nommu.c revision b951bf2c4693bfc9744e11293be859209f65f579
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) 2007-2010 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#include <linux/audit.h> 33 34#include <asm/uaccess.h> 35#include <asm/tlb.h> 36#include <asm/tlbflush.h> 37#include <asm/mmu_context.h> 38#include "internal.h" 39 40#if 0 41#define kenter(FMT, ...) \ 42 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 43#define kleave(FMT, ...) \ 44 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 45#define kdebug(FMT, ...) \ 46 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 47#else 48#define kenter(FMT, ...) \ 49 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 50#define kleave(FMT, ...) \ 51 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 52#define kdebug(FMT, ...) \ 53 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 54#endif 55 56void *high_memory; 57struct page *mem_map; 58unsigned long max_mapnr; 59unsigned long num_physpages; 60unsigned long highest_memmap_pfn; 61struct percpu_counter vm_committed_as; 62int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 63int sysctl_overcommit_ratio = 50; /* default is 50% */ 64int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 65int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 66int heap_stack_gap = 0; 67 68atomic_long_t mmap_pages_allocated; 69 70EXPORT_SYMBOL(mem_map); 71EXPORT_SYMBOL(num_physpages); 72 73/* list of mapped, potentially shareable regions */ 74static struct kmem_cache *vm_region_jar; 75struct rb_root nommu_region_tree = RB_ROOT; 76DECLARE_RWSEM(nommu_region_sem); 77 78const struct vm_operations_struct generic_file_vm_ops = { 79}; 80 81/* 82 * Return the total memory allocated for this pointer, not 83 * just what the caller asked for. 84 * 85 * Doesn't have to be accurate, i.e. may have races. 86 */ 87unsigned int kobjsize(const void *objp) 88{ 89 struct page *page; 90 91 /* 92 * If the object we have should not have ksize performed on it, 93 * return size of 0 94 */ 95 if (!objp || !virt_addr_valid(objp)) 96 return 0; 97 98 page = virt_to_head_page(objp); 99 100 /* 101 * If the allocator sets PageSlab, we know the pointer came from 102 * kmalloc(). 103 */ 104 if (PageSlab(page)) 105 return ksize(objp); 106 107 /* 108 * If it's not a compound page, see if we have a matching VMA 109 * region. This test is intentionally done in reverse order, 110 * so if there's no VMA, we still fall through and hand back 111 * PAGE_SIZE for 0-order pages. 112 */ 113 if (!PageCompound(page)) { 114 struct vm_area_struct *vma; 115 116 vma = find_vma(current->mm, (unsigned long)objp); 117 if (vma) 118 return vma->vm_end - vma->vm_start; 119 } 120 121 /* 122 * The ksize() function is only guaranteed to work for pointers 123 * returned by kmalloc(). So handle arbitrary pointers here. 124 */ 125 return PAGE_SIZE << compound_order(page); 126} 127 128int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 129 unsigned long start, int nr_pages, unsigned int foll_flags, 130 struct page **pages, struct vm_area_struct **vmas, 131 int *retry) 132{ 133 struct vm_area_struct *vma; 134 unsigned long vm_flags; 135 int i; 136 137 /* calculate required read or write permissions. 138 * If FOLL_FORCE is set, we only require the "MAY" flags. 139 */ 140 vm_flags = (foll_flags & FOLL_WRITE) ? 141 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 142 vm_flags &= (foll_flags & FOLL_FORCE) ? 143 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 144 145 for (i = 0; i < nr_pages; i++) { 146 vma = find_vma(mm, start); 147 if (!vma) 148 goto finish_or_fault; 149 150 /* protect what we can, including chardevs */ 151 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 152 !(vm_flags & vma->vm_flags)) 153 goto finish_or_fault; 154 155 if (pages) { 156 pages[i] = virt_to_page(start); 157 if (pages[i]) 158 page_cache_get(pages[i]); 159 } 160 if (vmas) 161 vmas[i] = vma; 162 start = (start + PAGE_SIZE) & PAGE_MASK; 163 } 164 165 return i; 166 167finish_or_fault: 168 return i ? : -EFAULT; 169} 170 171/* 172 * get a list of pages in an address range belonging to the specified process 173 * and indicate the VMA that covers each page 174 * - this is potentially dodgy as we may end incrementing the page count of a 175 * slab page or a secondary page from a compound page 176 * - don't permit access to VMAs that don't support it, such as I/O mappings 177 */ 178int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 179 unsigned long start, int nr_pages, int write, int force, 180 struct page **pages, struct vm_area_struct **vmas) 181{ 182 int flags = 0; 183 184 if (write) 185 flags |= FOLL_WRITE; 186 if (force) 187 flags |= FOLL_FORCE; 188 189 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 190 NULL); 191} 192EXPORT_SYMBOL(get_user_pages); 193 194/** 195 * follow_pfn - look up PFN at a user virtual address 196 * @vma: memory mapping 197 * @address: user virtual address 198 * @pfn: location to store found PFN 199 * 200 * Only IO mappings and raw PFN mappings are allowed. 201 * 202 * Returns zero and the pfn at @pfn on success, -ve otherwise. 203 */ 204int follow_pfn(struct vm_area_struct *vma, unsigned long address, 205 unsigned long *pfn) 206{ 207 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 208 return -EINVAL; 209 210 *pfn = address >> PAGE_SHIFT; 211 return 0; 212} 213EXPORT_SYMBOL(follow_pfn); 214 215DEFINE_RWLOCK(vmlist_lock); 216struct vm_struct *vmlist; 217 218void vfree(const void *addr) 219{ 220 kfree(addr); 221} 222EXPORT_SYMBOL(vfree); 223 224void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 225{ 226 /* 227 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 228 * returns only a logical address. 229 */ 230 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 231} 232EXPORT_SYMBOL(__vmalloc); 233 234void *vmalloc_user(unsigned long size) 235{ 236 void *ret; 237 238 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 239 PAGE_KERNEL); 240 if (ret) { 241 struct vm_area_struct *vma; 242 243 down_write(¤t->mm->mmap_sem); 244 vma = find_vma(current->mm, (unsigned long)ret); 245 if (vma) 246 vma->vm_flags |= VM_USERMAP; 247 up_write(¤t->mm->mmap_sem); 248 } 249 250 return ret; 251} 252EXPORT_SYMBOL(vmalloc_user); 253 254struct page *vmalloc_to_page(const void *addr) 255{ 256 return virt_to_page(addr); 257} 258EXPORT_SYMBOL(vmalloc_to_page); 259 260unsigned long vmalloc_to_pfn(const void *addr) 261{ 262 return page_to_pfn(virt_to_page(addr)); 263} 264EXPORT_SYMBOL(vmalloc_to_pfn); 265 266long vread(char *buf, char *addr, unsigned long count) 267{ 268 memcpy(buf, addr, count); 269 return count; 270} 271 272long vwrite(char *buf, char *addr, unsigned long count) 273{ 274 /* Don't allow overflow */ 275 if ((unsigned long) addr + count < count) 276 count = -(unsigned long) addr; 277 278 memcpy(addr, buf, count); 279 return(count); 280} 281 282/* 283 * vmalloc - allocate virtually continguos memory 284 * 285 * @size: allocation size 286 * 287 * Allocate enough pages to cover @size from the page level 288 * allocator and map them into continguos kernel virtual space. 289 * 290 * For tight control over page level allocator and protection flags 291 * use __vmalloc() instead. 292 */ 293void *vmalloc(unsigned long size) 294{ 295 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 296} 297EXPORT_SYMBOL(vmalloc); 298 299/* 300 * vzalloc - allocate virtually continguos memory with zero fill 301 * 302 * @size: allocation size 303 * 304 * Allocate enough pages to cover @size from the page level 305 * allocator and map them into continguos kernel virtual space. 306 * The memory allocated is set to zero. 307 * 308 * For tight control over page level allocator and protection flags 309 * use __vmalloc() instead. 310 */ 311void *vzalloc(unsigned long size) 312{ 313 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 314 PAGE_KERNEL); 315} 316EXPORT_SYMBOL(vzalloc); 317 318/** 319 * vmalloc_node - allocate memory on a specific node 320 * @size: allocation size 321 * @node: numa node 322 * 323 * Allocate enough pages to cover @size from the page level 324 * allocator and map them into contiguous kernel virtual space. 325 * 326 * For tight control over page level allocator and protection flags 327 * use __vmalloc() instead. 328 */ 329void *vmalloc_node(unsigned long size, int node) 330{ 331 return vmalloc(size); 332} 333EXPORT_SYMBOL(vmalloc_node); 334 335/** 336 * vzalloc_node - allocate memory on a specific node with zero fill 337 * @size: allocation size 338 * @node: numa node 339 * 340 * Allocate enough pages to cover @size from the page level 341 * allocator and map them into contiguous kernel virtual space. 342 * The memory allocated is set to zero. 343 * 344 * For tight control over page level allocator and protection flags 345 * use __vmalloc() instead. 346 */ 347void *vzalloc_node(unsigned long size, int node) 348{ 349 return vzalloc(size); 350} 351EXPORT_SYMBOL(vzalloc_node); 352 353#ifndef PAGE_KERNEL_EXEC 354# define PAGE_KERNEL_EXEC PAGE_KERNEL 355#endif 356 357/** 358 * vmalloc_exec - allocate virtually contiguous, executable memory 359 * @size: allocation size 360 * 361 * Kernel-internal function to allocate enough pages to cover @size 362 * the page level allocator and map them into contiguous and 363 * executable kernel virtual space. 364 * 365 * For tight control over page level allocator and protection flags 366 * use __vmalloc() instead. 367 */ 368 369void *vmalloc_exec(unsigned long size) 370{ 371 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 372} 373 374/** 375 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 376 * @size: allocation size 377 * 378 * Allocate enough 32bit PA addressable pages to cover @size from the 379 * page level allocator and map them into continguos kernel virtual space. 380 */ 381void *vmalloc_32(unsigned long size) 382{ 383 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 384} 385EXPORT_SYMBOL(vmalloc_32); 386 387/** 388 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 389 * @size: allocation size 390 * 391 * The resulting memory area is 32bit addressable and zeroed so it can be 392 * mapped to userspace without leaking data. 393 * 394 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 395 * remap_vmalloc_range() are permissible. 396 */ 397void *vmalloc_32_user(unsigned long size) 398{ 399 /* 400 * We'll have to sort out the ZONE_DMA bits for 64-bit, 401 * but for now this can simply use vmalloc_user() directly. 402 */ 403 return vmalloc_user(size); 404} 405EXPORT_SYMBOL(vmalloc_32_user); 406 407void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 408{ 409 BUG(); 410 return NULL; 411} 412EXPORT_SYMBOL(vmap); 413 414void vunmap(const void *addr) 415{ 416 BUG(); 417} 418EXPORT_SYMBOL(vunmap); 419 420void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 421{ 422 BUG(); 423 return NULL; 424} 425EXPORT_SYMBOL(vm_map_ram); 426 427void vm_unmap_ram(const void *mem, unsigned int count) 428{ 429 BUG(); 430} 431EXPORT_SYMBOL(vm_unmap_ram); 432 433void vm_unmap_aliases(void) 434{ 435} 436EXPORT_SYMBOL_GPL(vm_unmap_aliases); 437 438/* 439 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 440 * have one. 441 */ 442void __attribute__((weak)) vmalloc_sync_all(void) 443{ 444} 445 446/** 447 * alloc_vm_area - allocate a range of kernel address space 448 * @size: size of the area 449 * 450 * Returns: NULL on failure, vm_struct on success 451 * 452 * This function reserves a range of kernel address space, and 453 * allocates pagetables to map that range. No actual mappings 454 * are created. If the kernel address space is not shared 455 * between processes, it syncs the pagetable across all 456 * processes. 457 */ 458struct vm_struct *alloc_vm_area(size_t size) 459{ 460 BUG(); 461 return NULL; 462} 463EXPORT_SYMBOL_GPL(alloc_vm_area); 464 465void free_vm_area(struct vm_struct *area) 466{ 467 BUG(); 468} 469EXPORT_SYMBOL_GPL(free_vm_area); 470 471int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 472 struct page *page) 473{ 474 return -EINVAL; 475} 476EXPORT_SYMBOL(vm_insert_page); 477 478/* 479 * sys_brk() for the most part doesn't need the global kernel 480 * lock, except when an application is doing something nasty 481 * like trying to un-brk an area that has already been mapped 482 * to a regular file. in this case, the unmapping will need 483 * to invoke file system routines that need the global lock. 484 */ 485SYSCALL_DEFINE1(brk, unsigned long, brk) 486{ 487 struct mm_struct *mm = current->mm; 488 489 if (brk < mm->start_brk || brk > mm->context.end_brk) 490 return mm->brk; 491 492 if (mm->brk == brk) 493 return mm->brk; 494 495 /* 496 * Always allow shrinking brk 497 */ 498 if (brk <= mm->brk) { 499 mm->brk = brk; 500 return brk; 501 } 502 503 /* 504 * Ok, looks good - let it rip. 505 */ 506 flush_icache_range(mm->brk, brk); 507 return mm->brk = brk; 508} 509 510/* 511 * initialise the VMA and region record slabs 512 */ 513void __init mmap_init(void) 514{ 515 int ret; 516 517 ret = percpu_counter_init(&vm_committed_as, 0); 518 VM_BUG_ON(ret); 519 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 520} 521 522/* 523 * validate the region tree 524 * - the caller must hold the region lock 525 */ 526#ifdef CONFIG_DEBUG_NOMMU_REGIONS 527static noinline void validate_nommu_regions(void) 528{ 529 struct vm_region *region, *last; 530 struct rb_node *p, *lastp; 531 532 lastp = rb_first(&nommu_region_tree); 533 if (!lastp) 534 return; 535 536 last = rb_entry(lastp, struct vm_region, vm_rb); 537 BUG_ON(unlikely(last->vm_end <= last->vm_start)); 538 BUG_ON(unlikely(last->vm_top < last->vm_end)); 539 540 while ((p = rb_next(lastp))) { 541 region = rb_entry(p, struct vm_region, vm_rb); 542 last = rb_entry(lastp, struct vm_region, vm_rb); 543 544 BUG_ON(unlikely(region->vm_end <= region->vm_start)); 545 BUG_ON(unlikely(region->vm_top < region->vm_end)); 546 BUG_ON(unlikely(region->vm_start < last->vm_top)); 547 548 lastp = p; 549 } 550} 551#else 552static void validate_nommu_regions(void) 553{ 554} 555#endif 556 557/* 558 * add a region into the global tree 559 */ 560static void add_nommu_region(struct vm_region *region) 561{ 562 struct vm_region *pregion; 563 struct rb_node **p, *parent; 564 565 validate_nommu_regions(); 566 567 parent = NULL; 568 p = &nommu_region_tree.rb_node; 569 while (*p) { 570 parent = *p; 571 pregion = rb_entry(parent, struct vm_region, vm_rb); 572 if (region->vm_start < pregion->vm_start) 573 p = &(*p)->rb_left; 574 else if (region->vm_start > pregion->vm_start) 575 p = &(*p)->rb_right; 576 else if (pregion == region) 577 return; 578 else 579 BUG(); 580 } 581 582 rb_link_node(®ion->vm_rb, parent, p); 583 rb_insert_color(®ion->vm_rb, &nommu_region_tree); 584 585 validate_nommu_regions(); 586} 587 588/* 589 * delete a region from the global tree 590 */ 591static void delete_nommu_region(struct vm_region *region) 592{ 593 BUG_ON(!nommu_region_tree.rb_node); 594 595 validate_nommu_regions(); 596 rb_erase(®ion->vm_rb, &nommu_region_tree); 597 validate_nommu_regions(); 598} 599 600/* 601 * free a contiguous series of pages 602 */ 603static void free_page_series(unsigned long from, unsigned long to) 604{ 605 for (; from < to; from += PAGE_SIZE) { 606 struct page *page = virt_to_page(from); 607 608 kdebug("- free %lx", from); 609 atomic_long_dec(&mmap_pages_allocated); 610 if (page_count(page) != 1) 611 kdebug("free page %p: refcount not one: %d", 612 page, page_count(page)); 613 put_page(page); 614 } 615} 616 617/* 618 * release a reference to a region 619 * - the caller must hold the region semaphore for writing, which this releases 620 * - the region may not have been added to the tree yet, in which case vm_top 621 * will equal vm_start 622 */ 623static void __put_nommu_region(struct vm_region *region) 624 __releases(nommu_region_sem) 625{ 626 kenter("%p{%d}", region, region->vm_usage); 627 628 BUG_ON(!nommu_region_tree.rb_node); 629 630 if (--region->vm_usage == 0) { 631 if (region->vm_top > region->vm_start) 632 delete_nommu_region(region); 633 up_write(&nommu_region_sem); 634 635 if (region->vm_file) 636 fput(region->vm_file); 637 638 /* IO memory and memory shared directly out of the pagecache 639 * from ramfs/tmpfs mustn't be released here */ 640 if (region->vm_flags & VM_MAPPED_COPY) { 641 kdebug("free series"); 642 free_page_series(region->vm_start, region->vm_top); 643 } 644 kmem_cache_free(vm_region_jar, region); 645 } else { 646 up_write(&nommu_region_sem); 647 } 648} 649 650/* 651 * release a reference to a region 652 */ 653static void put_nommu_region(struct vm_region *region) 654{ 655 down_write(&nommu_region_sem); 656 __put_nommu_region(region); 657} 658 659/* 660 * update protection on a vma 661 */ 662static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 663{ 664#ifdef CONFIG_MPU 665 struct mm_struct *mm = vma->vm_mm; 666 long start = vma->vm_start & PAGE_MASK; 667 while (start < vma->vm_end) { 668 protect_page(mm, start, flags); 669 start += PAGE_SIZE; 670 } 671 update_protections(mm); 672#endif 673} 674 675/* 676 * add a VMA into a process's mm_struct in the appropriate place in the list 677 * and tree and add to the address space's page tree also if not an anonymous 678 * page 679 * - should be called with mm->mmap_sem held writelocked 680 */ 681static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 682{ 683 struct vm_area_struct *pvma, *prev; 684 struct address_space *mapping; 685 struct rb_node **p, *parent, *rb_prev; 686 687 kenter(",%p", vma); 688 689 BUG_ON(!vma->vm_region); 690 691 mm->map_count++; 692 vma->vm_mm = mm; 693 694 protect_vma(vma, vma->vm_flags); 695 696 /* add the VMA to the mapping */ 697 if (vma->vm_file) { 698 mapping = vma->vm_file->f_mapping; 699 700 flush_dcache_mmap_lock(mapping); 701 vma_prio_tree_insert(vma, &mapping->i_mmap); 702 flush_dcache_mmap_unlock(mapping); 703 } 704 705 /* add the VMA to the tree */ 706 parent = rb_prev = NULL; 707 p = &mm->mm_rb.rb_node; 708 while (*p) { 709 parent = *p; 710 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 711 712 /* sort by: start addr, end addr, VMA struct addr in that order 713 * (the latter is necessary as we may get identical VMAs) */ 714 if (vma->vm_start < pvma->vm_start) 715 p = &(*p)->rb_left; 716 else if (vma->vm_start > pvma->vm_start) { 717 rb_prev = parent; 718 p = &(*p)->rb_right; 719 } else if (vma->vm_end < pvma->vm_end) 720 p = &(*p)->rb_left; 721 else if (vma->vm_end > pvma->vm_end) { 722 rb_prev = parent; 723 p = &(*p)->rb_right; 724 } else if (vma < pvma) 725 p = &(*p)->rb_left; 726 else if (vma > pvma) { 727 rb_prev = parent; 728 p = &(*p)->rb_right; 729 } else 730 BUG(); 731 } 732 733 rb_link_node(&vma->vm_rb, parent, p); 734 rb_insert_color(&vma->vm_rb, &mm->mm_rb); 735 736 /* add VMA to the VMA list also */ 737 prev = NULL; 738 if (rb_prev) 739 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 740 741 __vma_link_list(mm, vma, prev, parent); 742} 743 744/* 745 * delete a VMA from its owning mm_struct and address space 746 */ 747static void delete_vma_from_mm(struct vm_area_struct *vma) 748{ 749 struct address_space *mapping; 750 struct mm_struct *mm = vma->vm_mm; 751 752 kenter("%p", vma); 753 754 protect_vma(vma, 0); 755 756 mm->map_count--; 757 if (mm->mmap_cache == vma) 758 mm->mmap_cache = NULL; 759 760 /* remove the VMA from the mapping */ 761 if (vma->vm_file) { 762 mapping = vma->vm_file->f_mapping; 763 764 flush_dcache_mmap_lock(mapping); 765 vma_prio_tree_remove(vma, &mapping->i_mmap); 766 flush_dcache_mmap_unlock(mapping); 767 } 768 769 /* remove from the MM's tree and list */ 770 rb_erase(&vma->vm_rb, &mm->mm_rb); 771 772 if (vma->vm_prev) 773 vma->vm_prev->vm_next = vma->vm_next; 774 else 775 mm->mmap = vma->vm_next; 776 777 if (vma->vm_next) 778 vma->vm_next->vm_prev = vma->vm_prev; 779 780 vma->vm_mm = NULL; 781} 782 783/* 784 * destroy a VMA record 785 */ 786static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 787{ 788 kenter("%p", vma); 789 if (vma->vm_ops && vma->vm_ops->close) 790 vma->vm_ops->close(vma); 791 if (vma->vm_file) { 792 fput(vma->vm_file); 793 if (vma->vm_flags & VM_EXECUTABLE) 794 removed_exe_file_vma(mm); 795 } 796 put_nommu_region(vma->vm_region); 797 kmem_cache_free(vm_area_cachep, vma); 798} 799 800/* 801 * look up the first VMA in which addr resides, NULL if none 802 * - should be called with mm->mmap_sem at least held readlocked 803 */ 804struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 805{ 806 struct vm_area_struct *vma; 807 struct rb_node *n = mm->mm_rb.rb_node; 808 809 /* check the cache first */ 810 vma = mm->mmap_cache; 811 if (vma && vma->vm_start <= addr && vma->vm_end > addr) 812 return vma; 813 814 /* trawl the tree (there may be multiple mappings in which addr 815 * resides) */ 816 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { 817 vma = rb_entry(n, struct vm_area_struct, vm_rb); 818 if (vma->vm_start > addr) 819 return NULL; 820 if (vma->vm_end > addr) { 821 mm->mmap_cache = vma; 822 return vma; 823 } 824 } 825 826 return NULL; 827} 828EXPORT_SYMBOL(find_vma); 829 830/* 831 * find a VMA 832 * - we don't extend stack VMAs under NOMMU conditions 833 */ 834struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 835{ 836 return find_vma(mm, addr); 837} 838 839/* 840 * expand a stack to a given address 841 * - not supported under NOMMU conditions 842 */ 843int expand_stack(struct vm_area_struct *vma, unsigned long address) 844{ 845 return -ENOMEM; 846} 847 848/* 849 * look up the first VMA exactly that exactly matches addr 850 * - should be called with mm->mmap_sem at least held readlocked 851 */ 852static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 853 unsigned long addr, 854 unsigned long len) 855{ 856 struct vm_area_struct *vma; 857 struct rb_node *n = mm->mm_rb.rb_node; 858 unsigned long end = addr + len; 859 860 /* check the cache first */ 861 vma = mm->mmap_cache; 862 if (vma && vma->vm_start == addr && vma->vm_end == end) 863 return vma; 864 865 /* trawl the tree (there may be multiple mappings in which addr 866 * resides) */ 867 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { 868 vma = rb_entry(n, struct vm_area_struct, vm_rb); 869 if (vma->vm_start < addr) 870 continue; 871 if (vma->vm_start > addr) 872 return NULL; 873 if (vma->vm_end == end) { 874 mm->mmap_cache = vma; 875 return vma; 876 } 877 } 878 879 return NULL; 880} 881 882/* 883 * determine whether a mapping should be permitted and, if so, what sort of 884 * mapping we're capable of supporting 885 */ 886static int validate_mmap_request(struct file *file, 887 unsigned long addr, 888 unsigned long len, 889 unsigned long prot, 890 unsigned long flags, 891 unsigned long pgoff, 892 unsigned long *_capabilities) 893{ 894 unsigned long capabilities, rlen; 895 unsigned long reqprot = prot; 896 int ret; 897 898 /* do the simple checks first */ 899 if (flags & MAP_FIXED) { 900 printk(KERN_DEBUG 901 "%d: Can't do fixed-address/overlay mmap of RAM\n", 902 current->pid); 903 return -EINVAL; 904 } 905 906 if ((flags & MAP_TYPE) != MAP_PRIVATE && 907 (flags & MAP_TYPE) != MAP_SHARED) 908 return -EINVAL; 909 910 if (!len) 911 return -EINVAL; 912 913 /* Careful about overflows.. */ 914 rlen = PAGE_ALIGN(len); 915 if (!rlen || rlen > TASK_SIZE) 916 return -ENOMEM; 917 918 /* offset overflow? */ 919 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 920 return -EOVERFLOW; 921 922 if (file) { 923 /* validate file mapping requests */ 924 struct address_space *mapping; 925 926 /* files must support mmap */ 927 if (!file->f_op || !file->f_op->mmap) 928 return -ENODEV; 929 930 /* work out if what we've got could possibly be shared 931 * - we support chardevs that provide their own "memory" 932 * - we support files/blockdevs that are memory backed 933 */ 934 mapping = file->f_mapping; 935 if (!mapping) 936 mapping = file->f_path.dentry->d_inode->i_mapping; 937 938 capabilities = 0; 939 if (mapping && mapping->backing_dev_info) 940 capabilities = mapping->backing_dev_info->capabilities; 941 942 if (!capabilities) { 943 /* no explicit capabilities set, so assume some 944 * defaults */ 945 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 946 case S_IFREG: 947 case S_IFBLK: 948 capabilities = BDI_CAP_MAP_COPY; 949 break; 950 951 case S_IFCHR: 952 capabilities = 953 BDI_CAP_MAP_DIRECT | 954 BDI_CAP_READ_MAP | 955 BDI_CAP_WRITE_MAP; 956 break; 957 958 default: 959 return -EINVAL; 960 } 961 } 962 963 /* eliminate any capabilities that we can't support on this 964 * device */ 965 if (!file->f_op->get_unmapped_area) 966 capabilities &= ~BDI_CAP_MAP_DIRECT; 967 if (!file->f_op->read) 968 capabilities &= ~BDI_CAP_MAP_COPY; 969 970 /* The file shall have been opened with read permission. */ 971 if (!(file->f_mode & FMODE_READ)) 972 return -EACCES; 973 974 if (flags & MAP_SHARED) { 975 /* do checks for writing, appending and locking */ 976 if ((prot & PROT_WRITE) && 977 !(file->f_mode & FMODE_WRITE)) 978 return -EACCES; 979 980 if (IS_APPEND(file->f_path.dentry->d_inode) && 981 (file->f_mode & FMODE_WRITE)) 982 return -EACCES; 983 984 if (locks_verify_locked(file->f_path.dentry->d_inode)) 985 return -EAGAIN; 986 987 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 988 return -ENODEV; 989 990 /* we mustn't privatise shared mappings */ 991 capabilities &= ~BDI_CAP_MAP_COPY; 992 } 993 else { 994 /* we're going to read the file into private memory we 995 * allocate */ 996 if (!(capabilities & BDI_CAP_MAP_COPY)) 997 return -ENODEV; 998 999 /* we don't permit a private writable mapping to be 1000 * shared with the backing device */ 1001 if (prot & PROT_WRITE) 1002 capabilities &= ~BDI_CAP_MAP_DIRECT; 1003 } 1004 1005 if (capabilities & BDI_CAP_MAP_DIRECT) { 1006 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 1007 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 1008 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 1009 ) { 1010 capabilities &= ~BDI_CAP_MAP_DIRECT; 1011 if (flags & MAP_SHARED) { 1012 printk(KERN_WARNING 1013 "MAP_SHARED not completely supported on !MMU\n"); 1014 return -EINVAL; 1015 } 1016 } 1017 } 1018 1019 /* handle executable mappings and implied executable 1020 * mappings */ 1021 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 1022 if (prot & PROT_EXEC) 1023 return -EPERM; 1024 } 1025 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 1026 /* handle implication of PROT_EXEC by PROT_READ */ 1027 if (current->personality & READ_IMPLIES_EXEC) { 1028 if (capabilities & BDI_CAP_EXEC_MAP) 1029 prot |= PROT_EXEC; 1030 } 1031 } 1032 else if ((prot & PROT_READ) && 1033 (prot & PROT_EXEC) && 1034 !(capabilities & BDI_CAP_EXEC_MAP) 1035 ) { 1036 /* backing file is not executable, try to copy */ 1037 capabilities &= ~BDI_CAP_MAP_DIRECT; 1038 } 1039 } 1040 else { 1041 /* anonymous mappings are always memory backed and can be 1042 * privately mapped 1043 */ 1044 capabilities = BDI_CAP_MAP_COPY; 1045 1046 /* handle PROT_EXEC implication by PROT_READ */ 1047 if ((prot & PROT_READ) && 1048 (current->personality & READ_IMPLIES_EXEC)) 1049 prot |= PROT_EXEC; 1050 } 1051 1052 /* allow the security API to have its say */ 1053 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 1054 if (ret < 0) 1055 return ret; 1056 1057 /* looks okay */ 1058 *_capabilities = capabilities; 1059 return 0; 1060} 1061 1062/* 1063 * we've determined that we can make the mapping, now translate what we 1064 * now know into VMA flags 1065 */ 1066static unsigned long determine_vm_flags(struct file *file, 1067 unsigned long prot, 1068 unsigned long flags, 1069 unsigned long capabilities) 1070{ 1071 unsigned long vm_flags; 1072 1073 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 1074 /* vm_flags |= mm->def_flags; */ 1075 1076 if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 1077 /* attempt to share read-only copies of mapped file chunks */ 1078 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 1079 if (file && !(prot & PROT_WRITE)) 1080 vm_flags |= VM_MAYSHARE; 1081 } else { 1082 /* overlay a shareable mapping on the backing device or inode 1083 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 1084 * romfs/cramfs */ 1085 vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); 1086 if (flags & MAP_SHARED) 1087 vm_flags |= VM_SHARED; 1088 } 1089 1090 /* refuse to let anyone share private mappings with this process if 1091 * it's being traced - otherwise breakpoints set in it may interfere 1092 * with another untraced process 1093 */ 1094 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) 1095 vm_flags &= ~VM_MAYSHARE; 1096 1097 return vm_flags; 1098} 1099 1100/* 1101 * set up a shared mapping on a file (the driver or filesystem provides and 1102 * pins the storage) 1103 */ 1104static int do_mmap_shared_file(struct vm_area_struct *vma) 1105{ 1106 int ret; 1107 1108 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1109 if (ret == 0) { 1110 vma->vm_region->vm_top = vma->vm_region->vm_end; 1111 return 0; 1112 } 1113 if (ret != -ENOSYS) 1114 return ret; 1115 1116 /* getting -ENOSYS indicates that direct mmap isn't possible (as 1117 * opposed to tried but failed) so we can only give a suitable error as 1118 * it's not possible to make a private copy if MAP_SHARED was given */ 1119 return -ENODEV; 1120} 1121 1122/* 1123 * set up a private mapping or an anonymous shared mapping 1124 */ 1125static int do_mmap_private(struct vm_area_struct *vma, 1126 struct vm_region *region, 1127 unsigned long len, 1128 unsigned long capabilities) 1129{ 1130 struct page *pages; 1131 unsigned long total, point, n, rlen; 1132 void *base; 1133 int ret, order; 1134 1135 /* invoke the file's mapping function so that it can keep track of 1136 * shared mappings on devices or memory 1137 * - VM_MAYSHARE will be set if it may attempt to share 1138 */ 1139 if (capabilities & BDI_CAP_MAP_DIRECT) { 1140 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1141 if (ret == 0) { 1142 /* shouldn't return success if we're not sharing */ 1143 BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1144 vma->vm_region->vm_top = vma->vm_region->vm_end; 1145 return 0; 1146 } 1147 if (ret != -ENOSYS) 1148 return ret; 1149 1150 /* getting an ENOSYS error indicates that direct mmap isn't 1151 * possible (as opposed to tried but failed) so we'll try to 1152 * make a private copy of the data and map that instead */ 1153 } 1154 1155 rlen = PAGE_ALIGN(len); 1156 1157 /* allocate some memory to hold the mapping 1158 * - note that this may not return a page-aligned address if the object 1159 * we're allocating is smaller than a page 1160 */ 1161 order = get_order(rlen); 1162 kdebug("alloc order %d for %lx", order, len); 1163 1164 pages = alloc_pages(GFP_KERNEL, order); 1165 if (!pages) 1166 goto enomem; 1167 1168 total = 1 << order; 1169 atomic_long_add(total, &mmap_pages_allocated); 1170 1171 point = rlen >> PAGE_SHIFT; 1172 1173 /* we allocated a power-of-2 sized page set, so we may want to trim off 1174 * the excess */ 1175 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 1176 while (total > point) { 1177 order = ilog2(total - point); 1178 n = 1 << order; 1179 kdebug("shave %lu/%lu @%lu", n, total - point, total); 1180 atomic_long_sub(n, &mmap_pages_allocated); 1181 total -= n; 1182 set_page_refcounted(pages + total); 1183 __free_pages(pages + total, order); 1184 } 1185 } 1186 1187 for (point = 1; point < total; point++) 1188 set_page_refcounted(&pages[point]); 1189 1190 base = page_address(pages); 1191 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 1192 region->vm_start = (unsigned long) base; 1193 region->vm_end = region->vm_start + rlen; 1194 region->vm_top = region->vm_start + (total << PAGE_SHIFT); 1195 1196 vma->vm_start = region->vm_start; 1197 vma->vm_end = region->vm_start + len; 1198 1199 if (vma->vm_file) { 1200 /* read the contents of a file into the copy */ 1201 mm_segment_t old_fs; 1202 loff_t fpos; 1203 1204 fpos = vma->vm_pgoff; 1205 fpos <<= PAGE_SHIFT; 1206 1207 old_fs = get_fs(); 1208 set_fs(KERNEL_DS); 1209 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); 1210 set_fs(old_fs); 1211 1212 if (ret < 0) 1213 goto error_free; 1214 1215 /* clear the last little bit */ 1216 if (ret < rlen) 1217 memset(base + ret, 0, rlen - ret); 1218 1219 } 1220 1221 return 0; 1222 1223error_free: 1224 free_page_series(region->vm_start, region->vm_end); 1225 region->vm_start = vma->vm_start = 0; 1226 region->vm_end = vma->vm_end = 0; 1227 region->vm_top = 0; 1228 return ret; 1229 1230enomem: 1231 printk("Allocation of length %lu from process %d (%s) failed\n", 1232 len, current->pid, current->comm); 1233 show_free_areas(0); 1234 return -ENOMEM; 1235} 1236 1237/* 1238 * handle mapping creation for uClinux 1239 */ 1240unsigned long do_mmap_pgoff(struct file *file, 1241 unsigned long addr, 1242 unsigned long len, 1243 unsigned long prot, 1244 unsigned long flags, 1245 unsigned long pgoff) 1246{ 1247 struct vm_area_struct *vma; 1248 struct vm_region *region; 1249 struct rb_node *rb; 1250 unsigned long capabilities, vm_flags, result; 1251 int ret; 1252 1253 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 1254 1255 /* decide whether we should attempt the mapping, and if so what sort of 1256 * mapping */ 1257 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 1258 &capabilities); 1259 if (ret < 0) { 1260 kleave(" = %d [val]", ret); 1261 return ret; 1262 } 1263 1264 /* we ignore the address hint */ 1265 addr = 0; 1266 1267 /* we've determined that we can make the mapping, now translate what we 1268 * now know into VMA flags */ 1269 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 1270 1271 /* we're going to need to record the mapping */ 1272 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 1273 if (!region) 1274 goto error_getting_region; 1275 1276 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 1277 if (!vma) 1278 goto error_getting_vma; 1279 1280 region->vm_usage = 1; 1281 region->vm_flags = vm_flags; 1282 region->vm_pgoff = pgoff; 1283 1284 INIT_LIST_HEAD(&vma->anon_vma_chain); 1285 vma->vm_flags = vm_flags; 1286 vma->vm_pgoff = pgoff; 1287 1288 if (file) { 1289 region->vm_file = file; 1290 get_file(file); 1291 vma->vm_file = file; 1292 get_file(file); 1293 if (vm_flags & VM_EXECUTABLE) { 1294 added_exe_file_vma(current->mm); 1295 vma->vm_mm = current->mm; 1296 } 1297 } 1298 1299 down_write(&nommu_region_sem); 1300 1301 /* if we want to share, we need to check for regions created by other 1302 * mmap() calls that overlap with our proposed mapping 1303 * - we can only share with a superset match on most regular files 1304 * - shared mappings on character devices and memory backed files are 1305 * permitted to overlap inexactly as far as we are concerned for in 1306 * these cases, sharing is handled in the driver or filesystem rather 1307 * than here 1308 */ 1309 if (vm_flags & VM_MAYSHARE) { 1310 struct vm_region *pregion; 1311 unsigned long pglen, rpglen, pgend, rpgend, start; 1312 1313 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1314 pgend = pgoff + pglen; 1315 1316 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 1317 pregion = rb_entry(rb, struct vm_region, vm_rb); 1318 1319 if (!(pregion->vm_flags & VM_MAYSHARE)) 1320 continue; 1321 1322 /* search for overlapping mappings on the same file */ 1323 if (pregion->vm_file->f_path.dentry->d_inode != 1324 file->f_path.dentry->d_inode) 1325 continue; 1326 1327 if (pregion->vm_pgoff >= pgend) 1328 continue; 1329 1330 rpglen = pregion->vm_end - pregion->vm_start; 1331 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 1332 rpgend = pregion->vm_pgoff + rpglen; 1333 if (pgoff >= rpgend) 1334 continue; 1335 1336 /* handle inexactly overlapping matches between 1337 * mappings */ 1338 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 1339 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 1340 /* new mapping is not a subset of the region */ 1341 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 1342 goto sharing_violation; 1343 continue; 1344 } 1345 1346 /* we've found a region we can share */ 1347 pregion->vm_usage++; 1348 vma->vm_region = pregion; 1349 start = pregion->vm_start; 1350 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 1351 vma->vm_start = start; 1352 vma->vm_end = start + len; 1353 1354 if (pregion->vm_flags & VM_MAPPED_COPY) { 1355 kdebug("share copy"); 1356 vma->vm_flags |= VM_MAPPED_COPY; 1357 } else { 1358 kdebug("share mmap"); 1359 ret = do_mmap_shared_file(vma); 1360 if (ret < 0) { 1361 vma->vm_region = NULL; 1362 vma->vm_start = 0; 1363 vma->vm_end = 0; 1364 pregion->vm_usage--; 1365 pregion = NULL; 1366 goto error_just_free; 1367 } 1368 } 1369 fput(region->vm_file); 1370 kmem_cache_free(vm_region_jar, region); 1371 region = pregion; 1372 result = start; 1373 goto share; 1374 } 1375 1376 /* obtain the address at which to make a shared mapping 1377 * - this is the hook for quasi-memory character devices to 1378 * tell us the location of a shared mapping 1379 */ 1380 if (capabilities & BDI_CAP_MAP_DIRECT) { 1381 addr = file->f_op->get_unmapped_area(file, addr, len, 1382 pgoff, flags); 1383 if (IS_ERR((void *) addr)) { 1384 ret = addr; 1385 if (ret != (unsigned long) -ENOSYS) 1386 goto error_just_free; 1387 1388 /* the driver refused to tell us where to site 1389 * the mapping so we'll have to attempt to copy 1390 * it */ 1391 ret = (unsigned long) -ENODEV; 1392 if (!(capabilities & BDI_CAP_MAP_COPY)) 1393 goto error_just_free; 1394 1395 capabilities &= ~BDI_CAP_MAP_DIRECT; 1396 } else { 1397 vma->vm_start = region->vm_start = addr; 1398 vma->vm_end = region->vm_end = addr + len; 1399 } 1400 } 1401 } 1402 1403 vma->vm_region = region; 1404 1405 /* set up the mapping 1406 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set 1407 */ 1408 if (file && vma->vm_flags & VM_SHARED) 1409 ret = do_mmap_shared_file(vma); 1410 else 1411 ret = do_mmap_private(vma, region, len, capabilities); 1412 if (ret < 0) 1413 goto error_just_free; 1414 add_nommu_region(region); 1415 1416 /* clear anonymous mappings that don't ask for uninitialized data */ 1417 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1418 memset((void *)region->vm_start, 0, 1419 region->vm_end - region->vm_start); 1420 1421 /* okay... we have a mapping; now we have to register it */ 1422 result = vma->vm_start; 1423 1424 current->mm->total_vm += len >> PAGE_SHIFT; 1425 1426share: 1427 add_vma_to_mm(current->mm, vma); 1428 1429 /* we flush the region from the icache only when the first executable 1430 * mapping of it is made */ 1431 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1432 flush_icache_range(region->vm_start, region->vm_end); 1433 region->vm_icache_flushed = true; 1434 } 1435 1436 up_write(&nommu_region_sem); 1437 1438 kleave(" = %lx", result); 1439 return result; 1440 1441error_just_free: 1442 up_write(&nommu_region_sem); 1443error: 1444 if (region->vm_file) 1445 fput(region->vm_file); 1446 kmem_cache_free(vm_region_jar, region); 1447 if (vma->vm_file) 1448 fput(vma->vm_file); 1449 if (vma->vm_flags & VM_EXECUTABLE) 1450 removed_exe_file_vma(vma->vm_mm); 1451 kmem_cache_free(vm_area_cachep, vma); 1452 kleave(" = %d", ret); 1453 return ret; 1454 1455sharing_violation: 1456 up_write(&nommu_region_sem); 1457 printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 1458 ret = -EINVAL; 1459 goto error; 1460 1461error_getting_vma: 1462 kmem_cache_free(vm_region_jar, region); 1463 printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 1464 " from process %d failed\n", 1465 len, current->pid); 1466 show_free_areas(0); 1467 return -ENOMEM; 1468 1469error_getting_region: 1470 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 1471 " from process %d failed\n", 1472 len, current->pid); 1473 show_free_areas(0); 1474 return -ENOMEM; 1475} 1476EXPORT_SYMBOL(do_mmap_pgoff); 1477 1478SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1479 unsigned long, prot, unsigned long, flags, 1480 unsigned long, fd, unsigned long, pgoff) 1481{ 1482 struct file *file = NULL; 1483 unsigned long retval = -EBADF; 1484 1485 audit_mmap_fd(fd, flags); 1486 if (!(flags & MAP_ANONYMOUS)) { 1487 file = fget(fd); 1488 if (!file) 1489 goto out; 1490 } 1491 1492 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 1493 1494 down_write(¤t->mm->mmap_sem); 1495 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1496 up_write(¤t->mm->mmap_sem); 1497 1498 if (file) 1499 fput(file); 1500out: 1501 return retval; 1502} 1503 1504#ifdef __ARCH_WANT_SYS_OLD_MMAP 1505struct mmap_arg_struct { 1506 unsigned long addr; 1507 unsigned long len; 1508 unsigned long prot; 1509 unsigned long flags; 1510 unsigned long fd; 1511 unsigned long offset; 1512}; 1513 1514SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1515{ 1516 struct mmap_arg_struct a; 1517 1518 if (copy_from_user(&a, arg, sizeof(a))) 1519 return -EFAULT; 1520 if (a.offset & ~PAGE_MASK) 1521 return -EINVAL; 1522 1523 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1524 a.offset >> PAGE_SHIFT); 1525} 1526#endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1527 1528/* 1529 * split a vma into two pieces at address 'addr', a new vma is allocated either 1530 * for the first part or the tail. 1531 */ 1532int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 1533 unsigned long addr, int new_below) 1534{ 1535 struct vm_area_struct *new; 1536 struct vm_region *region; 1537 unsigned long npages; 1538 1539 kenter(""); 1540 1541 /* we're only permitted to split anonymous regions (these should have 1542 * only a single usage on the region) */ 1543 if (vma->vm_file) 1544 return -ENOMEM; 1545 1546 if (mm->map_count >= sysctl_max_map_count) 1547 return -ENOMEM; 1548 1549 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 1550 if (!region) 1551 return -ENOMEM; 1552 1553 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 1554 if (!new) { 1555 kmem_cache_free(vm_region_jar, region); 1556 return -ENOMEM; 1557 } 1558 1559 /* most fields are the same, copy all, and then fixup */ 1560 *new = *vma; 1561 *region = *vma->vm_region; 1562 new->vm_region = region; 1563 1564 npages = (addr - vma->vm_start) >> PAGE_SHIFT; 1565 1566 if (new_below) { 1567 region->vm_top = region->vm_end = new->vm_end = addr; 1568 } else { 1569 region->vm_start = new->vm_start = addr; 1570 region->vm_pgoff = new->vm_pgoff += npages; 1571 } 1572 1573 if (new->vm_ops && new->vm_ops->open) 1574 new->vm_ops->open(new); 1575 1576 delete_vma_from_mm(vma); 1577 down_write(&nommu_region_sem); 1578 delete_nommu_region(vma->vm_region); 1579 if (new_below) { 1580 vma->vm_region->vm_start = vma->vm_start = addr; 1581 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 1582 } else { 1583 vma->vm_region->vm_end = vma->vm_end = addr; 1584 vma->vm_region->vm_top = addr; 1585 } 1586 add_nommu_region(vma->vm_region); 1587 add_nommu_region(new->vm_region); 1588 up_write(&nommu_region_sem); 1589 add_vma_to_mm(mm, vma); 1590 add_vma_to_mm(mm, new); 1591 return 0; 1592} 1593 1594/* 1595 * shrink a VMA by removing the specified chunk from either the beginning or 1596 * the end 1597 */ 1598static int shrink_vma(struct mm_struct *mm, 1599 struct vm_area_struct *vma, 1600 unsigned long from, unsigned long to) 1601{ 1602 struct vm_region *region; 1603 1604 kenter(""); 1605 1606 /* adjust the VMA's pointers, which may reposition it in the MM's tree 1607 * and list */ 1608 delete_vma_from_mm(vma); 1609 if (from > vma->vm_start) 1610 vma->vm_end = from; 1611 else 1612 vma->vm_start = to; 1613 add_vma_to_mm(mm, vma); 1614 1615 /* cut the backing region down to size */ 1616 region = vma->vm_region; 1617 BUG_ON(region->vm_usage != 1); 1618 1619 down_write(&nommu_region_sem); 1620 delete_nommu_region(region); 1621 if (from > region->vm_start) { 1622 to = region->vm_top; 1623 region->vm_top = region->vm_end = from; 1624 } else { 1625 region->vm_start = to; 1626 } 1627 add_nommu_region(region); 1628 up_write(&nommu_region_sem); 1629 1630 free_page_series(from, to); 1631 return 0; 1632} 1633 1634/* 1635 * release a mapping 1636 * - under NOMMU conditions the chunk to be unmapped must be backed by a single 1637 * VMA, though it need not cover the whole VMA 1638 */ 1639int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 1640{ 1641 struct vm_area_struct *vma; 1642 struct rb_node *rb; 1643 unsigned long end = start + len; 1644 int ret; 1645 1646 kenter(",%lx,%zx", start, len); 1647 1648 if (len == 0) 1649 return -EINVAL; 1650 1651 /* find the first potentially overlapping VMA */ 1652 vma = find_vma(mm, start); 1653 if (!vma) { 1654 static int limit = 0; 1655 if (limit < 5) { 1656 printk(KERN_WARNING 1657 "munmap of memory not mmapped by process %d" 1658 " (%s): 0x%lx-0x%lx\n", 1659 current->pid, current->comm, 1660 start, start + len - 1); 1661 limit++; 1662 } 1663 return -EINVAL; 1664 } 1665 1666 /* we're allowed to split an anonymous VMA but not a file-backed one */ 1667 if (vma->vm_file) { 1668 do { 1669 if (start > vma->vm_start) { 1670 kleave(" = -EINVAL [miss]"); 1671 return -EINVAL; 1672 } 1673 if (end == vma->vm_end) 1674 goto erase_whole_vma; 1675 rb = rb_next(&vma->vm_rb); 1676 vma = rb_entry(rb, struct vm_area_struct, vm_rb); 1677 } while (rb); 1678 kleave(" = -EINVAL [split file]"); 1679 return -EINVAL; 1680 } else { 1681 /* the chunk must be a subset of the VMA found */ 1682 if (start == vma->vm_start && end == vma->vm_end) 1683 goto erase_whole_vma; 1684 if (start < vma->vm_start || end > vma->vm_end) { 1685 kleave(" = -EINVAL [superset]"); 1686 return -EINVAL; 1687 } 1688 if (start & ~PAGE_MASK) { 1689 kleave(" = -EINVAL [unaligned start]"); 1690 return -EINVAL; 1691 } 1692 if (end != vma->vm_end && end & ~PAGE_MASK) { 1693 kleave(" = -EINVAL [unaligned split]"); 1694 return -EINVAL; 1695 } 1696 if (start != vma->vm_start && end != vma->vm_end) { 1697 ret = split_vma(mm, vma, start, 1); 1698 if (ret < 0) { 1699 kleave(" = %d [split]", ret); 1700 return ret; 1701 } 1702 } 1703 return shrink_vma(mm, vma, start, end); 1704 } 1705 1706erase_whole_vma: 1707 delete_vma_from_mm(vma); 1708 delete_vma(mm, vma); 1709 kleave(" = 0"); 1710 return 0; 1711} 1712EXPORT_SYMBOL(do_munmap); 1713 1714SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1715{ 1716 int ret; 1717 struct mm_struct *mm = current->mm; 1718 1719 down_write(&mm->mmap_sem); 1720 ret = do_munmap(mm, addr, len); 1721 up_write(&mm->mmap_sem); 1722 return ret; 1723} 1724 1725/* 1726 * release all the mappings made in a process's VM space 1727 */ 1728void exit_mmap(struct mm_struct *mm) 1729{ 1730 struct vm_area_struct *vma; 1731 1732 if (!mm) 1733 return; 1734 1735 kenter(""); 1736 1737 mm->total_vm = 0; 1738 1739 while ((vma = mm->mmap)) { 1740 mm->mmap = vma->vm_next; 1741 delete_vma_from_mm(vma); 1742 delete_vma(mm, vma); 1743 cond_resched(); 1744 } 1745 1746 kleave(""); 1747} 1748 1749unsigned long do_brk(unsigned long addr, unsigned long len) 1750{ 1751 return -ENOMEM; 1752} 1753 1754/* 1755 * expand (or shrink) an existing mapping, potentially moving it at the same 1756 * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 1757 * 1758 * under NOMMU conditions, we only permit changing a mapping's size, and only 1759 * as long as it stays within the region allocated by do_mmap_private() and the 1760 * block is not shareable 1761 * 1762 * MREMAP_FIXED is not supported under NOMMU conditions 1763 */ 1764unsigned long do_mremap(unsigned long addr, 1765 unsigned long old_len, unsigned long new_len, 1766 unsigned long flags, unsigned long new_addr) 1767{ 1768 struct vm_area_struct *vma; 1769 1770 /* insanity checks first */ 1771 if (old_len == 0 || new_len == 0) 1772 return (unsigned long) -EINVAL; 1773 1774 if (addr & ~PAGE_MASK) 1775 return -EINVAL; 1776 1777 if (flags & MREMAP_FIXED && new_addr != addr) 1778 return (unsigned long) -EINVAL; 1779 1780 vma = find_vma_exact(current->mm, addr, old_len); 1781 if (!vma) 1782 return (unsigned long) -EINVAL; 1783 1784 if (vma->vm_end != vma->vm_start + old_len) 1785 return (unsigned long) -EFAULT; 1786 1787 if (vma->vm_flags & VM_MAYSHARE) 1788 return (unsigned long) -EPERM; 1789 1790 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 1791 return (unsigned long) -ENOMEM; 1792 1793 /* all checks complete - do it */ 1794 vma->vm_end = vma->vm_start + new_len; 1795 return vma->vm_start; 1796} 1797EXPORT_SYMBOL(do_mremap); 1798 1799SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 1800 unsigned long, new_len, unsigned long, flags, 1801 unsigned long, new_addr) 1802{ 1803 unsigned long ret; 1804 1805 down_write(¤t->mm->mmap_sem); 1806 ret = do_mremap(addr, old_len, new_len, flags, new_addr); 1807 up_write(¤t->mm->mmap_sem); 1808 return ret; 1809} 1810 1811struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1812 unsigned int foll_flags) 1813{ 1814 return NULL; 1815} 1816 1817int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, 1818 unsigned long to, unsigned long size, pgprot_t prot) 1819{ 1820 vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; 1821 return 0; 1822} 1823EXPORT_SYMBOL(remap_pfn_range); 1824 1825int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1826 unsigned long pgoff) 1827{ 1828 unsigned int size = vma->vm_end - vma->vm_start; 1829 1830 if (!(vma->vm_flags & VM_USERMAP)) 1831 return -EINVAL; 1832 1833 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1834 vma->vm_end = vma->vm_start + size; 1835 1836 return 0; 1837} 1838EXPORT_SYMBOL(remap_vmalloc_range); 1839 1840unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 1841 unsigned long len, unsigned long pgoff, unsigned long flags) 1842{ 1843 return -ENOMEM; 1844} 1845 1846void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 1847{ 1848} 1849 1850void unmap_mapping_range(struct address_space *mapping, 1851 loff_t const holebegin, loff_t const holelen, 1852 int even_cows) 1853{ 1854} 1855EXPORT_SYMBOL(unmap_mapping_range); 1856 1857/* 1858 * Check that a process has enough memory to allocate a new virtual 1859 * mapping. 0 means there is enough memory for the allocation to 1860 * succeed and -ENOMEM implies there is not. 1861 * 1862 * We currently support three overcommit policies, which are set via the 1863 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1864 * 1865 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1866 * Additional code 2002 Jul 20 by Robert Love. 1867 * 1868 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1869 * 1870 * Note this is a helper function intended to be used by LSMs which 1871 * wish to use this logic. 1872 */ 1873int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1874{ 1875 unsigned long free, allowed; 1876 1877 vm_acct_memory(pages); 1878 1879 /* 1880 * Sometimes we want to use more memory than we have 1881 */ 1882 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1883 return 0; 1884 1885 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1886 unsigned long n; 1887 1888 free = global_page_state(NR_FILE_PAGES); 1889 free += nr_swap_pages; 1890 1891 /* 1892 * Any slabs which are created with the 1893 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1894 * which are reclaimable, under pressure. The dentry 1895 * cache and most inode caches should fall into this 1896 */ 1897 free += global_page_state(NR_SLAB_RECLAIMABLE); 1898 1899 /* 1900 * Leave the last 3% for root 1901 */ 1902 if (!cap_sys_admin) 1903 free -= free / 32; 1904 1905 if (free > pages) 1906 return 0; 1907 1908 /* 1909 * nr_free_pages() is very expensive on large systems, 1910 * only call if we're about to fail. 1911 */ 1912 n = nr_free_pages(); 1913 1914 /* 1915 * Leave reserved pages. The pages are not for anonymous pages. 1916 */ 1917 if (n <= totalreserve_pages) 1918 goto error; 1919 else 1920 n -= totalreserve_pages; 1921 1922 /* 1923 * Leave the last 3% for root 1924 */ 1925 if (!cap_sys_admin) 1926 n -= n / 32; 1927 free += n; 1928 1929 if (free > pages) 1930 return 0; 1931 1932 goto error; 1933 } 1934 1935 allowed = totalram_pages * sysctl_overcommit_ratio / 100; 1936 /* 1937 * Leave the last 3% for root 1938 */ 1939 if (!cap_sys_admin) 1940 allowed -= allowed / 32; 1941 allowed += total_swap_pages; 1942 1943 /* Don't let a single process grow too big: 1944 leave 3% of the size of this process for other processes */ 1945 if (mm) 1946 allowed -= mm->total_vm / 32; 1947 1948 if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1949 return 0; 1950 1951error: 1952 vm_unacct_memory(pages); 1953 1954 return -ENOMEM; 1955} 1956 1957int in_gate_area_no_mm(unsigned long addr) 1958{ 1959 return 0; 1960} 1961 1962int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1963{ 1964 BUG(); 1965 return 0; 1966} 1967EXPORT_SYMBOL(filemap_fault); 1968 1969static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 1970 unsigned long addr, void *buf, int len, int write) 1971{ 1972 struct vm_area_struct *vma; 1973 1974 down_read(&mm->mmap_sem); 1975 1976 /* the access must start within one of the target process's mappings */ 1977 vma = find_vma(mm, addr); 1978 if (vma) { 1979 /* don't overrun this mapping */ 1980 if (addr + len >= vma->vm_end) 1981 len = vma->vm_end - addr; 1982 1983 /* only read or write mappings where it is permitted */ 1984 if (write && vma->vm_flags & VM_MAYWRITE) 1985 copy_to_user_page(vma, NULL, addr, 1986 (void *) addr, buf, len); 1987 else if (!write && vma->vm_flags & VM_MAYREAD) 1988 copy_from_user_page(vma, NULL, addr, 1989 buf, (void *) addr, len); 1990 else 1991 len = 0; 1992 } else { 1993 len = 0; 1994 } 1995 1996 up_read(&mm->mmap_sem); 1997 1998 return len; 1999} 2000 2001/** 2002 * @access_remote_vm - access another process' address space 2003 * @mm: the mm_struct of the target address space 2004 * @addr: start address to access 2005 * @buf: source or destination buffer 2006 * @len: number of bytes to transfer 2007 * @write: whether the access is a write 2008 * 2009 * The caller must hold a reference on @mm. 2010 */ 2011int access_remote_vm(struct mm_struct *mm, unsigned long addr, 2012 void *buf, int len, int write) 2013{ 2014 return __access_remote_vm(NULL, mm, addr, buf, len, write); 2015} 2016 2017/* 2018 * Access another process' address space. 2019 * - source/target buffer must be kernel space 2020 */ 2021int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 2022{ 2023 struct mm_struct *mm; 2024 2025 if (addr + len < addr) 2026 return 0; 2027 2028 mm = get_task_mm(tsk); 2029 if (!mm) 2030 return 0; 2031 2032 len = __access_remote_vm(tsk, mm, addr, buf, len, write); 2033 2034 mmput(mm); 2035 return len; 2036} 2037 2038/** 2039 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 2040 * @inode: The inode to check 2041 * @size: The current filesize of the inode 2042 * @newsize: The proposed filesize of the inode 2043 * 2044 * Check the shared mappings on an inode on behalf of a shrinking truncate to 2045 * make sure that that any outstanding VMAs aren't broken and then shrink the 2046 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 2047 * automatically grant mappings that are too large. 2048 */ 2049int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 2050 size_t newsize) 2051{ 2052 struct vm_area_struct *vma; 2053 struct prio_tree_iter iter; 2054 struct vm_region *region; 2055 pgoff_t low, high; 2056 size_t r_size, r_top; 2057 2058 low = newsize >> PAGE_SHIFT; 2059 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 2060 2061 down_write(&nommu_region_sem); 2062 2063 /* search for VMAs that fall within the dead zone */ 2064 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, 2065 low, high) { 2066 /* found one - only interested if it's shared out of the page 2067 * cache */ 2068 if (vma->vm_flags & VM_SHARED) { 2069 up_write(&nommu_region_sem); 2070 return -ETXTBSY; /* not quite true, but near enough */ 2071 } 2072 } 2073 2074 /* reduce any regions that overlap the dead zone - if in existence, 2075 * these will be pointed to by VMAs that don't overlap the dead zone 2076 * 2077 * we don't check for any regions that start beyond the EOF as there 2078 * shouldn't be any 2079 */ 2080 vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, 2081 0, ULONG_MAX) { 2082 if (!(vma->vm_flags & VM_SHARED)) 2083 continue; 2084 2085 region = vma->vm_region; 2086 r_size = region->vm_top - region->vm_start; 2087 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 2088 2089 if (r_top > newsize) { 2090 region->vm_top -= r_top - newsize; 2091 if (region->vm_end > region->vm_top) 2092 region->vm_end = region->vm_top; 2093 } 2094 } 2095 2096 up_write(&nommu_region_sem); 2097 return 0; 2098} 2099