1/* 2 * linux/mm/nommu.c 3 * 4 * Replacement code for mm functions to support CPU's that don't 5 * have any form of memory management unit (thus no virtual memory). 6 * 7 * See Documentation/nommu-mmap.txt 8 * 9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> 10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> 11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> 12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> 13 * Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org> 14 */ 15 16#include <linux/module.h> 17#include <linux/mm.h> 18#include <linux/mman.h> 19#include <linux/swap.h> 20#include <linux/file.h> 21#include <linux/highmem.h> 22#include <linux/pagemap.h> 23#include <linux/slab.h> 24#include <linux/vmalloc.h> 25#include <linux/tracehook.h> 26#include <linux/blkdev.h> 27#include <linux/backing-dev.h> 28#include <linux/mount.h> 29#include <linux/personality.h> 30#include <linux/security.h> 31#include <linux/syscalls.h> 32 33#include <asm/uaccess.h> 34#include <asm/tlb.h> 35#include <asm/tlbflush.h> 36#include <asm/mmu_context.h> 37#include "internal.h" 38 39#define kenter(FMT, ...) \ 40 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 41#define kleave(FMT, ...) \ 42 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 43#define kdebug(FMT, ...) \ 44 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 45 46void *high_memory; 47struct page *mem_map; 48unsigned long max_mapnr; 49unsigned long num_physpages; 50unsigned long highest_memmap_pfn; 51struct percpu_counter vm_committed_as; 52int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 53int sysctl_overcommit_ratio = 50; /* default is 50% */ 54int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 55int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 56int heap_stack_gap = 0; 57 58atomic_long_t mmap_pages_allocated; 59 60EXPORT_SYMBOL(mem_map); 61EXPORT_SYMBOL(num_physpages); 62 63/* list of mapped, potentially shareable regions */ 64static struct kmem_cache *vm_region_jar; 65struct rb_root nommu_region_tree = RB_ROOT; 66DECLARE_RWSEM(nommu_region_sem); 67 68const struct vm_operations_struct generic_file_vm_ops = { 69}; 70 71/* 72 * Return the total memory allocated for this pointer, not 73 * just what the caller asked for. 74 * 75 * Doesn't have to be accurate, i.e. may have races. 76 */ 77unsigned int kobjsize(const void *objp) 78{ 79 struct page *page; 80 81 /* 82 * If the object we have should not have ksize performed on it, 83 * return size of 0 84 */ 85 if (!objp || !virt_addr_valid(objp)) 86 return 0; 87 88 page = virt_to_head_page(objp); 89 90 /* 91 * If the allocator sets PageSlab, we know the pointer came from 92 * kmalloc(). 93 */ 94 if (PageSlab(page)) 95 return ksize(objp); 96 97 /* 98 * If it's not a compound page, see if we have a matching VMA 99 * region. This test is intentionally done in reverse order, 100 * so if there's no VMA, we still fall through and hand back 101 * PAGE_SIZE for 0-order pages. 102 */ 103 if (!PageCompound(page)) { 104 struct vm_area_struct *vma; 105 106 vma = find_vma(current->mm, (unsigned long)objp); 107 if (vma) 108 return vma->vm_end - vma->vm_start; 109 } 110 111 /* 112 * The ksize() function is only guaranteed to work for pointers 113 * returned by kmalloc(). So handle arbitrary pointers here. 114 */ 115 return PAGE_SIZE << compound_order(page); 116} 117 118int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 119 unsigned long start, int nr_pages, unsigned int foll_flags, 120 struct page **pages, struct vm_area_struct **vmas) 121{ 122 struct vm_area_struct *vma; 123 unsigned long vm_flags; 124 int i; 125 126 /* calculate required read or write permissions. 127 * If FOLL_FORCE is set, we only require the "MAY" flags. 128 */ 129 vm_flags = (foll_flags & FOLL_WRITE) ? 130 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 131 vm_flags &= (foll_flags & FOLL_FORCE) ? 132 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 133 134 for (i = 0; i < nr_pages; i++) { 135 vma = find_vma(mm, start); 136 if (!vma) 137 goto finish_or_fault; 138 139 /* protect what we can, including chardevs */ 140 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 141 !(vm_flags & vma->vm_flags)) 142 goto finish_or_fault; 143 144 if (pages) { 145 pages[i] = virt_to_page(start); 146 if (pages[i]) 147 page_cache_get(pages[i]); 148 } 149 if (vmas) 150 vmas[i] = vma; 151 start = (start + PAGE_SIZE) & PAGE_MASK; 152 } 153 154 return i; 155 156finish_or_fault: 157 return i ? : -EFAULT; 158} 159 160/* 161 * get a list of pages in an address range belonging to the specified process 162 * and indicate the VMA that covers each page 163 * - this is potentially dodgy as we may end incrementing the page count of a 164 * slab page or a secondary page from a compound page 165 * - don't permit access to VMAs that don't support it, such as I/O mappings 166 */ 167int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 168 unsigned long start, int nr_pages, int write, int force, 169 struct page **pages, struct vm_area_struct **vmas) 170{ 171 int flags = 0; 172 173 if (write) 174 flags |= FOLL_WRITE; 175 if (force) 176 flags |= FOLL_FORCE; 177 178 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); 179} 180EXPORT_SYMBOL(get_user_pages); 181 182/** 183 * follow_pfn - look up PFN at a user virtual address 184 * @vma: memory mapping 185 * @address: user virtual address 186 * @pfn: location to store found PFN 187 * 188 * Only IO mappings and raw PFN mappings are allowed. 189 * 190 * Returns zero and the pfn at @pfn on success, -ve otherwise. 191 */ 192int follow_pfn(struct vm_area_struct *vma, unsigned long address, 193 unsigned long *pfn) 194{ 195 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 196 return -EINVAL; 197 198 *pfn = address >> PAGE_SHIFT; 199 return 0; 200} 201EXPORT_SYMBOL(follow_pfn); 202 203DEFINE_RWLOCK(vmlist_lock); 204struct vm_struct *vmlist; 205 206void vfree(const void *addr) 207{ 208 kfree(addr); 209} 210EXPORT_SYMBOL(vfree); 211 212void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 213{ 214 /* 215 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 216 * returns only a logical address. 217 */ 218 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 219} 220EXPORT_SYMBOL(__vmalloc); 221 222void *vmalloc_user(unsigned long size) 223{ 224 void *ret; 225 226 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 227 PAGE_KERNEL); 228 if (ret) { 229 struct vm_area_struct *vma; 230 231 down_write(¤t->mm->mmap_sem); 232 vma = find_vma(current->mm, (unsigned long)ret); 233 if (vma) 234 vma->vm_flags |= VM_USERMAP; 235 up_write(¤t->mm->mmap_sem); 236 } 237 238 return ret; 239} 240EXPORT_SYMBOL(vmalloc_user); 241 242struct page *vmalloc_to_page(const void *addr) 243{ 244 return virt_to_page(addr); 245} 246EXPORT_SYMBOL(vmalloc_to_page); 247 248unsigned long vmalloc_to_pfn(const void *addr) 249{ 250 return page_to_pfn(virt_to_page(addr)); 251} 252EXPORT_SYMBOL(vmalloc_to_pfn); 253 254long vread(char *buf, char *addr, unsigned long count) 255{ 256 memcpy(buf, addr, count); 257 return count; 258} 259 260long vwrite(char *buf, char *addr, unsigned long count) 261{ 262 /* Don't allow overflow */ 263 if ((unsigned long) addr + count < count) 264 count = -(unsigned long) addr; 265 266 memcpy(addr, buf, count); 267 return(count); 268} 269 270/* 271 * vmalloc - allocate virtually continguos memory 272 * 273 * @size: allocation size 274 * 275 * Allocate enough pages to cover @size from the page level 276 * allocator and map them into continguos kernel virtual space. 277 * 278 * For tight control over page level allocator and protection flags 279 * use __vmalloc() instead. 280 */ 281void *vmalloc(unsigned long size) 282{ 283 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 284} 285EXPORT_SYMBOL(vmalloc); 286 287void *vmalloc_node(unsigned long size, int node) 288{ 289 return vmalloc(size); 290} 291EXPORT_SYMBOL(vmalloc_node); 292 293#ifndef PAGE_KERNEL_EXEC 294# define PAGE_KERNEL_EXEC PAGE_KERNEL 295#endif 296 297/** 298 * vmalloc_exec - allocate virtually contiguous, executable memory 299 * @size: allocation size 300 * 301 * Kernel-internal function to allocate enough pages to cover @size 302 * the page level allocator and map them into contiguous and 303 * executable kernel virtual space. 304 * 305 * For tight control over page level allocator and protection flags 306 * use __vmalloc() instead. 307 */ 308 309void *vmalloc_exec(unsigned long size) 310{ 311 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 312} 313 314/** 315 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 316 * @size: allocation size 317 * 318 * Allocate enough 32bit PA addressable pages to cover @size from the 319 * page level allocator and map them into continguos kernel virtual space. 320 */ 321void *vmalloc_32(unsigned long size) 322{ 323 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 324} 325EXPORT_SYMBOL(vmalloc_32); 326 327/** 328 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 329 * @size: allocation size 330 * 331 * The resulting memory area is 32bit addressable and zeroed so it can be 332 * mapped to userspace without leaking data. 333 * 334 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 335 * remap_vmalloc_range() are permissible. 336 */ 337void *vmalloc_32_user(unsigned long size) 338{ 339 /* 340 * We'll have to sort out the ZONE_DMA bits for 64-bit, 341 * but for now this can simply use vmalloc_user() directly. 342 */ 343 return vmalloc_user(size); 344} 345EXPORT_SYMBOL(vmalloc_32_user); 346 347void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 348{ 349 BUG(); 350 return NULL; 351} 352EXPORT_SYMBOL(vmap); 353 354void vunmap(const void *addr) 355{ 356 BUG(); 357} 358EXPORT_SYMBOL(vunmap); 359 360void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 361{ 362 BUG(); 363 return NULL; 364} 365EXPORT_SYMBOL(vm_map_ram); 366 367void vm_unmap_ram(const void *mem, unsigned int count) 368{ 369 BUG(); 370} 371EXPORT_SYMBOL(vm_unmap_ram); 372 373void vm_unmap_aliases(void) 374{ 375} 376EXPORT_SYMBOL_GPL(vm_unmap_aliases); 377 378/* 379 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 380 * have one. 381 */ 382void __attribute__((weak)) vmalloc_sync_all(void) 383{ 384} 385 386int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 387 struct page *page) 388{ 389 return -EINVAL; 390} 391EXPORT_SYMBOL(vm_insert_page); 392 393/* 394 * sys_brk() for the most part doesn't need the global kernel 395 * lock, except when an application is doing something nasty 396 * like trying to un-brk an area that has already been mapped 397 * to a regular file. in this case, the unmapping will need 398 * to invoke file system routines that need the global lock. 399 */ 400SYSCALL_DEFINE1(brk, unsigned long, brk) 401{ 402 struct mm_struct *mm = current->mm; 403 404 if (brk < mm->start_brk || brk > mm->context.end_brk) 405 return mm->brk; 406 407 if (mm->brk == brk) 408 return mm->brk; 409 410 /* 411 * Always allow shrinking brk 412 */ 413 if (brk <= mm->brk) { 414 mm->brk = brk; 415 return brk; 416 } 417 418 /* 419 * Ok, looks good - let it rip. 420 */ 421 flush_icache_range(mm->brk, brk); 422 return mm->brk = brk; 423} 424 425/* 426 * initialise the VMA and region record slabs 427 */ 428void __init mmap_init(void) 429{ 430 int ret; 431 432 ret = percpu_counter_init(&vm_committed_as, 0); 433 VM_BUG_ON(ret); 434 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 435} 436 437/* 438 * validate the region tree 439 * - the caller must hold the region lock 440 */ 441#ifdef CONFIG_DEBUG_NOMMU_REGIONS 442static noinline void validate_nommu_regions(void) 443{ 444 struct vm_region *region, *last; 445 struct rb_node *p, *lastp; 446 447 lastp = rb_first(&nommu_region_tree); 448 if (!lastp) 449 return; 450 451 last = rb_entry(lastp, struct vm_region, vm_rb); 452 BUG_ON(unlikely(last->vm_end <= last->vm_start)); 453 BUG_ON(unlikely(last->vm_top < last->vm_end)); 454 455 while ((p = rb_next(lastp))) { 456 region = rb_entry(p, struct vm_region, vm_rb); 457 last = rb_entry(lastp, struct vm_region, vm_rb); 458 459 BUG_ON(unlikely(region->vm_end <= region->vm_start)); 460 BUG_ON(unlikely(region->vm_top < region->vm_end)); 461 BUG_ON(unlikely(region->vm_start < last->vm_top)); 462 463 lastp = p; 464 } 465} 466#else 467static void validate_nommu_regions(void) 468{ 469} 470#endif 471 472/* 473 * add a region into the global tree 474 */ 475static void add_nommu_region(struct vm_region *region) 476{ 477 struct vm_region *pregion; 478 struct rb_node **p, *parent; 479 480 validate_nommu_regions(); 481 482 parent = NULL; 483 p = &nommu_region_tree.rb_node; 484 while (*p) { 485 parent = *p; 486 pregion = rb_entry(parent, struct vm_region, vm_rb); 487 if (region->vm_start < pregion->vm_start) 488 p = &(*p)->rb_left; 489 else if (region->vm_start > pregion->vm_start) 490 p = &(*p)->rb_right; 491 else if (pregion == region) 492 return; 493 else 494 BUG(); 495 } 496 497 rb_link_node(®ion->vm_rb, parent, p); 498 rb_insert_color(®ion->vm_rb, &nommu_region_tree); 499 500 validate_nommu_regions(); 501} 502 503/* 504 * delete a region from the global tree 505 */ 506static void delete_nommu_region(struct vm_region *region) 507{ 508 BUG_ON(!nommu_region_tree.rb_node); 509 510 validate_nommu_regions(); 511 rb_erase(®ion->vm_rb, &nommu_region_tree); 512 validate_nommu_regions(); 513} 514 515/* 516 * free a contiguous series of pages 517 */ 518static void free_page_series(unsigned long from, unsigned long to) 519{ 520 for (; from < to; from += PAGE_SIZE) { 521 struct page *page = virt_to_page(from); 522 523 kdebug("- free %lx", from); 524 atomic_long_dec(&mmap_pages_allocated); 525 if (page_count(page) != 1) 526 kdebug("free page %p: refcount not one: %d", 527 page, page_count(page)); 528 put_page(page); 529 } 530} 531 532/* 533 * release a reference to a region 534 * - the caller must hold the region semaphore for writing, which this releases 535 * - the region may not have been added to the tree yet, in which case vm_top 536 * will equal vm_start 537 */ 538static void __put_nommu_region(struct vm_region *region) 539 __releases(nommu_region_sem) 540{ 541 kenter("%p{%d}", region, region->vm_usage); 542 543 BUG_ON(!nommu_region_tree.rb_node); 544 545 if (--region->vm_usage == 0) { 546 if (region->vm_top > region->vm_start) 547 delete_nommu_region(region); 548 up_write(&nommu_region_sem); 549 550 if (region->vm_file) 551 fput(region->vm_file); 552 553 /* IO memory and memory shared directly out of the pagecache 554 * from ramfs/tmpfs mustn't be released here */ 555 if (region->vm_flags & VM_MAPPED_COPY) { 556 kdebug("free series"); 557 free_page_series(region->vm_start, region->vm_top); 558 } 559 kmem_cache_free(vm_region_jar, region); 560 } else { 561 up_write(&nommu_region_sem); 562 } 563} 564 565/* 566 * release a reference to a region 567 */ 568static void put_nommu_region(struct vm_region *region) 569{ 570 down_write(&nommu_region_sem); 571 __put_nommu_region(region); 572} 573 574/* 575 * update protection on a vma 576 */ 577static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 578{ 579#ifdef CONFIG_MPU 580 struct mm_struct *mm = vma->vm_mm; 581 long start = vma->vm_start & PAGE_MASK; 582 while (start < vma->vm_end) { 583 protect_page(mm, start, flags); 584 start += PAGE_SIZE; 585 } 586 update_protections(mm); 587#endif 588} 589 590/* 591 * add a VMA into a process's mm_struct in the appropriate place in the list 592 * and tree and add to the address space's page tree also if not an anonymous 593 * page 594 * - should be called with mm->mmap_sem held writelocked 595 */ 596static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 597{ 598 struct vm_area_struct *pvma, **pp, *next; 599 struct address_space *mapping; 600 struct rb_node **p, *parent; 601 602 kenter(",%p", vma); 603 604 BUG_ON(!vma->vm_region); 605 606 mm->map_count++; 607 vma->vm_mm = mm; 608 609 protect_vma(vma, vma->vm_flags); 610 611 /* add the VMA to the mapping */ 612 if (vma->vm_file) { 613 mapping = vma->vm_file->f_mapping; 614 615 flush_dcache_mmap_lock(mapping); 616 vma_prio_tree_insert(vma, &mapping->i_mmap); 617 flush_dcache_mmap_unlock(mapping); 618 } 619 620 /* add the VMA to the tree */ 621 parent = NULL; 622 p = &mm->mm_rb.rb_node; 623 while (*p) { 624 parent = *p; 625 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 626 627 /* sort by: start addr, end addr, VMA struct addr in that order 628 * (the latter is necessary as we may get identical VMAs) */ 629 if (vma->vm_start < pvma->vm_start) 630 p = &(*p)->rb_left; 631 else if (vma->vm_start > pvma->vm_start) 632 p = &(*p)->rb_right; 633 else if (vma->vm_end < pvma->vm_end) 634 p = &(*p)->rb_left; 635 else if (vma->vm_end > pvma->vm_end) 636 p = &(*p)->rb_right; 637 else if (vma < pvma) 638 p = &(*p)->rb_left; 639 else if (vma > pvma) 640 p = &(*p)->rb_right; 641 else 642 BUG(); 643 } 644 645 rb_link_node(&vma->vm_rb, parent, p); 646 rb_insert_color(&vma->vm_rb, &mm->mm_rb); 647 648 /* add VMA to the VMA list also */ 649 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) { 650 if (pvma->vm_start > vma->vm_start) 651 break; 652 if (pvma->vm_start < vma->vm_start) 653 continue; 654 if (pvma->vm_end < vma->vm_end) 655 break; 656 } 657 658 next = *pp; 659 *pp = vma; 660 vma->vm_next = next; 661 if (next) 662 next->vm_prev = vma; 663} 664 665/* 666 * delete a VMA from its owning mm_struct and address space 667 */ 668static void delete_vma_from_mm(struct vm_area_struct *vma) 669{ 670 struct vm_area_struct **pp; 671 struct address_space *mapping; 672 struct mm_struct *mm = vma->vm_mm; 673 674 kenter("%p", vma); 675 676 protect_vma(vma, 0); 677 678 mm->map_count--; 679 if (mm->mmap_cache == vma) 680 mm->mmap_cache = NULL; 681 682 /* remove the VMA from the mapping */ 683 if (vma->vm_file) { 684 mapping = vma->vm_file->f_mapping; 685 686 flush_dcache_mmap_lock(mapping); 687 vma_prio_tree_remove(vma, &mapping->i_mmap); 688 flush_dcache_mmap_unlock(mapping); 689 } 690 691 /* remove from the MM's tree and list */ 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 */ 706static 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 */ 724struct 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} 748EXPORT_SYMBOL(find_vma); 749 750/* 751 * find a VMA 752 * - we don't extend stack VMAs under NOMMU conditions 753 */ 754struct 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 */ 763int 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 */ 772static 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; 800} 801 802/* 803 * determine whether a mapping should be permitted and, if so, what sort of 804 * mapping we're capable of supporting 805 */ 806static int validate_mmap_request(struct file *file, 807 unsigned long addr, 808 unsigned long len, 809 unsigned long prot, 810 unsigned long flags, 811 unsigned long pgoff, 812 unsigned long *_capabilities) 813{ 814 unsigned long capabilities, rlen; 815 unsigned long reqprot = prot; 816 int ret; 817 818 /* do the simple checks first */ 819 if (flags & MAP_FIXED) { 820 printk(KERN_DEBUG 821 "%d: Can't do fixed-address/overlay mmap of RAM\n", 822 current->pid); 823 return -EINVAL; 824 } 825 826 if ((flags & MAP_TYPE) != MAP_PRIVATE && 827 (flags & MAP_TYPE) != MAP_SHARED) 828 return -EINVAL; 829 830 if (!len) 831 return -EINVAL; 832 833 /* Careful about overflows.. */ 834 rlen = PAGE_ALIGN(len); 835 if (!rlen || rlen > TASK_SIZE) 836 return -ENOMEM; 837 838 /* offset overflow? */ 839 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 840 return -EOVERFLOW; 841 842 if (file) { 843 /* validate file mapping requests */ 844 struct address_space *mapping; 845 846 /* files must support mmap */ 847 if (!file->f_op || !file->f_op->mmap) 848 return -ENODEV; 849 850 /* work out if what we've got could possibly be shared 851 * - we support chardevs that provide their own "memory" 852 * - we support files/blockdevs that are memory backed 853 */ 854 mapping = file->f_mapping; 855 if (!mapping) 856 mapping = file->f_path.dentry->d_inode->i_mapping; 857 858 capabilities = 0; 859 if (mapping && mapping->backing_dev_info) 860 capabilities = mapping->backing_dev_info->capabilities; 861 862 if (!capabilities) { 863 /* no explicit capabilities set, so assume some 864 * defaults */ 865 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 866 case S_IFREG: 867 case S_IFBLK: 868 capabilities = BDI_CAP_MAP_COPY; 869 break; 870 871 case S_IFCHR: 872 capabilities = 873 BDI_CAP_MAP_DIRECT | 874 BDI_CAP_READ_MAP | 875 BDI_CAP_WRITE_MAP; 876 break; 877 878 default: 879 return -EINVAL; 880 } 881 } 882 883 /* eliminate any capabilities that we can't support on this 884 * device */ 885 if (!file->f_op->get_unmapped_area) 886 capabilities &= ~BDI_CAP_MAP_DIRECT; 887 if (!file->f_op->read) 888 capabilities &= ~BDI_CAP_MAP_COPY; 889 890 /* The file shall have been opened with read permission. */ 891 if (!(file->f_mode & FMODE_READ)) 892 return -EACCES; 893 894 if (flags & MAP_SHARED) { 895 /* do checks for writing, appending and locking */ 896 if ((prot & PROT_WRITE) && 897 !(file->f_mode & FMODE_WRITE)) 898 return -EACCES; 899 900 if (IS_APPEND(file->f_path.dentry->d_inode) && 901 (file->f_mode & FMODE_WRITE)) 902 return -EACCES; 903 904 if (locks_verify_locked(file->f_path.dentry->d_inode)) 905 return -EAGAIN; 906 907 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 908 return -ENODEV; 909 910 /* we mustn't privatise shared mappings */ 911 capabilities &= ~BDI_CAP_MAP_COPY; 912 } 913 else { 914 /* we're going to read the file into private memory we 915 * allocate */ 916 if (!(capabilities & BDI_CAP_MAP_COPY)) 917 return -ENODEV; 918 919 /* we don't permit a private writable mapping to be 920 * shared with the backing device */ 921 if (prot & PROT_WRITE) 922 capabilities &= ~BDI_CAP_MAP_DIRECT; 923 } 924 925 if (capabilities & BDI_CAP_MAP_DIRECT) { 926 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 927 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 928 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 929 ) { 930 capabilities &= ~BDI_CAP_MAP_DIRECT; 931 if (flags & MAP_SHARED) { 932 printk(KERN_WARNING 933 "MAP_SHARED not completely supported on !MMU\n"); 934 return -EINVAL; 935 } 936 } 937 } 938 939 /* handle executable mappings and implied executable 940 * mappings */ 941 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 942 if (prot & PROT_EXEC) 943 return -EPERM; 944 } 945 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 946 /* handle implication of PROT_EXEC by PROT_READ */ 947 if (current->personality & READ_IMPLIES_EXEC) { 948 if (capabilities & BDI_CAP_EXEC_MAP) 949 prot |= PROT_EXEC; 950 } 951 } 952 else if ((prot & PROT_READ) && 953 (prot & PROT_EXEC) && 954 !(capabilities & BDI_CAP_EXEC_MAP) 955 ) { 956 /* backing file is not executable, try to copy */ 957 capabilities &= ~BDI_CAP_MAP_DIRECT; 958 } 959 } 960 else { 961 /* anonymous mappings are always memory backed and can be 962 * privately mapped 963 */ 964 capabilities = BDI_CAP_MAP_COPY; 965 966 /* handle PROT_EXEC implication by PROT_READ */ 967 if ((prot & PROT_READ) && 968 (current->personality & READ_IMPLIES_EXEC)) 969 prot |= PROT_EXEC; 970 } 971 972 /* allow the security API to have its say */ 973 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 974 if (ret < 0) 975 return ret; 976 977 /* looks okay */ 978 *_capabilities = capabilities; 979 return 0; 980} 981 982/* 983 * we've determined that we can make the mapping, now translate what we 984 * now know into VMA flags 985 */ 986static unsigned long determine_vm_flags(struct file *file, 987 unsigned long prot, 988 unsigned long flags, 989 unsigned long capabilities) 990{ 991 unsigned long vm_flags; 992 993 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 994 /* vm_flags |= mm->def_flags; */ 995 996 if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 997 /* attempt to share read-only copies of mapped file chunks */ 998 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 999 if (file && !(prot & PROT_WRITE)) 1000 vm_flags |= VM_MAYSHARE; 1001 } else { 1002 /* overlay a shareable mapping on the backing device or inode 1003 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 1004 * romfs/cramfs */ 1005 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); 1006 if (flags & MAP_SHARED) 1007 vm_flags |= VM_SHARED; 1008 } 1009 1010 /* refuse to let anyone share private mappings with this process if 1011 * it's being traced - otherwise breakpoints set in it may interfere 1012 * with another untraced process 1013 */ 1014 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) 1015 vm_flags &= ~VM_MAYSHARE; 1016 1017 return vm_flags; 1018} 1019 1020/* 1021 * set up a shared mapping on a file (the driver or filesystem provides and 1022 * pins the storage) 1023 */ 1024static int do_mmap_shared_file(struct vm_area_struct *vma) 1025{ 1026 int ret; 1027 1028 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1029 if (ret == 0) { 1030 vma->vm_region->vm_top = vma->vm_region->vm_end; 1031 return 0; 1032 } 1033 if (ret != -ENOSYS) 1034 return ret; 1035 1036 /* getting -ENOSYS indicates that direct mmap isn't possible (as 1037 * opposed to tried but failed) so we can only give a suitable error as 1038 * it's not possible to make a private copy if MAP_SHARED was given */ 1039 return -ENODEV; 1040} 1041 1042/* 1043 * set up a private mapping or an anonymous shared mapping 1044 */ 1045static int do_mmap_private(struct vm_area_struct *vma, 1046 struct vm_region *region, 1047 unsigned long len, 1048 unsigned long capabilities) 1049{ 1050 struct page *pages; 1051 unsigned long total, point, n, rlen; 1052 void *base; 1053 int ret, order; 1054 1055 /* invoke the file's mapping function so that it can keep track of 1056 * shared mappings on devices or memory 1057 * - VM_MAYSHARE will be set if it may attempt to share 1058 */ 1059 if (capabilities & BDI_CAP_MAP_DIRECT) { 1060 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1061 if (ret == 0) { 1062 /* shouldn't return success if we're not sharing */ 1063 BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1064 vma->vm_region->vm_top = vma->vm_region->vm_end; 1065 return 0; 1066 } 1067 if (ret != -ENOSYS) 1068 return ret; 1069 1070 /* getting an ENOSYS error indicates that direct mmap isn't 1071 * possible (as opposed to tried but failed) so we'll try to 1072 * make a private copy of the data and map that instead */ 1073 } 1074 1075 rlen = PAGE_ALIGN(len); 1076 1077 /* allocate some memory to hold the mapping 1078 * - note that this may not return a page-aligned address if the object 1079 * we're allocating is smaller than a page 1080 */ 1081 order = get_order(rlen); 1082 kdebug("alloc order %d for %lx", order, len); 1083 1084 pages = alloc_pages(GFP_KERNEL, order); 1085 if (!pages) 1086 goto enomem; 1087 1088 total = 1 << order; 1089 atomic_long_add(total, &mmap_pages_allocated); 1090 1091 point = rlen >> PAGE_SHIFT; 1092 1093 /* we allocated a power-of-2 sized page set, so we may want to trim off 1094 * the excess */ 1095 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 1096 while (total > point) { 1097 order = ilog2(total - point); 1098 n = 1 << order; 1099 kdebug("shave %lu/%lu @%lu", n, total - point, total); 1100 atomic_long_sub(n, &mmap_pages_allocated); 1101 total -= n; 1102 set_page_refcounted(pages + total); 1103 __free_pages(pages + total, order); 1104 } 1105 } 1106 1107 for (point = 1; point < total; point++) 1108 set_page_refcounted(&pages[point]); 1109 1110 base = page_address(pages); 1111 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 1112 region->vm_start = (unsigned long) base; 1113 region->vm_end = region->vm_start + rlen; 1114 region->vm_top = region->vm_start + (total << PAGE_SHIFT); 1115 1116 vma->vm_start = region->vm_start; 1117 vma->vm_end = region->vm_start + len; 1118 1119 if (vma->vm_file) { 1120 /* read the contents of a file into the copy */ 1121 mm_segment_t old_fs; 1122 loff_t fpos; 1123 1124 fpos = vma->vm_pgoff; 1125 fpos <<= PAGE_SHIFT; 1126 1127 old_fs = get_fs(); 1128 set_fs(KERNEL_DS); 1129 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); 1130 set_fs(old_fs); 1131 1132 if (ret < 0) 1133 goto error_free; 1134 1135 /* clear the last little bit */ 1136 if (ret < rlen) 1137 memset(base + ret, 0, rlen - ret); 1138 1139 } 1140 1141 return 0; 1142 1143error_free: 1144 free_page_series(region->vm_start, region->vm_end); 1145 region->vm_start = vma->vm_start = 0; 1146 region->vm_end = vma->vm_end = 0; 1147 region->vm_top = 0; 1148 return ret; 1149 1150enomem: 1151 printk("Allocation of length %lu from process %d (%s) failed\n", 1152 len, current->pid, current->comm); 1153 show_free_areas(); 1154 return -ENOMEM; 1155} 1156 1157/* 1158 * handle mapping creation for uClinux 1159 */ 1160unsigned long do_mmap_pgoff(struct file *file, 1161 unsigned long addr, 1162 unsigned long len, 1163 unsigned long prot, 1164 unsigned long flags, 1165 unsigned long pgoff) 1166{ 1167 struct vm_area_struct *vma; 1168 struct vm_region *region; 1169 struct rb_node *rb; 1170 unsigned long capabilities, vm_flags, result; 1171 int ret; 1172 1173 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 1174 1175 /* decide whether we should attempt the mapping, and if so what sort of 1176 * mapping */ 1177 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 1178 &capabilities); 1179 if (ret < 0) { 1180 kleave(" = %d [val]", ret); 1181 return ret; 1182 } 1183 1184 /* we ignore the address hint */ 1185 addr = 0; 1186 1187 /* we've determined that we can make the mapping, now translate what we 1188 * now know into VMA flags */ 1189 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 1190 1191 /* we're going to need to record the mapping */ 1192 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 1193 if (!region) 1194 goto error_getting_region; 1195 1196 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 1197 if (!vma) 1198 goto error_getting_vma; 1199 1200 region->vm_usage = 1; 1201 region->vm_flags = vm_flags; 1202 region->vm_pgoff = pgoff; 1203 1204 INIT_LIST_HEAD(&vma->anon_vma_chain); 1205 vma->vm_flags = vm_flags; 1206 vma->vm_pgoff = pgoff; 1207 1208 if (file) { 1209 region->vm_file = file; 1210 get_file(file); 1211 vma->vm_file = file; 1212 get_file(file); 1213 if (vm_flags & VM_EXECUTABLE) { 1214 added_exe_file_vma(current->mm); 1215 vma->vm_mm = current->mm; 1216 } 1217 } 1218 1219 down_write(&nommu_region_sem); 1220 1221 /* if we want to share, we need to check for regions created by other 1222 * mmap() calls that overlap with our proposed mapping 1223 * - we can only share with a superset match on most regular files 1224 * - shared mappings on character devices and memory backed files are 1225 * permitted to overlap inexactly as far as we are concerned for in 1226 * these cases, sharing is handled in the driver or filesystem rather 1227 * than here 1228 */ 1229 if (vm_flags & VM_MAYSHARE) { 1230 struct vm_region *pregion; 1231 unsigned long pglen, rpglen, pgend, rpgend, start; 1232 1233 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1234 pgend = pgoff + pglen; 1235 1236 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 1237 pregion = rb_entry(rb, struct vm_region, vm_rb); 1238 1239 if (!(pregion->vm_flags & VM_MAYSHARE)) 1240 continue; 1241 1242 /* search for overlapping mappings on the same file */ 1243 if (pregion->vm_file->f_path.dentry->d_inode != 1244 file->f_path.dentry->d_inode) 1245 continue; 1246 1247 if (pregion->vm_pgoff >= pgend) 1248 continue; 1249 1250 rpglen = pregion->vm_end - pregion->vm_start; 1251 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 1252 rpgend = pregion->vm_pgoff + rpglen; 1253 if (pgoff >= rpgend) 1254 continue; 1255 1256 /* handle inexactly overlapping matches between 1257 * mappings */ 1258 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 1259 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 1260 /* new mapping is not a subset of the region */ 1261 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 1262 goto sharing_violation; 1263 continue; 1264 } 1265 1266 /* we've found a region we can share */ 1267 pregion->vm_usage++; 1268 vma->vm_region = pregion; 1269 start = pregion->vm_start; 1270 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 1271 vma->vm_start = start; 1272 vma->vm_end = start + len; 1273 1274 if (pregion->vm_flags & VM_MAPPED_COPY) { 1275 kdebug("share copy"); 1276 vma->vm_flags |= VM_MAPPED_COPY; 1277 } else { 1278 kdebug("share mmap"); 1279 ret = do_mmap_shared_file(vma); 1280 if (ret < 0) { 1281 vma->vm_region = NULL; 1282 vma->vm_start = 0; 1283 vma->vm_end = 0; 1284 pregion->vm_usage--; 1285 pregion = NULL; 1286 goto error_just_free; 1287 } 1288 } 1289 fput(region->vm_file); 1290 kmem_cache_free(vm_region_jar, region); 1291 region = pregion; 1292 result = start; 1293 goto share; 1294 } 1295 1296 /* obtain the address at which to make a shared mapping 1297 * - this is the hook for quasi-memory character devices to 1298 * tell us the location of a shared mapping 1299 */ 1300 if (capabilities & BDI_CAP_MAP_DIRECT) { 1301 addr = file->f_op->get_unmapped_area(file, addr, len, 1302 pgoff, flags); 1303 if (IS_ERR((void *) addr)) { 1304 ret = addr; 1305 if (ret != (unsigned long) -ENOSYS) 1306 goto error_just_free; 1307 1308 /* the driver refused to tell us where to site 1309 * the mapping so we'll have to attempt to copy 1310 * it */ 1311 ret = (unsigned long) -ENODEV; 1312 if (!(capabilities & BDI_CAP_MAP_COPY)) 1313 goto error_just_free; 1314 1315 capabilities &= ~BDI_CAP_MAP_DIRECT; 1316 } else { 1317 vma->vm_start = region->vm_start = addr; 1318 vma->vm_end = region->vm_end = addr + len; 1319 } 1320 } 1321 } 1322 1323 vma->vm_region = region; 1324 1325 /* set up the mapping 1326 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set 1327 */ 1328 if (file && vma->vm_flags & VM_SHARED) 1329 ret = do_mmap_shared_file(vma); 1330 else 1331 ret = do_mmap_private(vma, region, len, capabilities); 1332 if (ret < 0) 1333 goto error_just_free; 1334 add_nommu_region(region); 1335 1336 /* clear anonymous mappings that don't ask for uninitialized data */ 1337 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1338 memset((void *)region->vm_start, 0, 1339 region->vm_end - region->vm_start); 1340 1341 /* okay... we have a mapping; now we have to register it */ 1342 result = vma->vm_start; 1343 1344 current->mm->total_vm += len >> PAGE_SHIFT; 1345 1346share: 1347 add_vma_to_mm(current->mm, vma); 1348 1349 /* we flush the region from the icache only when the first executable 1350 * mapping of it is made */ 1351 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1352 flush_icache_range(region->vm_start, region->vm_end); 1353 region->vm_icache_flushed = true; 1354 } 1355 1356 up_write(&nommu_region_sem); 1357 1358 kleave(" = %lx", result); 1359 return result; 1360 1361error_just_free: 1362 up_write(&nommu_region_sem); 1363error: 1364 if (region->vm_file) 1365 fput(region->vm_file); 1366 kmem_cache_free(vm_region_jar, region); 1367 if (vma->vm_file) 1368 fput(vma->vm_file); 1369 if (vma->vm_flags & VM_EXECUTABLE) 1370 removed_exe_file_vma(vma->vm_mm); 1371 kmem_cache_free(vm_area_cachep, vma); 1372 kleave(" = %d", ret); 1373 return ret; 1374 1375sharing_violation: 1376 up_write(&nommu_region_sem); 1377 printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 1378 ret = -EINVAL; 1379 goto error; 1380 1381error_getting_vma: 1382 kmem_cache_free(vm_region_jar, region); 1383 printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 1384 " from process %d failed\n", 1385 len, current->pid); 1386 show_free_areas(); 1387 return -ENOMEM; 1388 1389error_getting_region: 1390 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 1391 " from process %d failed\n", 1392 len, current->pid); 1393 show_free_areas(); 1394 return -ENOMEM; 1395} 1396EXPORT_SYMBOL(do_mmap_pgoff); 1397 1398SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1399 unsigned long, prot, unsigned long, flags, 1400 unsigned long, fd, unsigned long, pgoff) 1401{ 1402 struct file *file = NULL; 1403 unsigned long retval = -EBADF; 1404 1405 if (!(flags & MAP_ANONYMOUS)) { 1406 file = fget(fd); 1407 if (!file) 1408 goto out; 1409 } 1410 1411 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 1412 1413 down_write(¤t->mm->mmap_sem); 1414 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1415 up_write(¤t->mm->mmap_sem); 1416 1417 if (file) 1418 fput(file); 1419out: 1420 return retval; 1421} 1422 1423#ifdef __ARCH_WANT_SYS_OLD_MMAP 1424struct mmap_arg_struct { 1425 unsigned long addr; 1426 unsigned long len; 1427 unsigned long prot; 1428 unsigned long flags; 1429 unsigned long fd; 1430 unsigned long offset; 1431}; 1432 1433SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1434{ 1435 struct mmap_arg_struct a; 1436 1437 if (copy_from_user(&a, arg, sizeof(a))) 1438 return -EFAULT; 1439 if (a.offset & ~PAGE_MASK) 1440 return -EINVAL; 1441 1442 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1443 a.offset >> PAGE_SHIFT); 1444} 1445#endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1446 1447/* 1448 * split a vma into two pieces at address 'addr', a new vma is allocated either 1449 * for the first part or the tail. 1450 */ 1451int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 1452 unsigned long addr, int new_below) 1453{ 1454 struct vm_area_struct *new; 1455 struct vm_region *region; 1456 unsigned long npages; 1457 1458 kenter(""); 1459 1460 /* we're only permitted to split anonymous regions (these should have 1461 * only a single usage on the region) */ 1462 if (vma->vm_file) 1463 return -ENOMEM; 1464 1465 if (mm->map_count >= sysctl_max_map_count) 1466 return -ENOMEM; 1467 1468 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 1469 if (!region) 1470 return -ENOMEM; 1471 1472 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 1473 if (!new) { 1474 kmem_cache_free(vm_region_jar, region); 1475 return -ENOMEM; 1476 } 1477 1478 /* most fields are the same, copy all, and then fixup */ 1479 *new = *vma; 1480 *region = *vma->vm_region; 1481 new->vm_region = region; 1482 1483 npages = (addr - vma->vm_start) >> PAGE_SHIFT; 1484 1485 if (new_below) { 1486 region->vm_top = region->vm_end = new->vm_end = addr; 1487 } else { 1488 region->vm_start = new->vm_start = addr; 1489 region->vm_pgoff = new->vm_pgoff += npages; 1490 } 1491 1492 if (new->vm_ops && new->vm_ops->open) 1493 new->vm_ops->open(new); 1494 1495 delete_vma_from_mm(vma); 1496 down_write(&nommu_region_sem); 1497 delete_nommu_region(vma->vm_region); 1498 if (new_below) { 1499 vma->vm_region->vm_start = vma->vm_start = addr; 1500 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 1501 } else { 1502 vma->vm_region->vm_end = vma->vm_end = addr; 1503 vma->vm_region->vm_top = addr; 1504 } 1505 add_nommu_region(vma->vm_region); 1506 add_nommu_region(new->vm_region); 1507 up_write(&nommu_region_sem); 1508 add_vma_to_mm(mm, vma); 1509 add_vma_to_mm(mm, new); 1510 return 0; 1511} 1512 1513/* 1514 * shrink a VMA by removing the specified chunk from either the beginning or 1515 * the end 1516 */ 1517static int shrink_vma(struct mm_struct *mm, 1518 struct vm_area_struct *vma, 1519 unsigned long from, unsigned long to) 1520{ 1521 struct vm_region *region; 1522 1523 kenter(""); 1524 1525 /* adjust the VMA's pointers, which may reposition it in the MM's tree 1526 * and list */ 1527 delete_vma_from_mm(vma); 1528 if (from > vma->vm_start) 1529 vma->vm_end = from; 1530 else 1531 vma->vm_start = to; 1532 add_vma_to_mm(mm, vma); 1533 1534 /* cut the backing region down to size */ 1535 region = vma->vm_region; 1536 BUG_ON(region->vm_usage != 1); 1537 1538 down_write(&nommu_region_sem); 1539 delete_nommu_region(region); 1540 if (from > region->vm_start) { 1541 to = region->vm_top; 1542 region->vm_top = region->vm_end = from; 1543 } else { 1544 region->vm_start = to; 1545 } 1546 add_nommu_region(region); 1547 up_write(&nommu_region_sem); 1548 1549 free_page_series(from, to); 1550 return 0; 1551} 1552 1553/* 1554 * release a mapping 1555 * - under NOMMU conditions the chunk to be unmapped must be backed by a single 1556 * VMA, though it need not cover the whole VMA 1557 */ 1558int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 1559{ 1560 struct vm_area_struct *vma; 1561 struct rb_node *rb; 1562 unsigned long end = start + len; 1563 int ret; 1564 1565 kenter(",%lx,%zx", start, len); 1566 1567 if (len == 0) 1568 return -EINVAL; 1569 1570 /* find the first potentially overlapping VMA */ 1571 vma = find_vma(mm, start); 1572 if (!vma) { 1573 static int limit = 0; 1574 if (limit < 5) { 1575 printk(KERN_WARNING 1576 "munmap of memory not mmapped by process %d" 1577 " (%s): 0x%lx-0x%lx\n", 1578 current->pid, current->comm, 1579 start, start + len - 1); 1580 limit++; 1581 } 1582 return -EINVAL; 1583 } 1584 1585 /* we're allowed to split an anonymous VMA but not a file-backed one */ 1586 if (vma->vm_file) { 1587 do { 1588 if (start > vma->vm_start) { 1589 kleave(" = -EINVAL [miss]"); 1590 return -EINVAL; 1591 } 1592 if (end == vma->vm_end) 1593 goto erase_whole_vma; 1594 rb = rb_next(&vma->vm_rb); 1595 vma = rb_entry(rb, struct vm_area_struct, vm_rb); 1596 } while (rb); 1597 kleave(" = -EINVAL [split file]"); 1598 return -EINVAL; 1599 } else { 1600 /* the chunk must be a subset of the VMA found */ 1601 if (start == vma->vm_start && end == vma->vm_end) 1602 goto erase_whole_vma; 1603 if (start < vma->vm_start || end > vma->vm_end) { 1604 kleave(" = -EINVAL [superset]"); 1605 return -EINVAL; 1606 } 1607 if (start & ~PAGE_MASK) { 1608 kleave(" = -EINVAL [unaligned start]"); 1609 return -EINVAL; 1610 } 1611 if (end != vma->vm_end && end & ~PAGE_MASK) { 1612 kleave(" = -EINVAL [unaligned split]"); 1613 return -EINVAL; 1614 } 1615 if (start != vma->vm_start && end != vma->vm_end) { 1616 ret = split_vma(mm, vma, start, 1); 1617 if (ret < 0) { 1618 kleave(" = %d [split]", ret); 1619 return ret; 1620 } 1621 } 1622 return shrink_vma(mm, vma, start, end); 1623 } 1624 1625erase_whole_vma: 1626 delete_vma_from_mm(vma); 1627 delete_vma(mm, vma); 1628 kleave(" = 0"); 1629 return 0; 1630} 1631EXPORT_SYMBOL(do_munmap); 1632 1633SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1634{ 1635 int ret; 1636 struct mm_struct *mm = current->mm; 1637 1638 down_write(&mm->mmap_sem); 1639 ret = do_munmap(mm, addr, len); 1640 up_write(&mm->mmap_sem); 1641 return ret; 1642} 1643 1644/* 1645 * release all the mappings made in a process's VM space 1646 */ 1647void exit_mmap(struct mm_struct *mm) 1648{ 1649 struct vm_area_struct *vma; 1650 1651 if (!mm) 1652 return; 1653 1654 kenter(""); 1655 1656 mm->total_vm = 0; 1657 1658 while ((vma = mm->mmap)) { 1659 mm->mmap = vma->vm_next; 1660 delete_vma_from_mm(vma); 1661 delete_vma(mm, vma); 1662 cond_resched(); 1663 } 1664 1665 kleave(""); 1666} 1667 1668unsigned long do_brk(unsigned long addr, unsigned long len) 1669{ 1670 return -ENOMEM; 1671} 1672 1673/* 1674 * expand (or shrink) an existing mapping, potentially moving it at the same 1675 * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 1676 * 1677 * under NOMMU conditions, we only permit changing a mapping's size, and only 1678 * as long as it stays within the region allocated by do_mmap_private() and the 1679 * block is not shareable 1680 * 1681 * MREMAP_FIXED is not supported under NOMMU conditions 1682 */ 1683unsigned long do_mremap(unsigned long addr, 1684 unsigned long old_len, unsigned long new_len, 1685 unsigned long flags, unsigned long new_addr) 1686{ 1687 struct vm_area_struct *vma; 1688 1689 /* insanity checks first */ 1690 if (old_len == 0 || new_len == 0) 1691 return (unsigned long) -EINVAL; 1692 1693 if (addr & ~PAGE_MASK) 1694 return -EINVAL; 1695 1696 if (flags & MREMAP_FIXED && new_addr != addr) 1697 return (unsigned long) -EINVAL; 1698 1699 vma = find_vma_exact(current->mm, addr, old_len); 1700 if (!vma) 1701 return (unsigned long) -EINVAL; 1702 1703 if (vma->vm_end != vma->vm_start + old_len) 1704 return (unsigned long) -EFAULT; 1705 1706 if (vma->vm_flags & VM_MAYSHARE) 1707 return (unsigned long) -EPERM; 1708 1709 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 1710 return (unsigned long) -ENOMEM; 1711 1712 /* all checks complete - do it */ 1713 vma->vm_end = vma->vm_start + new_len; 1714 return vma->vm_start; 1715} 1716EXPORT_SYMBOL(do_mremap); 1717 1718SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 1719 unsigned long, new_len, unsigned long, flags, 1720 unsigned long, new_addr) 1721{ 1722 unsigned long ret; 1723 1724 down_write(¤t->mm->mmap_sem); 1725 ret = do_mremap(addr, old_len, new_len, flags, new_addr); 1726 up_write(¤t->mm->mmap_sem); 1727 return ret; 1728} 1729 1730struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1731 unsigned int foll_flags) 1732{ 1733 return NULL; 1734} 1735 1736int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, 1737 unsigned long to, unsigned long size, pgprot_t prot) 1738{ 1739 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; 1740 return 0; 1741} 1742EXPORT_SYMBOL(remap_pfn_range); 1743 1744int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1745 unsigned long pgoff) 1746{ 1747 unsigned int size = vma->vm_end - vma->vm_start; 1748 1749 if (!(vma->vm_flags & VM_USERMAP)) 1750 return -EINVAL; 1751 1752 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1753 vma->vm_end = vma->vm_start + size; 1754 1755 return 0; 1756} 1757EXPORT_SYMBOL(remap_vmalloc_range); 1758 1759void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 1760{ 1761} 1762 1763unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 1764 unsigned long len, unsigned long pgoff, unsigned long flags) 1765{ 1766 return -ENOMEM; 1767} 1768 1769void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 1770{ 1771} 1772 1773void unmap_mapping_range(struct address_space *mapping, 1774 loff_t const holebegin, loff_t const holelen, 1775 int even_cows) 1776{ 1777} 1778EXPORT_SYMBOL(unmap_mapping_range); 1779 1780/* 1781 * Check that a process has enough memory to allocate a new virtual 1782 * mapping. 0 means there is enough memory for the allocation to 1783 * succeed and -ENOMEM implies there is not. 1784 * 1785 * We currently support three overcommit policies, which are set via the 1786 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1787 * 1788 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1789 * Additional code 2002 Jul 20 by Robert Love. 1790 * 1791 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1792 * 1793 * Note this is a helper function intended to be used by LSMs which 1794 * wish to use this logic. 1795 */ 1796int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1797{ 1798 unsigned long free, allowed; 1799 1800 vm_acct_memory(pages); 1801 1802 /* 1803 * Sometimes we want to use more memory than we have 1804 */ 1805 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1806 return 0; 1807 1808 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1809 unsigned long n; 1810 1811 free = global_page_state(NR_FILE_PAGES); 1812 free += nr_swap_pages; 1813 1814 /* 1815 * Any slabs which are created with the 1816 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1817 * which are reclaimable, under pressure. The dentry 1818 * cache and most inode caches should fall into this 1819 */ 1820 free += global_page_state(NR_SLAB_RECLAIMABLE); 1821 1822 /* 1823 * Leave the last 3% for root 1824 */ 1825 if (!cap_sys_admin) 1826 free -= free / 32; 1827 1828 if (free > pages) 1829 return 0; 1830 1831 /* 1832 * nr_free_pages() is very expensive on large systems, 1833 * only call if we're about to fail. 1834 */ 1835 n = nr_free_pages(); 1836 1837 /* 1838 * Leave reserved pages. The pages are not for anonymous pages. 1839 */ 1840 if (n <= totalreserve_pages) 1841 goto error; 1842 else 1843 n -= totalreserve_pages; 1844 1845 /* 1846 * Leave the last 3% for root 1847 */ 1848 if (!cap_sys_admin) 1849 n -= n / 32; 1850 free += n; 1851 1852 if (free > pages) 1853 return 0; 1854 1855 goto error; 1856 } 1857 1858 allowed = totalram_pages * sysctl_overcommit_ratio / 100; 1859 /* 1860 * Leave the last 3% for root 1861 */ 1862 if (!cap_sys_admin) 1863 allowed -= allowed / 32; 1864 allowed += total_swap_pages; 1865 1866 /* Don't let a single process grow too big: 1867 leave 3% of the size of this process for other processes */ 1868 if (mm) 1869 allowed -= mm->total_vm / 32; 1870 1871 if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1872 return 0; 1873 1874error: 1875 vm_unacct_memory(pages); 1876 1877 return -ENOMEM; 1878} 1879 1880int in_gate_area_no_task(unsigned long addr) 1881{ 1882 return 0; 1883} 1884 1885int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1886{ 1887 BUG(); 1888 return 0; 1889} 1890EXPORT_SYMBOL(filemap_fault); 1891 1892/* 1893 * Access another process' address space. 1894 * - source/target buffer must be kernel space 1895 */ 1896int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 1897{ 1898 struct vm_area_struct *vma; 1899 struct mm_struct *mm; 1900 1901 if (addr + len < addr) 1902 return 0; 1903 1904 mm = get_task_mm(tsk); 1905 if (!mm) 1906 return 0; 1907 1908 down_read(&mm->mmap_sem); 1909 1910 /* the access must start within one of the target process's mappings */ 1911 vma = find_vma(mm, addr); 1912 if (vma) { 1913 /* don't overrun this mapping */ 1914 if (addr + len >= vma->vm_end) 1915 len = vma->vm_end - addr; 1916 1917 /* only read or write mappings where it is permitted */ 1918 if (write && vma->vm_flags & VM_MAYWRITE) 1919 copy_to_user_page(vma, NULL, addr, 1920 (void *) addr, buf, len); 1921 else if (!write && vma->vm_flags & VM_MAYREAD) 1922 copy_from_user_page(vma, NULL, addr, 1923 buf, (void *) addr, len); 1924 else 1925 len = 0; 1926 } else { 1927 len = 0; 1928 } 1929 1930 up_read(&mm->mmap_sem); 1931 mmput(mm); 1932 return len; 1933} 1934 1935/** 1936 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 1937 * @inode: The inode to check 1938 * @size: The current filesize of the inode 1939 * @newsize: The proposed filesize of the inode 1940 * 1941 * Check the shared mappings on an inode on behalf of a shrinking truncate to 1942 * make sure that that any outstanding VMAs aren't broken and then shrink the 1943 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 1944 * automatically grant mappings that are too large. 1945 */ 1946int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 1947 size_t newsize) 1948{ 1949 struct vm_area_struct *vma; 1950 struct prio_tree_iter iter; 1951 struct vm_region *region; 1952 pgoff_t low, high; 1953 size_t r_size, r_top; 1954 1955 low = newsize >> PAGE_SHIFT; 1956 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 1957 1958 down_write(&nommu_region_sem); 1959 1960 /* search for VMAs that fall within the dead zone */ 1961 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, 1962 low, high) { 1963 /* found one - only interested if it's shared out of the page 1964 * cache */ 1965 if (vma->vm_flags & VM_SHARED) { 1966 up_write(&nommu_region_sem); 1967 return -ETXTBSY; /* not quite true, but near enough */ 1968 } 1969 } 1970 1971 /* reduce any regions that overlap the dead zone - if in existence, 1972 * these will be pointed to by VMAs that don't overlap the dead zone 1973 * 1974 * we don't check for any regions that start beyond the EOF as there 1975 * shouldn't be any 1976 */ 1977 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, 1978 0, ULONG_MAX) { 1979 if (!(vma->vm_flags & VM_SHARED)) 1980 continue; 1981 1982 region = vma->vm_region; 1983 r_size = region->vm_top - region->vm_start; 1984 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 1985 1986 if (r_top > newsize) { 1987 region->vm_top -= r_top - newsize; 1988 if (region->vm_end > region->vm_top) 1989 region->vm_end = region->vm_top; 1990 } 1991 } 1992 1993 up_write(&nommu_region_sem); 1994 return 0; 1995} 1996