Process blocks %d:", current->pid); | 450 | SLAB_PANIC, NULL); |
410 | 451 | vm_area_cachep = kmem_cache_create("vm_area_struct", | |
411 | for (vml = ¤t->mm->context.vmlist; vml; vml = vml->next) { | 452 | sizeof(struct vm_area_struct), 0, |
412 | printk(" %p: %p", vml, vml->vma); | 453 | SLAB_PANIC, NULL); |
413 | if (vml->vma) | ||
414 | printk(" (%d @%lx #%d)", | ||
415 | kobjsize((void *) vml->vma->vm_start), | ||
416 | vml->vma->vm_start, | ||
417 | atomic_read(&vml->vma->vm_usage)); | ||
418 | printk(vml->next ? " ->" : ".\n"); | ||
419 | } | ||
420 | } | 454 | } |
421 | #endif /* DEBUG */ | ||
422 | 455 | ||
423 | /* | 456 | /* |
424 | * add a VMA into a process's mm_struct in the appropriate place in the list | 457 | * validate the region tree |
425 | * - should be called with mm->mmap_sem held writelocked | 458 | * - the caller must hold the region lock |
426 | */ | 459 | */ |
427 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_list_struct *vml) | 460 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
461 | static noinline void validate_nommu_regions(void) | ||
428 | { | 462 | { |
429 | struct vm_list_struct **ppv; | 463 | struct vm_region *region, *last; |
430 | 464 | struct rb_node *p, *lastp; | |
431 | for (ppv = ¤t->mm->context.vmlist; *ppv; ppv = &(*ppv)->next) | 465 | |
432 | if ((*ppv)->vma->vm_start > vml->vma->vm_start) | 466 | lastp = rb_first(&nommu_region_tree); |
433 | break; | 467 | if (!lastp) |
434 | 468 | return; | |
435 | vml->next = *ppv; | 469 | |
436 | *ppv = vml; | 470 | last = rb_entry(lastp, struct vm_region, vm_rb); |
471 | if (unlikely(last->vm_end <= last->vm_start)) | ||
472 | BUG(); | ||
473 | if (unlikely(last->vm_top < last->vm_end)) | ||
474 | BUG(); | ||
475 | |||
476 | while ((p = rb_next(lastp))) { | ||
477 | region = rb_entry(p, struct vm_region, vm_rb); | ||
478 | last = rb_entry(lastp, struct vm_region, vm_rb); | ||
479 | |||
480 | if (unlikely(region->vm_end <= region->vm_start)) | ||
481 | BUG(); | ||
482 | if (unlikely(region->vm_top < region->vm_end)) | ||
483 | BUG(); | ||
484 | if (unlikely(region->vm_start < last->vm_top)) | ||
485 | BUG(); | ||
486 | |||
487 | lastp = p; | ||
488 | } | ||
437 | } | 489 | } |
490 | #else | ||
491 | #define validate_nommu_regions() do {} while(0) | ||
492 | #endif | ||
438 | 493 | ||
439 | /* | 494 | /* |
440 | * look up the first VMA in which addr resides, NULL if none | 495 | * add a region into the global tree |
441 | * - should be called with mm->mmap_sem at least held readlocked | ||
442 | */ | 496 | */ |
443 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | 497 | static void add_nommu_region(struct vm_region *region) |
444 | { | 498 | { |
445 | struct vm_list_struct *loop, *vml; | 499 | struct vm_region *pregion; |
500 | struct rb_node **p, *parent; | ||
446 | 501 | ||
447 | /* search the vm_start ordered list */ | 502 | validate_nommu_regions(); |
448 | vml = NULL; | 503 | |
449 | for (loop = mm->context.vmlist; loop; loop = loop->next) { | 504 | BUG_ON(region->vm_start & ~PAGE_MASK); |
450 | if (loop->vma->vm_start > addr) | 505 | |
451 | break; | 506 | parent = NULL; |
452 | vml = loop; | 507 | p = &nommu_region_tree.rb_node; |
508 | while (*p) { | ||
509 | parent = *p; | ||
510 | pregion = rb_entry(parent, struct vm_region, vm_rb); | ||
511 | if (region->vm_start < pregion->vm_start) | ||
512 | p = &(*p)->rb_left; | ||
513 | else if (region->vm_start > pregion->vm_start) | ||
514 | p = &(*p)->rb_right; | ||
515 | else if (pregion == region) | ||
516 | return; | ||
517 | else | ||
518 | BUG(); | ||
453 | } | 519 | } |
454 | 520 | ||
455 | if (vml && vml->vma->vm_end > addr) | 521 | rb_link_node(®ion->vm_rb, parent, p); |
456 | return vml->vma; | 522 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); |
457 | 523 | ||
458 | return NULL; | 524 | validate_nommu_regions(); |
459 | } | 525 | } |
460 | EXPORT_SYMBOL(find_vma); | ||
461 | 526 | ||
462 | /* | 527 | /* |
463 | * find a VMA | 528 | * delete a region from the global tree |
464 | * - we don't extend stack VMAs under NOMMU conditions | ||
465 | */ | 529 | */ |
466 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | 530 | static void delete_nommu_region(struct vm_region *region) |
467 | { | 531 | { |
468 | return find_vma(mm, addr); | 532 | BUG_ON(!nommu_region_tree.rb_node); |
469 | } | ||
470 | 533 | ||
471 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | 534 | validate_nommu_regions(); |
472 | { | 535 | rb_erase(®ion->vm_rb, &nommu_region_tree); |
473 | return -ENOMEM; | 536 | validate_nommu_regions(); |
474 | } | 537 | } |
475 | 538 | ||
476 | /* | 539 | /* |
477 | * look up the first VMA exactly that exactly matches addr | 540 | * free a contiguous series of pages |
478 | * - should be called with mm->mmap_sem at least held readlocked | ||
479 | */ | 541 | */ |
480 | static inline struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | 542 | static void free_page_series(unsigned long from, unsigned long to) |
481 | unsigned long addr) | ||
482 | { | 543 | { |
483 | struct vm_list_struct *vml; | 544 | for (; from < to; from += PAGE_SIZE) { |
484 | 545 | struct page *page = virt_to_page(from); | |
485 | /* search the vm_start ordered list */ | 546 | |
486 | for (vml = mm->context.vmlist; vml; vml = vml->next) { | 547 | kdebug("- free %lx", from); |
487 | if (vml->vma->vm_start == addr) | 548 | atomic_dec(&mmap_pages_allocated); |
488 | return vml->vma; | 549 | if (page_count(page) != 1) |
489 | if (vml->vma->vm_start > addr) | 550 | kdebug("free page %p [%d]", page, page_count(page)); |
490 | break; | 551 | put_page(page); |
491 | } | 552 | } |
492 | |||
493 | return NULL; | ||
494 | } | 553 | } |
495 | 554 | ||
496 | /* | 555 | /* |
497 | * find a VMA in the global tree | 556 | * release a reference to a region |
557 | * - the caller must hold the region semaphore, which this releases | ||
558 | * - the region may not have been added to the tree yet, in which case vm_top | ||
559 | * will equal vm_start | ||
498 | */ | 560 | */ |
499 | static inline struct vm_area_struct *find_nommu_vma(unsigned long start) | 561 | static void __put_nommu_region(struct vm_region *region) |
562 | __releases(nommu_region_sem) | ||
500 | { | 563 | { |
501 | struct vm_area_struct *vma; | 564 | kenter("%p{%d}", region, atomic_read(®ion->vm_usage)); |
502 | struct rb_node *n = nommu_vma_tree.rb_node; | ||
503 | 565 | ||
504 | while (n) { | 566 | BUG_ON(!nommu_region_tree.rb_node); |
505 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | ||
506 | 567 | ||
507 | if (start < vma->vm_start) | 568 | if (atomic_dec_and_test(®ion->vm_usage)) { |
508 | n = n->rb_left; | 569 | if (region->vm_top > region->vm_start) |
509 | else if (start > vma->vm_start) | 570 | delete_nommu_region(region); |
510 | n = n->rb_right; | 571 | up_write(&nommu_region_sem); |
511 | else | 572 | |
512 | return vma; | 573 | if (region->vm_file) |
574 | fput(region->vm_file); | ||
575 | |||
576 | /* IO memory and memory shared directly out of the pagecache | ||
577 | * from ramfs/tmpfs mustn't be released here */ | ||
578 | if (region->vm_flags & VM_MAPPED_COPY) { | ||
579 | kdebug("free series"); | ||
580 | free_page_series(region->vm_start, region->vm_top); | ||
581 | } | ||
582 | kmem_cache_free(vm_region_jar, region); | ||
583 | } else { | ||
584 | up_write(&nommu_region_sem); | ||
513 | } | 585 | } |
586 | } | ||
514 | 587 | ||
515 | return NULL; | 588 | /* |
589 | * release a reference to a region | ||
590 | */ | ||
591 | static void put_nommu_region(struct vm_region *region) | ||
592 | { | ||
593 | down_write(&nommu_region_sem); | ||
594 | __put_nommu_region(region); | ||
516 | } | 595 | } |
517 | 596 | ||
518 | /* | 597 | /* |
519 | * add a VMA in the global tree | 598 | * add a VMA into a process's mm_struct in the appropriate place in the list |
599 | * and tree and add to the address space's page tree also if not an anonymous | ||
600 | * page | ||
601 | * - should be called with mm->mmap_sem held writelocked | ||
520 | */ | 602 | */ |
521 | static void add_nommu_vma(struct vm_area_struct *vma) | 603 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
522 | { | 604 | { |
523 | struct vm_area_struct *pvma; | 605 | struct vm_area_struct *pvma, **pp; |
524 | struct address_space *mapping; | 606 | struct address_space *mapping; |
525 | struct rb_node **p = &nommu_vma_tree.rb_node; | 607 | struct rb_node **p, *parent; |
526 | struct rb_node *parent = NULL; | 608 | |
609 | kenter(",%p", vma); | ||
610 | |||
611 | BUG_ON(!vma->vm_region); | ||
612 | |||
613 | mm->map_count++; | ||
614 | vma->vm_mm = mm; | ||
527 | 615 | ||
528 | /* add the VMA to the mapping */ | 616 | /* add the VMA to the mapping */ |
529 | if (vma->vm_file) { | 617 | if (vma->vm_file) { |
@@ -534,42 +622,62 @@ static void add_nommu_vma(struct vm_area_struct *vma) | |||
534 | flush_dcache_mmap_unlock(mapping); | 622 | flush_dcache_mmap_unlock(mapping); |
535 | } | 623 | } |
536 | 624 | ||
537 | /* add the VMA to the master list */ | 625 | /* add the VMA to the tree */ |
626 | parent = NULL; | ||
627 | p = &mm->mm_rb.rb_node; | ||
538 | while (*p) { | 628 | while (*p) { |
539 | parent = *p; | 629 | parent = *p; |
540 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | 630 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); |
541 | 631 | ||
542 | if (vma->vm_start < pvma->vm_start) { | 632 | /* sort by: start addr, end addr, VMA struct addr in that order |
633 | * (the latter is necessary as we may get identical VMAs) */ | ||
634 | if (vma->vm_start < pvma->vm_start) | ||
543 | p = &(*p)->rb_left; | 635 | p = &(*p)->rb_left; |
544 | } | 636 | else if (vma->vm_start > pvma->vm_start) |
545 | else if (vma->vm_start > pvma->vm_start) { | ||
546 | p = &(*p)->rb_right; | 637 | p = &(*p)->rb_right; |
547 | } | 638 | else if (vma->vm_end < pvma->vm_end) |
548 | else { | 639 | p = &(*p)->rb_left; |
549 | /* mappings are at the same address - this can only | 640 | else if (vma->vm_end > pvma->vm_end) |
550 | * happen for shared-mem chardevs and shared file | 641 | p = &(*p)->rb_right; |
551 | * mappings backed by ramfs/tmpfs */ | 642 | else if (vma < pvma) |
552 | BUG_ON(!(pvma->vm_flags & VM_SHARED)); | 643 | p = &(*p)->rb_left; |
553 | 644 | else if (vma > pvma) | |
554 | if (vma < pvma) | 645 | p = &(*p)->rb_right; |
555 | p = &(*p)->rb_left; | 646 | else |
556 | else if (vma > pvma) | 647 | BUG(); |
557 | p = &(*p)->rb_right; | ||
558 | else | ||
559 | BUG(); | ||
560 | } | ||
561 | } | 648 | } |
562 | 649 | ||
563 | rb_link_node(&vma->vm_rb, parent, p); | 650 | rb_link_node(&vma->vm_rb, parent, p); |
564 | rb_insert_color(&vma->vm_rb, &nommu_vma_tree); | 651 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
652 | |||
653 | /* add VMA to the VMA list also */ | ||
654 | for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) { | ||
655 | if (pvma->vm_start > vma->vm_start) | ||
656 | break; | ||
657 | if (pvma->vm_start < vma->vm_start) | ||
658 | continue; | ||
659 | if (pvma->vm_end < vma->vm_end) | ||
660 | break; | ||
661 | } | ||
662 | |||
663 | vma->vm_next = *pp; | ||
664 | *pp = vma; | ||
565 | } | 665 | } |
566 | 666 | ||
567 | /* | 667 | /* |
568 | * delete a VMA from the global list | 668 | * delete a VMA from its owning mm_struct and address space |
569 | */ | 669 | */ |
570 | static void delete_nommu_vma(struct vm_area_struct *vma) | 670 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
571 | { | 671 | { |
672 | struct vm_area_struct **pp; | ||
572 | struct address_space *mapping; | 673 | struct address_space *mapping; |
674 | struct mm_struct *mm = vma->vm_mm; | ||
675 | |||
676 | kenter("%p", vma); | ||
677 | |||
678 | mm->map_count--; | ||
679 | if (mm->mmap_cache == vma) | ||
680 | mm->mmap_cache = NULL; | ||
573 | 681 | ||
574 | /* remove the VMA from the mapping */ | 682 | /* remove the VMA from the mapping */ |
575 | if (vma->vm_file) { | 683 | if (vma->vm_file) { |
@@ -580,8 +688,115 @@ static void delete_nommu_vma(struct vm_area_struct *vma) | |||
580 | flush_dcache_mmap_unlock(mapping); | 688 | flush_dcache_mmap_unlock(mapping); |
581 | } | 689 | } |
582 | 690 | ||
583 | /* remove from the master list */ | 691 | /* remove from the MM's tree and list */ |
584 | rb_erase(&vma->vm_rb, &nommu_vma_tree); | 692 | rb_erase(&vma->vm_rb, &mm->mm_rb); |
693 | for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) { | ||
694 | if (*pp == vma) { | ||
695 | *pp = vma->vm_next; | ||
696 | break; | ||
697 | } | ||
698 | } | ||
699 | |||
700 | vma->vm_mm = NULL; | ||
701 | } | ||
702 | |||
703 | /* | ||
704 | * destroy a VMA record | ||
705 | */ | ||
706 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | ||
707 | { | ||
708 | kenter("%p", vma); | ||
709 | if (vma->vm_ops && vma->vm_ops->close) | ||
710 | vma->vm_ops->close(vma); | ||
711 | if (vma->vm_file) { | ||
712 | fput(vma->vm_file); | ||
713 | if (vma->vm_flags & VM_EXECUTABLE) | ||
714 | removed_exe_file_vma(mm); | ||
715 | } | ||
716 | put_nommu_region(vma->vm_region); | ||
717 | kmem_cache_free(vm_area_cachep, vma); | ||
718 | } | ||
719 | |||
720 | /* | ||
721 | * look up the first VMA in which addr resides, NULL if none | ||
722 | * - should be called with mm->mmap_sem at least held readlocked | ||
723 | */ | ||
724 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | ||
725 | { | ||
726 | struct vm_area_struct *vma; | ||
727 | struct rb_node *n = mm->mm_rb.rb_node; | ||
728 | |||
729 | /* check the cache first */ | ||
730 | vma = mm->mmap_cache; | ||
731 | if (vma && vma->vm_start <= addr && vma->vm_end > addr) | ||
732 | return vma; | ||
733 | |||
734 | /* trawl the tree (there may be multiple mappings in which addr | ||
735 | * resides) */ | ||
736 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | ||
737 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | ||
738 | if (vma->vm_start > addr) | ||
739 | return NULL; | ||
740 | if (vma->vm_end > addr) { | ||
741 | mm->mmap_cache = vma; | ||
742 | return vma; | ||
743 | } | ||
744 | } | ||
745 | |||
746 | return NULL; | ||
747 | } | ||
748 | EXPORT_SYMBOL(find_vma); | ||
749 | |||
750 | /* | ||
751 | * find a VMA | ||
752 | * - we don't extend stack VMAs under NOMMU conditions | ||
753 | */ | ||
754 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | ||
755 | { | ||
756 | return find_vma(mm, addr); | ||
757 | } | ||
758 | |||
759 | /* | ||
760 | * expand a stack to a given address | ||
761 | * - not supported under NOMMU conditions | ||
762 | */ | ||
763 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | ||
764 | { | ||
765 | return -ENOMEM; | ||
766 | } | ||
767 | |||
768 | /* | ||
769 | * look up the first VMA exactly that exactly matches addr | ||
770 | * - should be called with mm->mmap_sem at least held readlocked | ||
771 | */ | ||
772 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | ||
773 | unsigned long addr, | ||
774 | unsigned long len) | ||
775 | { | ||
776 | struct vm_area_struct *vma; | ||
777 | struct rb_node *n = mm->mm_rb.rb_node; | ||
778 | unsigned long end = addr + len; | ||
779 | |||
780 | /* check the cache first */ | ||
781 | vma = mm->mmap_cache; | ||
782 | if (vma && vma->vm_start == addr && vma->vm_end == end) | ||
783 | return vma; | ||
784 | |||
785 | /* trawl the tree (there may be multiple mappings in which addr | ||
786 | * resides) */ | ||
787 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | ||
788 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | ||
789 | if (vma->vm_start < addr) | ||
790 | continue; | ||
791 | if (vma->vm_start > addr) | ||
792 | return NULL; | ||
793 | if (vma->vm_end == end) { | ||
794 | mm->mmap_cache = vma; | ||
795 | return vma; | ||
796 | } | ||
797 | } | ||
798 | |||
799 | return NULL; | ||
585 | } | 800 | } |
586 | 801 | ||
587 | /* | 802 | /* |
@@ -596,7 +811,7 @@ static int validate_mmap_request(struct file *file, | |||
596 | unsigned long pgoff, | 811 | unsigned long pgoff, |
597 | unsigned long *_capabilities) | 812 | unsigned long *_capabilities) |
598 | { | 813 | { |
599 | unsigned long capabilities; | 814 | unsigned long capabilities, rlen; |
600 | unsigned long reqprot = prot; | 815 | unsigned long reqprot = prot; |
601 | int ret; | 816 | int ret; |
602 | 817 | ||
@@ -616,12 +831,12 @@ static int validate_mmap_request(struct file *file, | |||
616 | return -EINVAL; | 831 | return -EINVAL; |
617 | 832 | ||
618 | /* Careful about overflows.. */ | 833 | /* Careful about overflows.. */ |
619 | len = PAGE_ALIGN(len); | 834 | rlen = PAGE_ALIGN(len); |
620 | if (!len || len > TASK_SIZE) | 835 | if (!rlen || rlen > TASK_SIZE) |
621 | return -ENOMEM; | 836 | return -ENOMEM; |
622 | 837 | ||
623 | /* offset overflow? */ | 838 | /* offset overflow? */ |
624 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) | 839 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
625 | return -EOVERFLOW; | 840 | return -EOVERFLOW; |
626 | 841 | ||
627 | if (file) { | 842 | if (file) { |
@@ -795,13 +1010,18 @@ static unsigned long determine_vm_flags(struct file *file, | |||
795 | } | 1010 | } |
796 | 1011 | ||
797 | /* | 1012 | /* |
798 | * set up a shared mapping on a file | 1013 | * set up a shared mapping on a file (the driver or filesystem provides and |
1014 | * pins the storage) | ||
799 | */ | 1015 | */ |
800 | static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len) | 1016 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
801 | { | 1017 | { |
802 | int ret; | 1018 | int ret; |
803 | 1019 | ||
804 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | 1020 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
1021 | if (ret == 0) { | ||
1022 | vma->vm_region->vm_top = vma->vm_region->vm_end; | ||
1023 | return ret; | ||
1024 | } | ||
805 | if (ret != -ENOSYS) | 1025 | if (ret != -ENOSYS) |
806 | return ret; | 1026 | return ret; |
807 | 1027 | ||
@@ -815,10 +1035,14 @@ static int do_mmap_shared_file(struct vm_area_struct *vma, unsigned long len) | |||
815 | /* | 1035 | /* |
816 | * set up a private mapping or an anonymous shared mapping | 1036 | * set up a private mapping or an anonymous shared mapping |
817 | */ | 1037 | */ |
818 | static int do_mmap_private(struct vm_area_struct *vma, unsigned long len) | 1038 | static int do_mmap_private(struct vm_area_struct *vma, |
1039 | struct vm_region *region, | ||
1040 | unsigned long len) | ||
819 | { | 1041 | { |
1042 | struct page *pages; | ||
1043 | unsigned long total, point, n, rlen; | ||
820 | void *base; | 1044 | void *base; |
821 | int ret; | 1045 | int ret, order; |
822 | 1046 | ||
823 | /* invoke the file's mapping function so that it can keep track of | 1047 | /* invoke the file's mapping function so that it can keep track of |
824 | * shared mappings on devices or memory | 1048 | * shared mappings on devices or memory |
@@ -826,34 +1050,63 @@ static int do_mmap_private(struct vm_area_struct *vma, unsigned long len) | |||
826 | */ | 1050 | */ |
827 | if (vma->vm_file) { | 1051 | if (vma->vm_file) { |
828 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | 1052 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
829 | if (ret != -ENOSYS) { | 1053 | if (ret == 0) { |
830 | /* shouldn't return success if we're not sharing */ | 1054 | /* shouldn't return success if we're not sharing */ |
831 | BUG_ON(ret == 0 && !(vma->vm_flags & VM_MAYSHARE)); | 1055 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
832 | return ret; /* success or a real error */ | 1056 | vma->vm_region->vm_top = vma->vm_region->vm_end; |
1057 | return ret; | ||
833 | } | 1058 | } |
1059 | if (ret != -ENOSYS) | ||
1060 | return ret; | ||
834 | 1061 | ||
835 | /* getting an ENOSYS error indicates that direct mmap isn't | 1062 | /* getting an ENOSYS error indicates that direct mmap isn't |
836 | * possible (as opposed to tried but failed) so we'll try to | 1063 | * possible (as opposed to tried but failed) so we'll try to |
837 | * make a private copy of the data and map that instead */ | 1064 | * make a private copy of the data and map that instead */ |
838 | } | 1065 | } |
839 | 1066 | ||
1067 | rlen = PAGE_ALIGN(len); | ||
1068 | |||
840 | /* allocate some memory to hold the mapping | 1069 | /* allocate some memory to hold the mapping |
841 | * - note that this may not return a page-aligned address if the object | 1070 | * - note that this may not return a page-aligned address if the object |
842 | * we're allocating is smaller than a page | 1071 | * we're allocating is smaller than a page |
843 | */ | 1072 | */ |
844 | base = kmalloc(len, GFP_KERNEL|__GFP_COMP); | 1073 | order = get_order(rlen); |
845 | if (!base) | 1074 | kdebug("alloc order %d for %lx", order, len); |
1075 | |||
1076 | pages = alloc_pages(GFP_KERNEL, order); | ||
1077 | if (!pages) | ||
846 | goto enomem; | 1078 | goto enomem; |
847 | 1079 | ||
848 | vma->vm_start = (unsigned long) base; | 1080 | total = 1 << order; |
849 | vma->vm_end = vma->vm_start + len; | 1081 | atomic_add(total, &mmap_pages_allocated); |
850 | vma->vm_flags |= VM_MAPPED_COPY; | 1082 | |
1083 | point = rlen >> PAGE_SHIFT; | ||
1084 | |||
1085 | /* we allocated a power-of-2 sized page set, so we may want to trim off | ||
1086 | * the excess */ | ||
1087 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { | ||
1088 | while (total > point) { | ||
1089 | order = ilog2(total - point); | ||
1090 | n = 1 << order; | ||
1091 | kdebug("shave %lu/%lu @%lu", n, total - point, total); | ||
1092 | atomic_sub(n, &mmap_pages_allocated); | ||
1093 | total -= n; | ||
1094 | set_page_refcounted(pages + total); | ||
1095 | __free_pages(pages + total, order); | ||
1096 | } | ||
1097 | } | ||
1098 | |||
1099 | for (point = 1; point < total; point++) | ||
1100 | set_page_refcounted(&pages[point]); | ||
851 | 1101 | ||
852 | #ifdef WARN_ON_SLACK | 1102 | base = page_address(pages); |
853 | if (len + WARN_ON_SLACK <= kobjsize(result)) | 1103 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
854 | printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n", | 1104 | region->vm_start = (unsigned long) base; |
855 | len, current->pid, kobjsize(result) - len); | 1105 | region->vm_end = region->vm_start + rlen; |
856 | #endif | 1106 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
1107 | |||
1108 | vma->vm_start = region->vm_start; | ||
1109 | vma->vm_end = region->vm_start + len; | ||
857 | 1110 | ||
858 | if (vma->vm_file) { | 1111 | if (vma->vm_file) { |
859 | /* read the contents of a file into the copy */ | 1112 | /* read the contents of a file into the copy */ |
@@ -865,26 +1118,28 @@ static int do_mmap_private(struct vm_area_struct *vma, unsigned long len) | |||
865 | 1118 | ||
866 | old_fs = get_fs(); | 1119 | old_fs = get_fs(); |
867 | set_fs(KERNEL_DS); | 1120 | set_fs(KERNEL_DS); |
868 | ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); | 1121 | ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); |
869 | set_fs(old_fs); | 1122 | set_fs(old_fs); |
870 | 1123 | ||
871 | if (ret < 0) | 1124 | if (ret < 0) |
872 | goto error_free; | 1125 | goto error_free; |
873 | 1126 | ||
874 | /* clear the last little bit */ | 1127 | /* clear the last little bit */ |
875 | if (ret < len) | 1128 | if (ret < rlen) |
876 | memset(base + ret, 0, len - ret); | 1129 | memset(base + ret, 0, rlen - ret); |
877 | 1130 | ||
878 | } else { | 1131 | } else { |
879 | /* if it's an anonymous mapping, then just clear it */ | 1132 | /* if it's an anonymous mapping, then just clear it */ |
880 | memset(base, 0, len); | 1133 | memset(base, 0, rlen); |
881 | } | 1134 | } |
882 | 1135 | ||
883 | return 0; | 1136 | return 0; |
884 | 1137 | ||
885 | error_free: | 1138 | error_free: |
886 | kfree(base); | 1139 | free_page_series(region->vm_start, region->vm_end); |
887 | vma->vm_start = 0; | 1140 | region->vm_start = vma->vm_start = 0; |
1141 | region->vm_end = vma->vm_end = 0; | ||
1142 | region->vm_top = 0; | ||
888 | return ret; | 1143 | return ret; |
889 | 1144 | ||
890 | enomem: | 1145 | enomem: |
@@ -904,13 +1159,14 @@ unsigned long do_mmap_pgoff(struct file *file, | |||
904 | unsigned long flags, | 1159 | unsigned long flags, |
905 | unsigned long pgoff) | 1160 | unsigned long pgoff) |
906 | { | 1161 | { |
907 | struct vm_list_struct *vml = NULL; | 1162 | struct vm_area_struct *vma; |
908 | struct vm_area_struct *vma = NULL; | 1163 | struct vm_region *region; |
909 | struct rb_node *rb; | 1164 | struct rb_node *rb; |
910 | unsigned long capabilities, vm_flags; | 1165 | unsigned long capabilities, vm_flags, result; |
911 | void *result; | ||
912 | int ret; | 1166 | int ret; |
913 | 1167 | ||
1168 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); | ||
1169 | |||
914 | if (!(flags & MAP_FIXED)) | 1170 | if (!(flags & MAP_FIXED)) |
915 | addr = round_hint_to_min(addr); | 1171 | addr = round_hint_to_min(addr); |
916 | 1172 | ||
@@ -918,73 +1174,120 @@ unsigned long do_mmap_pgoff(struct file *file, | |||
918 | * mapping */ | 1174 | * mapping */ |
919 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | 1175 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, |
920 | &capabilities); | 1176 | &capabilities); |
921 | if (ret < 0) | 1177 | if (ret < 0) { |
1178 | kleave(" = %d [val]", ret); | ||
922 | return ret; | 1179 | return ret; |
1180 | } | ||
923 | 1181 | ||
924 | /* we've determined that we can make the mapping, now translate what we | 1182 | /* we've determined that we can make the mapping, now translate what we |
925 | * now know into VMA flags */ | 1183 | * now know into VMA flags */ |
926 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); | 1184 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); |
927 | 1185 | ||
928 | /* we're going to need to record the mapping if it works */ | 1186 | /* we're going to need to record the mapping */ |
929 | vml = kzalloc(sizeof(struct vm_list_struct), GFP_KERNEL); | 1187 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); |
930 | if (!vml) | 1188 | if (!region) |
931 | goto error_getting_vml; | 1189 | goto error_getting_region; |
1190 | |||
1191 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | ||
1192 | if (!vma) | ||
1193 | goto error_getting_vma; | ||
1194 | |||
1195 | atomic_set(®ion->vm_usage, 1); | ||
1196 | region->vm_flags = vm_flags; | ||
1197 | region->vm_pgoff = pgoff; | ||
1198 | |||
1199 | INIT_LIST_HEAD(&vma->anon_vma_node); | ||
1200 | vma->vm_flags = vm_flags; | ||
1201 | vma->vm_pgoff = pgoff; | ||
932 | 1202 | ||
933 | down_write(&nommu_vma_sem); | 1203 | if (file) { |
1204 | region->vm_file = file; | ||
1205 | get_file(file); | ||
1206 | vma->vm_file = file; | ||
1207 | get_file(file); | ||
1208 | if (vm_flags & VM_EXECUTABLE) { | ||
1209 | added_exe_file_vma(current->mm); | ||
1210 | vma->vm_mm = current->mm; | ||
1211 | } | ||
1212 | } | ||
934 | 1213 | ||
935 | /* if we want to share, we need to check for VMAs created by other | 1214 | down_write(&nommu_region_sem); |
1215 | |||
1216 | /* if we want to share, we need to check for regions created by other | ||
936 | * mmap() calls that overlap with our proposed mapping | 1217 | * mmap() calls that overlap with our proposed mapping |
937 | * - we can only share with an exact match on most regular files | 1218 | * - we can only share with a superset match on most regular files |
938 | * - shared mappings on character devices and memory backed files are | 1219 | * - shared mappings on character devices and memory backed files are |
939 | * permitted to overlap inexactly as far as we are concerned for in | 1220 | * permitted to overlap inexactly as far as we are concerned for in |
940 | * these cases, sharing is handled in the driver or filesystem rather | 1221 | * these cases, sharing is handled in the driver or filesystem rather |
941 | * than here | 1222 | * than here |
942 | */ | 1223 | */ |
943 | if (vm_flags & VM_MAYSHARE) { | 1224 | if (vm_flags & VM_MAYSHARE) { |
944 | unsigned long pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; | 1225 | struct vm_region *pregion; |
945 | unsigned long vmpglen; | 1226 | unsigned long pglen, rpglen, pgend, rpgend, start; |
946 | 1227 | ||
947 | /* suppress VMA sharing for shared regions */ | 1228 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
948 | if (vm_flags & VM_SHARED && | 1229 | pgend = pgoff + pglen; |
949 | capabilities & BDI_CAP_MAP_DIRECT) | ||
950 | goto dont_share_VMAs; | ||
951 | 1230 | ||
952 | for (rb = rb_first(&nommu_vma_tree); rb; rb = rb_next(rb)) { | 1231 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
953 | vma = rb_entry(rb, struct vm_area_struct, vm_rb); | 1232 | pregion = rb_entry(rb, struct vm_region, vm_rb); |
954 | 1233 | ||
955 | if (!(vma->vm_flags & VM_MAYSHARE)) | 1234 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
956 | continue; | 1235 | continue; |
957 | 1236 | ||
958 | /* search for overlapping mappings on the same file */ | 1237 | /* search for overlapping mappings on the same file */ |
959 | if (vma->vm_file->f_path.dentry->d_inode != file->f_path.dentry->d_inode) | 1238 | if (pregion->vm_file->f_path.dentry->d_inode != |
1239 | file->f_path.dentry->d_inode) | ||
960 | continue; | 1240 | continue; |
961 | 1241 | ||
962 | if (vma->vm_pgoff >= pgoff + pglen) | 1242 | if (pregion->vm_pgoff >= pgend) |
963 | continue; | 1243 | continue; |
964 | 1244 | ||
965 | vmpglen = vma->vm_end - vma->vm_start + PAGE_SIZE - 1; | 1245 | rpglen = pregion->vm_end - pregion->vm_start; |
966 | vmpglen >>= PAGE_SHIFT; | 1246 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
967 | if (pgoff >= vma->vm_pgoff + vmpglen) | 1247 | rpgend = pregion->vm_pgoff + rpglen; |
1248 | if (pgoff >= rpgend) | ||
968 | continue; | 1249 | continue; |
969 | 1250 | ||
970 | /* handle inexactly overlapping matches between mappings */ | 1251 | /* handle inexactly overlapping matches between |
971 | if (vma->vm_pgoff != pgoff || vmpglen != pglen) { | 1252 | * mappings */ |
1253 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | ||
1254 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | ||
1255 | /* new mapping is not a subset of the region */ | ||
972 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) | 1256 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
973 | goto sharing_violation; | 1257 | goto sharing_violation; |
974 | continue; | 1258 | continue; |
975 | } | 1259 | } |
976 | 1260 | ||
977 | /* we've found a VMA we can share */ | 1261 | /* we've found a region we can share */ |
978 | atomic_inc(&vma->vm_usage); | 1262 | atomic_inc(&pregion->vm_usage); |
979 | 1263 | vma->vm_region = pregion; | |
980 | vml->vma = vma; | 1264 | start = pregion->vm_start; |
981 | result = (void *) vma->vm_start; | 1265 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; |
982 | goto shared; | 1266 | vma->vm_start = start; |
1267 | vma->vm_end = start + len; | ||
1268 | |||
1269 | if (pregion->vm_flags & VM_MAPPED_COPY) { | ||
1270 | kdebug("share copy"); | ||
1271 | vma->vm_flags |= VM_MAPPED_COPY; | ||
1272 | } else { | ||
1273 | kdebug("share mmap"); | ||
1274 | ret = do_mmap_shared_file(vma); | ||
1275 | if (ret < 0) { | ||
1276 | vma->vm_region = NULL; | ||
1277 | vma->vm_start = 0; | ||
1278 | vma->vm_end = 0; | ||
1279 | atomic_dec(&pregion->vm_usage); | ||
1280 | pregion = NULL; | ||
1281 | goto error_just_free; | ||
1282 | } | ||
1283 | } | ||
1284 | fput(region->vm_file); | ||
1285 | kmem_cache_free(vm_region_jar, region); | ||
1286 | region = pregion; | ||
1287 | result = start; | ||
1288 | goto share; | ||
983 | } | 1289 | } |
984 | 1290 | ||
985 | dont_share_VMAs: | ||
986 | vma = NULL; | ||
987 | |||
988 | /* obtain the address at which to make a shared mapping | 1291 | /* obtain the address at which to make a shared mapping |
989 | * - this is the hook for quasi-memory character devices to | 1292 | * - this is the hook for quasi-memory character devices to |
990 | * tell us the location of a shared mapping | 1293 | * tell us the location of a shared mapping |
@@ -995,113 +1298,93 @@ unsigned long do_mmap_pgoff(struct file *file, | |||
995 | if (IS_ERR((void *) addr)) { | 1298 | if (IS_ERR((void *) addr)) { |
996 | ret = addr; | 1299 | ret = addr; |
997 | if (ret != (unsigned long) -ENOSYS) | 1300 | if (ret != (unsigned long) -ENOSYS) |
998 | goto error; | 1301 | goto error_just_free; |
999 | 1302 | ||
1000 | /* the driver refused to tell us where to site | 1303 | /* the driver refused to tell us where to site |
1001 | * the mapping so we'll have to attempt to copy | 1304 | * the mapping so we'll have to attempt to copy |
1002 | * it */ | 1305 | * it */ |
1003 | ret = (unsigned long) -ENODEV; | 1306 | ret = (unsigned long) -ENODEV; |
1004 | if (!(capabilities & BDI_CAP_MAP_COPY)) | 1307 | if (!(capabilities & BDI_CAP_MAP_COPY)) |
1005 | goto error; | 1308 | goto error_just_free; |
1006 | 1309 | ||
1007 | capabilities &= ~BDI_CAP_MAP_DIRECT; | 1310 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
1311 | } else { | ||
1312 | vma->vm_start = region->vm_start = addr; | ||
1313 | vma->vm_end = region->vm_end = addr + len; | ||
1008 | } | 1314 | } |
1009 | } | 1315 | } |
1010 | } | 1316 | } |
1011 | 1317 | ||
1012 | /* we're going to need a VMA struct as well */ | 1318 | vma->vm_region = region; |
1013 | vma = kzalloc(sizeof(struct vm_area_struct), GFP_KERNEL); | ||
1014 | if (!vma) | ||
1015 | goto error_getting_vma; | ||
1016 | |||
1017 | INIT_LIST_HEAD(&vma->anon_vma_node); | ||
1018 | atomic_set(&vma->vm_usage, 1); | ||
1019 | if (file) { | ||
1020 | get_file(file); | ||
1021 | if (vm_flags & VM_EXECUTABLE) { | ||
1022 | added_exe_file_vma(current->mm); | ||
1023 | vma->vm_mm = current->mm; | ||
1024 | } | ||
1025 | } | ||
1026 | vma->vm_file = file; | ||
1027 | vma->vm_flags = vm_flags; | ||
1028 | vma->vm_start = addr; | ||
1029 | vma->vm_end = addr + len; | ||
1030 | vma->vm_pgoff = pgoff; | ||
1031 | |||
1032 | vml->vma = vma; | ||
1033 | 1319 | ||
1034 | /* set up the mapping */ | 1320 | /* set up the mapping */ |
1035 | if (file && vma->vm_flags & VM_SHARED) | 1321 | if (file && vma->vm_flags & VM_SHARED) |
1036 | ret = do_mmap_shared_file(vma, len); | 1322 | ret = do_mmap_shared_file(vma); |
1037 | else | 1323 | else |
1038 | ret = do_mmap_private(vma, len); | 1324 | ret = do_mmap_private(vma, region, len); |
1039 | if (ret < 0) | 1325 | if (ret < 0) |
1040 | goto error; | 1326 | goto error_put_region; |
1041 | |||
1042 | /* okay... we have a mapping; now we have to register it */ | ||
1043 | result = (void *) vma->vm_start; | ||
1044 | 1327 | ||
1045 | if (vma->vm_flags & VM_MAPPED_COPY) { | 1328 | add_nommu_region(region); |
1046 | realalloc += kobjsize(result); | ||
1047 | askedalloc += len; | ||
1048 | } | ||
1049 | 1329 | ||
1050 | realalloc += kobjsize(vma); | 1330 | /* okay... we have a mapping; now we have to register it */ |
1051 | askedalloc += sizeof(*vma); | 1331 | result = vma->vm_start; |
1052 | 1332 | ||
1053 | current->mm->total_vm += len >> PAGE_SHIFT; | 1333 | current->mm->total_vm += len >> PAGE_SHIFT; |
1054 | 1334 | ||
1055 | add_nommu_vma(vma); | 1335 | share: |
1056 | 1336 | add_vma_to_mm(current->mm, vma); | |
1057 | shared: | ||
1058 | realalloc += kobjsize(vml); | ||
1059 | askedalloc += sizeof(*vml); | ||
1060 | |||
1061 | add_vma_to_mm(current->mm, vml); | ||
1062 | 1337 | ||
1063 | up_write(&nommu_vma_sem); | 1338 | up_write(&nommu_region_sem); |
1064 | 1339 | ||
1065 | if (prot & PROT_EXEC) | 1340 | if (prot & PROT_EXEC) |
1066 | flush_icache_range((unsigned long) result, | 1341 | flush_icache_range(result, result + len); |
1067 | (unsigned long) result + len); | ||
1068 | 1342 | ||
1069 | #ifdef DEBUG | 1343 | kleave(" = %lx", result); |
1070 | printk("do_mmap:\n"); | 1344 | return result; |
1071 | show_process_blocks(); | ||
1072 | #endif | ||
1073 | |||
1074 | return (unsigned long) result; | ||
1075 | 1345 | ||
1076 | error: | 1346 | error_put_region: |
1077 | up_write(&nommu_vma_sem); | 1347 | __put_nommu_region(region); |
1078 | kfree(vml); | ||
1079 | if (vma) { | 1348 | if (vma) { |
1080 | if (vma->vm_file) { | 1349 | if (vma->vm_file) { |
1081 | fput(vma->vm_file); | 1350 | fput(vma->vm_file); |
1082 | if (vma->vm_flags & VM_EXECUTABLE) | 1351 | if (vma->vm_flags & VM_EXECUTABLE) |
1083 | removed_exe_file_vma(vma->vm_mm); | 1352 | removed_exe_file_vma(vma->vm_mm); |
1084 | } | 1353 | } |
1085 | kfree(vma); | 1354 | kmem_cache_free(vm_area_cachep, vma); |
1086 | } | 1355 | } |
1356 | kleave(" = %d [pr]", ret); | ||
1087 | return ret; | 1357 | return ret; |
1088 | 1358 | ||
1089 | sharing_violation: | 1359 | error_just_free: |
1090 | up_write(&nommu_vma_sem); | 1360 | up_write(&nommu_region_sem); |
1091 | printk("Attempt to share mismatched mappings\n"); | 1361 | error: |
1092 | kfree(vml); | 1362 | fput(region->vm_file); |
1093 | return -EINVAL; | 1363 | kmem_cache_free(vm_region_jar, region); |
1364 | fput(vma->vm_file); | ||
1365 | if (vma->vm_flags & VM_EXECUTABLE) | ||
1366 | removed_exe_file_vma(vma->vm_mm); | ||
1367 | kmem_cache_free(vm_area_cachep, vma); | ||
1368 | kleave(" = %d", ret); | ||
1369 | return ret; | ||
1094 | 1370 | ||
1095 | error_getting_vma: | 1371 | sharing_violation: |
1096 | up_write(&nommu_vma_sem); | 1372 | up_write(&nommu_region_sem); |
1097 | kfree(vml); | 1373 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); |
1098 | printk("Allocation of vma for %lu byte allocation from process %d failed\n", | 1374 | ret = -EINVAL; |
1375 | goto error; | ||
1376 | |||
1377 | error_getting_vma: | ||
1378 | kmem_cache_free(vm_region_jar, region); | ||
1379 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" | ||
1380 | " from process %d failed\n", | ||
1099 | len, current->pid); | 1381 | len, current->pid); |
1100 | show_free_areas(); | 1382 | show_free_areas(); |
1101 | return -ENOMEM; | 1383 | return -ENOMEM; |
1102 | 1384 | ||
1103 | error_getting_vml: | 1385 | error_getting_region: |
1104 | printk("Allocation of vml for %lu byte allocation from process %d failed\n", | 1386 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" |
1387 | " from process %d failed\n", | ||
1105 | len, current->pid); | 1388 | len, current->pid); |
1106 | show_free_areas(); | 1389 | show_free_areas(); |
1107 | return -ENOMEM; | 1390 | return -ENOMEM; |
@@ -1109,85 +1392,183 @@ unsigned long do_mmap_pgoff(struct file *file, | |||
1109 | EXPORT_SYMBOL(do_mmap_pgoff); | 1392 | EXPORT_SYMBOL(do_mmap_pgoff); |
1110 | 1393 | ||
1111 | /* | 1394 | /* |
1112 | * handle mapping disposal for uClinux | 1395 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1396 | * for the first part or the tail. | ||
1113 | */ | 1397 | */ |
1114 | static void put_vma(struct mm_struct *mm, struct vm_area_struct *vma) | 1398 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1399 | unsigned long addr, int new_below) | ||
1115 | { | 1400 | { |
1116 | if (vma) { | 1401 | struct vm_area_struct *new; |
1117 | down_write(&nommu_vma_sem); | 1402 | struct vm_region *region; |
1403 | unsigned long npages; | ||
1118 | 1404 | ||
1119 | if (atomic_dec_and_test(&vma->vm_usage)) { | 1405 | kenter(""); |
1120 | delete_nommu_vma(vma); | ||
1121 | 1406 | ||
1122 | if (vma->vm_ops && vma->vm_ops->close) | 1407 | /* we're only permitted to split anonymous regions that have a single |
1123 | vma->vm_ops->close(vma); | 1408 | * owner */ |
1409 | if (vma->vm_file || | ||
1410 | atomic_read(&vma->vm_region->vm_usage) != 1) | ||
1411 | return -ENOMEM; | ||
1124 | 1412 | ||
1125 | /* IO memory and memory shared directly out of the pagecache from | 1413 | if (mm->map_count >= sysctl_max_map_count) |
1126 | * ramfs/tmpfs mustn't be released here */ | 1414 | return -ENOMEM; |
1127 | if (vma->vm_flags & VM_MAPPED_COPY) { | ||
1128 | realalloc -= kobjsize((void *) vma->vm_start); | ||
1129 | askedalloc -= vma->vm_end - vma->vm_start; | ||
1130 | kfree((void *) vma->vm_start); | ||
1131 | } | ||
1132 | 1415 | ||
1133 | realalloc -= kobjsize(vma); | 1416 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1134 | askedalloc -= sizeof(*vma); | 1417 | if (!region) |
1418 | return -ENOMEM; | ||
1135 | 1419 | ||
1136 | if (vma->vm_file) { | 1420 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1137 | fput(vma->vm_file); | 1421 | if (!new) { |
1138 | if (vma->vm_flags & VM_EXECUTABLE) | 1422 | kmem_cache_free(vm_region_jar, region); |
1139 | removed_exe_file_vma(mm); | 1423 | return -ENOMEM; |
1140 | } | 1424 | } |
1141 | kfree(vma); | 1425 | |
1142 | } | 1426 | /* most fields are the same, copy all, and then fixup */ |
1427 | *new = *vma; | ||
1428 | *region = *vma->vm_region; | ||
1429 | new->vm_region = region; | ||
1430 | |||
1431 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | ||
1143 | 1432 | ||
1144 | up_write(&nommu_vma_sem); | 1433 | if (new_below) { |
1434 | region->vm_top = region->vm_end = new->vm_end = addr; | ||
1435 | } else { | ||
1436 | region->vm_start = new->vm_start = addr; | ||
1437 | region->vm_pgoff = new->vm_pgoff += npages; | ||
1438 | } | ||
1439 | |||
1440 | if (new->vm_ops && new->vm_ops->open) | ||
1441 | new->vm_ops->open(new); | ||
1442 | |||
1443 | delete_vma_from_mm(vma); | ||
1444 | down_write(&nommu_region_sem); | ||
1445 | delete_nommu_region(vma->vm_region); | ||
1446 | if (new_below) { | ||
1447 | vma->vm_region->vm_start = vma->vm_start = addr; | ||
1448 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | ||
1449 | } else { | ||
1450 | vma->vm_region->vm_end = vma->vm_end = addr; | ||
1451 | vma->vm_region->vm_top = addr; | ||
1145 | } | 1452 | } |
1453 | add_nommu_region(vma->vm_region); | ||
1454 | add_nommu_region(new->vm_region); | ||
1455 | up_write(&nommu_region_sem); | ||
1456 | add_vma_to_mm(mm, vma); | ||
1457 | add_vma_to_mm(mm, new); | ||
1458 | return 0; | ||
1146 | } | 1459 | } |
1147 | 1460 | ||
1148 | /* | 1461 | /* |
1149 | * release a mapping | 1462 | * shrink a VMA by removing the specified chunk from either the beginning or |
1150 | * - under NOMMU conditions the parameters must match exactly to the mapping to | 1463 | * the end |
1151 | * be removed | ||
1152 | */ | 1464 | */ |
1153 | int do_munmap(struct mm_struct *mm, unsigned long addr, size_t len) | 1465 | static int shrink_vma(struct mm_struct *mm, |
1466 | struct vm_area_struct *vma, | ||
1467 | unsigned long from, unsigned long to) | ||
1154 | { | 1468 | { |
1155 | struct vm_list_struct *vml, **parent; | 1469 | struct vm_region *region; |
1156 | unsigned long end = addr + len; | ||
1157 | 1470 | ||
1158 | #ifdef DEBUG | 1471 | kenter(""); |
1159 | printk("do_munmap:\n"); | ||
1160 | #endif | ||
1161 | 1472 | ||
1162 | for (parent = &mm->context.vmlist; *parent; parent = &(*parent)->next) { | 1473 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1163 | if ((*parent)->vma->vm_start > addr) | 1474 | * and list */ |
1164 | break; | 1475 | delete_vma_from_mm(vma); |
1165 | if ((*parent)->vma->vm_start == addr && | 1476 | if (from > vma->vm_start) |
1166 | ((len == 0) || ((*parent)->vma->vm_end == end))) | 1477 | vma->vm_end = from; |
1167 | goto found; | 1478 | else |
1479 | vma->vm_start = to; | ||
1480 | add_vma_to_mm(mm, vma); | ||
1481 | |||
1482 | /* cut the backing region down to size */ | ||
1483 | region = vma->vm_region; | ||
1484 | BUG_ON(atomic_read(®ion->vm_usage) != 1); | ||
1485 | |||
1486 | down_write(&nommu_region_sem); | ||
1487 | delete_nommu_region(region); | ||
1488 | if (from > region->vm_start) { | ||
1489 | to = region->vm_top; | ||
1490 | region->vm_top = region->vm_end = from; | ||
1491 | } else { | ||
1492 | region->vm_start = to; | ||
1168 | } | 1493 | } |
1494 | add_nommu_region(region); | ||
1495 | up_write(&nommu_region_sem); | ||
1169 | 1496 | ||
1170 | printk("munmap of non-mmaped memory by process %d (%s): %p\n", | 1497 | free_page_series(from, to); |
1171 | current->pid, current->comm, (void *) addr); | 1498 | return 0; |
1172 | return -EINVAL; | 1499 | } |
1173 | 1500 | ||
1174 | found: | 1501 | /* |
1175 | vml = *parent; | 1502 | * release a mapping |
1503 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | ||
1504 | * VMA, though it need not cover the whole VMA | ||
1505 | */ | ||
1506 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | ||
1507 | { | ||
1508 | struct vm_area_struct *vma; | ||
1509 | struct rb_node *rb; | ||
1510 | unsigned long end = start + len; | ||
1511 | int ret; | ||
1176 | 1512 | ||
1177 | put_vma(mm, vml->vma); | 1513 | kenter(",%lx,%zx", start, len); |
1178 | 1514 | ||
1179 | *parent = vml->next; | 1515 | if (len == 0) |
1180 | realalloc -= kobjsize(vml); | 1516 | return -EINVAL; |
1181 | askedalloc -= sizeof(*vml); | ||
1182 | kfree(vml); | ||
1183 | 1517 | ||
1184 | update_hiwater_vm(mm); | 1518 | /* find the first potentially overlapping VMA */ |
1185 | mm->total_vm -= len >> PAGE_SHIFT; | 1519 | vma = find_vma(mm, start); |
1520 | if (!vma) { | ||
1521 | printk(KERN_WARNING | ||
1522 | "munmap of memory not mmapped by process %d (%s):" | ||
1523 | " 0x%lx-0x%lx\n", | ||
1524 | current->pid, current->comm, start, start + len - 1); | ||
1525 | return -EINVAL; | ||
1526 | } | ||
1186 | 1527 | ||
1187 | #ifdef DEBUG | 1528 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1188 | show_process_blocks(); | 1529 | if (vma->vm_file) { |
1189 | #endif | 1530 | do { |
1531 | if (start > vma->vm_start) { | ||
1532 | kleave(" = -EINVAL [miss]"); | ||
1533 | return -EINVAL; | ||
1534 | } | ||
1535 | if (end == vma->vm_end) | ||
1536 | goto erase_whole_vma; | ||
1537 | rb = rb_next(&vma->vm_rb); | ||
1538 | vma = rb_entry(rb, struct vm_area_struct, vm_rb); | ||
1539 | } while (rb); | ||
1540 | kleave(" = -EINVAL [split file]"); | ||
1541 | return -EINVAL; | ||
1542 | } else { | ||
1543 | /* the chunk must be a subset of the VMA found */ | ||
1544 | if (start == vma->vm_start && end == vma->vm_end) | ||
1545 | goto erase_whole_vma; | ||
1546 | if (start < vma->vm_start || end > vma->vm_end) { | ||
1547 | kleave(" = -EINVAL [superset]"); | ||
1548 | return -EINVAL; | ||
1549 | } | ||
1550 | if (start & ~PAGE_MASK) { | ||
1551 | kleave(" = -EINVAL [unaligned start]"); | ||
1552 | return -EINVAL; | ||
1553 | } | ||
1554 | if (end != vma->vm_end && end & ~PAGE_MASK) { | ||
1555 | kleave(" = -EINVAL [unaligned split]"); | ||
1556 | return -EINVAL; | ||
1557 | } | ||
1558 | if (start != vma->vm_start && end != vma->vm_end) { | ||
1559 | ret = split_vma(mm, vma, start, 1); | ||
1560 | if (ret < 0) { | ||
1561 | kleave(" = %d [split]", ret); | ||
1562 | return ret; | ||
1563 | } | ||
1564 | } | ||
1565 | return shrink_vma(mm, vma, start, end); | ||
1566 | } | ||
1190 | 1567 | ||
1568 | erase_whole_vma: | ||
1569 | delete_vma_from_mm(vma); | ||
1570 | delete_vma(mm, vma); | ||
1571 | kleave(" = 0"); | ||
1191 | return 0; | 1572 | return 0; |
1192 | } | 1573 | } |
1193 | EXPORT_SYMBOL(do_munmap); | 1574 | EXPORT_SYMBOL(do_munmap); |
@@ -1204,32 +1585,26 @@ asmlinkage long sys_munmap(unsigned long addr, size_t len) | |||
1204 | } | 1585 | } |
1205 | 1586 | ||
1206 | /* | 1587 | /* |
1207 | * Release all mappings | 1588 | * release all the mappings made in a process's VM space |
1208 | */ | 1589 | */ |
1209 | void exit_mmap(struct mm_struct * mm) | 1590 | void exit_mmap(struct mm_struct *mm) |
1210 | { | 1591 | { |
1211 | struct vm_list_struct *tmp; | 1592 | struct vm_area_struct *vma; |
1212 | |||
1213 | if (mm) { | ||
1214 | #ifdef DEBUG | ||
1215 | printk("Exit_mmap:\n"); | ||
1216 | #endif | ||
1217 | 1593 | ||
1218 | mm->total_vm = 0; | 1594 | if (!mm) |
1595 | return; | ||
1219 | 1596 | ||
1220 | while ((tmp = mm->context.vmlist)) { | 1597 | kenter(""); |
1221 | mm->context.vmlist = tmp->next; | ||
1222 | put_vma(mm, tmp->vma); | ||
1223 | 1598 | ||
1224 | realalloc -= kobjsize(tmp); | 1599 | mm->total_vm = 0; |
1225 | askedalloc -= sizeof(*tmp); | ||
1226 | kfree(tmp); | ||
1227 | } | ||
1228 | 1600 | ||
1229 | #ifdef DEBUG | 1601 | while ((vma = mm->mmap)) { |
1230 | show_process_blocks(); | 1602 | mm->mmap = vma->vm_next; |
1231 | #endif | 1603 | delete_vma_from_mm(vma); |
1604 | delete_vma(mm, vma); | ||
1232 | } | 1605 | } |
1606 | |||
1607 | kleave(""); | ||
1233 | } | 1608 | } |
1234 | 1609 | ||
1235 | unsigned long do_brk(unsigned long addr, unsigned long len) | 1610 | unsigned long do_brk(unsigned long addr, unsigned long len) |
@@ -1242,8 +1617,8 @@ unsigned long do_brk(unsigned long addr, unsigned long len) | |||
1242 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | 1617 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) |
1243 | * | 1618 | * |
1244 | * under NOMMU conditions, we only permit changing a mapping's size, and only | 1619 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
1245 | * as long as it stays within the hole allocated by the kmalloc() call in | 1620 | * as long as it stays within the region allocated by do_mmap_private() and the |
1246 | * do_mmap_pgoff() and the block is not shareable | 1621 | * block is not shareable |
1247 | * | 1622 | * |
1248 | * MREMAP_FIXED is not supported under NOMMU conditions | 1623 | * MREMAP_FIXED is not supported under NOMMU conditions |
1249 | */ | 1624 | */ |
@@ -1254,13 +1629,16 @@ unsigned long do_mremap(unsigned long addr, | |||
1254 | struct vm_area_struct *vma; | 1629 | struct vm_area_struct *vma; |
1255 | 1630 | ||
1256 | /* insanity checks first */ | 1631 | /* insanity checks first */ |
1257 | if (new_len == 0) | 1632 | if (old_len == 0 || new_len == 0) |
1258 | return (unsigned long) -EINVAL; | 1633 | return (unsigned long) -EINVAL; |
1259 | 1634 | ||
1635 | if (addr & ~PAGE_MASK) | ||
1636 | return -EINVAL; | ||
1637 | |||
1260 | if (flags & MREMAP_FIXED && new_addr != addr) | 1638 | if (flags & MREMAP_FIXED && new_addr != addr) |
1261 | return (unsigned long) -EINVAL; | 1639 | return (unsigned long) -EINVAL; |
1262 | 1640 | ||
1263 | vma = find_vma_exact(current->mm, addr); | 1641 | vma = find_vma_exact(current->mm, addr, old_len); |
1264 | if (!vma) | 1642 | if (!vma) |
1265 | return (unsigned long) -EINVAL; | 1643 | return (unsigned long) -EINVAL; |
1266 | 1644 | ||
@@ -1270,22 +1648,19 @@ unsigned long do_mremap(unsigned long addr, | |||
1270 | if (vma->vm_flags & VM_MAYSHARE) | 1648 | if (vma->vm_flags & VM_MAYSHARE) |
1271 | return (unsigned long) -EPERM; | 1649 | return (unsigned long) -EPERM; |
1272 | 1650 | ||
1273 | if (new_len > kobjsize((void *) addr)) | 1651 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1274 | return (unsigned long) -ENOMEM; | 1652 | return (unsigned long) -ENOMEM; |
1275 | 1653 | ||
1276 | /* all checks complete - do it */ | 1654 | /* all checks complete - do it */ |
1277 | vma->vm_end = vma->vm_start + new_len; | 1655 | vma->vm_end = vma->vm_start + new_len; |
1278 | |||
1279 | askedalloc -= old_len; | ||
1280 | askedalloc += new_len; | ||
1281 | |||
1282 | return vma->vm_start; | 1656 | return vma->vm_start; |
1283 | } | 1657 | } |
1284 | EXPORT_SYMBOL(do_mremap); | 1658 | EXPORT_SYMBOL(do_mremap); |
1285 | 1659 | ||
1286 | asmlinkage unsigned long sys_mremap(unsigned long addr, | 1660 | asmlinkage |
1287 | unsigned long old_len, unsigned long new_len, | 1661 | unsigned long sys_mremap(unsigned long addr, |
1288 | unsigned long flags, unsigned long new_addr) | 1662 | unsigned long old_len, unsigned long new_len, |
1663 | unsigned long flags, unsigned long new_addr) | ||
1289 | { | 1664 | { |
1290 | unsigned long ret; | 1665 | unsigned long ret; |
1291 | 1666 | ||