nommu.c revision 4266c97a3ef4604561a22212eb0eab8a3c338971
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 39static inline __attribute__((format(printf, 1, 2))) 40void no_printk(const char *fmt, ...) 41{ 42} 43 44#if 0 45#define kenter(FMT, ...) \ 46 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 47#define kleave(FMT, ...) \ 48 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 49#define kdebug(FMT, ...) \ 50 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 51#else 52#define kenter(FMT, ...) \ 53 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 54#define kleave(FMT, ...) \ 55 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 56#define kdebug(FMT, ...) \ 57 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 58#endif 59 60void *high_memory; 61struct page *mem_map; 62unsigned long max_mapnr; 63unsigned long num_physpages; 64unsigned long highest_memmap_pfn; 65struct percpu_counter vm_committed_as; 66int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 67int sysctl_overcommit_ratio = 50; /* default is 50% */ 68int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 69int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 70int heap_stack_gap = 0; 71 72atomic_long_t mmap_pages_allocated; 73 74EXPORT_SYMBOL(mem_map); 75EXPORT_SYMBOL(num_physpages); 76 77/* list of mapped, potentially shareable regions */ 78static struct kmem_cache *vm_region_jar; 79struct rb_root nommu_region_tree = RB_ROOT; 80DECLARE_RWSEM(nommu_region_sem); 81 82struct vm_operations_struct generic_file_vm_ops = { 83}; 84 85/* 86 * Handle all mappings that got truncated by a "truncate()" 87 * system call. 88 * 89 * NOTE! We have to be ready to update the memory sharing 90 * between the file and the memory map for a potential last 91 * incomplete page. Ugly, but necessary. 92 */ 93int vmtruncate(struct inode *inode, loff_t offset) 94{ 95 struct address_space *mapping = inode->i_mapping; 96 unsigned long limit; 97 98 if (inode->i_size < offset) 99 goto do_expand; 100 i_size_write(inode, offset); 101 102 truncate_inode_pages(mapping, offset); 103 goto out_truncate; 104 105do_expand: 106 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 107 if (limit != RLIM_INFINITY && offset > limit) 108 goto out_sig; 109 if (offset > inode->i_sb->s_maxbytes) 110 goto out; 111 i_size_write(inode, offset); 112 113out_truncate: 114 if (inode->i_op->truncate) 115 inode->i_op->truncate(inode); 116 return 0; 117out_sig: 118 send_sig(SIGXFSZ, current, 0); 119out: 120 return -EFBIG; 121} 122 123EXPORT_SYMBOL(vmtruncate); 124 125/* 126 * Return the total memory allocated for this pointer, not 127 * just what the caller asked for. 128 * 129 * Doesn't have to be accurate, i.e. may have races. 130 */ 131unsigned int kobjsize(const void *objp) 132{ 133 struct page *page; 134 135 /* 136 * If the object we have should not have ksize performed on it, 137 * return size of 0 138 */ 139 if (!objp || !virt_addr_valid(objp)) 140 return 0; 141 142 page = virt_to_head_page(objp); 143 144 /* 145 * If the allocator sets PageSlab, we know the pointer came from 146 * kmalloc(). 147 */ 148 if (PageSlab(page)) 149 return ksize(objp); 150 151 /* 152 * If it's not a compound page, see if we have a matching VMA 153 * region. This test is intentionally done in reverse order, 154 * so if there's no VMA, we still fall through and hand back 155 * PAGE_SIZE for 0-order pages. 156 */ 157 if (!PageCompound(page)) { 158 struct vm_area_struct *vma; 159 160 vma = find_vma(current->mm, (unsigned long)objp); 161 if (vma) 162 return vma->vm_end - vma->vm_start; 163 } 164 165 /* 166 * The ksize() function is only guaranteed to work for pointers 167 * returned by kmalloc(). So handle arbitrary pointers here. 168 */ 169 return PAGE_SIZE << compound_order(page); 170} 171 172int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 173 unsigned long start, int nr_pages, unsigned int foll_flags, 174 struct page **pages, struct vm_area_struct **vmas) 175{ 176 struct vm_area_struct *vma; 177 unsigned long vm_flags; 178 int i; 179 180 /* calculate required read or write permissions. 181 * If FOLL_FORCE is set, we only require the "MAY" flags. 182 */ 183 vm_flags = (foll_flags & FOLL_WRITE) ? 184 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 185 vm_flags &= (foll_flags & FOLL_FORCE) ? 186 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 187 188 for (i = 0; i < nr_pages; i++) { 189 vma = find_vma(mm, start); 190 if (!vma) 191 goto finish_or_fault; 192 193 /* protect what we can, including chardevs */ 194 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 195 !(vm_flags & vma->vm_flags)) 196 goto finish_or_fault; 197 198 if (pages) { 199 pages[i] = virt_to_page(start); 200 if (pages[i]) 201 page_cache_get(pages[i]); 202 } 203 if (vmas) 204 vmas[i] = vma; 205 start += PAGE_SIZE; 206 } 207 208 return i; 209 210finish_or_fault: 211 return i ? : -EFAULT; 212} 213 214/* 215 * get a list of pages in an address range belonging to the specified process 216 * and indicate the VMA that covers each page 217 * - this is potentially dodgy as we may end incrementing the page count of a 218 * slab page or a secondary page from a compound page 219 * - don't permit access to VMAs that don't support it, such as I/O mappings 220 */ 221int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 222 unsigned long start, int nr_pages, int write, int force, 223 struct page **pages, struct vm_area_struct **vmas) 224{ 225 int flags = 0; 226 227 if (write) 228 flags |= FOLL_WRITE; 229 if (force) 230 flags |= FOLL_FORCE; 231 232 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); 233} 234EXPORT_SYMBOL(get_user_pages); 235 236/** 237 * follow_pfn - look up PFN at a user virtual address 238 * @vma: memory mapping 239 * @address: user virtual address 240 * @pfn: location to store found PFN 241 * 242 * Only IO mappings and raw PFN mappings are allowed. 243 * 244 * Returns zero and the pfn at @pfn on success, -ve otherwise. 245 */ 246int follow_pfn(struct vm_area_struct *vma, unsigned long address, 247 unsigned long *pfn) 248{ 249 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 250 return -EINVAL; 251 252 *pfn = address >> PAGE_SHIFT; 253 return 0; 254} 255EXPORT_SYMBOL(follow_pfn); 256 257DEFINE_RWLOCK(vmlist_lock); 258struct vm_struct *vmlist; 259 260void vfree(const void *addr) 261{ 262 kfree(addr); 263} 264EXPORT_SYMBOL(vfree); 265 266void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 267{ 268 /* 269 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 270 * returns only a logical address. 271 */ 272 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 273} 274EXPORT_SYMBOL(__vmalloc); 275 276void *vmalloc_user(unsigned long size) 277{ 278 void *ret; 279 280 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 281 PAGE_KERNEL); 282 if (ret) { 283 struct vm_area_struct *vma; 284 285 down_write(¤t->mm->mmap_sem); 286 vma = find_vma(current->mm, (unsigned long)ret); 287 if (vma) 288 vma->vm_flags |= VM_USERMAP; 289 up_write(¤t->mm->mmap_sem); 290 } 291 292 return ret; 293} 294EXPORT_SYMBOL(vmalloc_user); 295 296struct page *vmalloc_to_page(const void *addr) 297{ 298 return virt_to_page(addr); 299} 300EXPORT_SYMBOL(vmalloc_to_page); 301 302unsigned long vmalloc_to_pfn(const void *addr) 303{ 304 return page_to_pfn(virt_to_page(addr)); 305} 306EXPORT_SYMBOL(vmalloc_to_pfn); 307 308long vread(char *buf, char *addr, unsigned long count) 309{ 310 memcpy(buf, addr, count); 311 return count; 312} 313 314long vwrite(char *buf, char *addr, unsigned long count) 315{ 316 /* Don't allow overflow */ 317 if ((unsigned long) addr + count < count) 318 count = -(unsigned long) addr; 319 320 memcpy(addr, buf, count); 321 return(count); 322} 323 324/* 325 * vmalloc - allocate virtually continguos memory 326 * 327 * @size: allocation size 328 * 329 * Allocate enough pages to cover @size from the page level 330 * allocator and map them into continguos kernel virtual space. 331 * 332 * For tight control over page level allocator and protection flags 333 * use __vmalloc() instead. 334 */ 335void *vmalloc(unsigned long size) 336{ 337 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 338} 339EXPORT_SYMBOL(vmalloc); 340 341void *vmalloc_node(unsigned long size, int node) 342{ 343 return vmalloc(size); 344} 345EXPORT_SYMBOL(vmalloc_node); 346 347#ifndef PAGE_KERNEL_EXEC 348# define PAGE_KERNEL_EXEC PAGE_KERNEL 349#endif 350 351/** 352 * vmalloc_exec - allocate virtually contiguous, executable memory 353 * @size: allocation size 354 * 355 * Kernel-internal function to allocate enough pages to cover @size 356 * the page level allocator and map them into contiguous and 357 * executable kernel virtual space. 358 * 359 * For tight control over page level allocator and protection flags 360 * use __vmalloc() instead. 361 */ 362 363void *vmalloc_exec(unsigned long size) 364{ 365 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 366} 367 368/** 369 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 370 * @size: allocation size 371 * 372 * Allocate enough 32bit PA addressable pages to cover @size from the 373 * page level allocator and map them into continguos kernel virtual space. 374 */ 375void *vmalloc_32(unsigned long size) 376{ 377 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 378} 379EXPORT_SYMBOL(vmalloc_32); 380 381/** 382 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 383 * @size: allocation size 384 * 385 * The resulting memory area is 32bit addressable and zeroed so it can be 386 * mapped to userspace without leaking data. 387 * 388 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 389 * remap_vmalloc_range() are permissible. 390 */ 391void *vmalloc_32_user(unsigned long size) 392{ 393 /* 394 * We'll have to sort out the ZONE_DMA bits for 64-bit, 395 * but for now this can simply use vmalloc_user() directly. 396 */ 397 return vmalloc_user(size); 398} 399EXPORT_SYMBOL(vmalloc_32_user); 400 401void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 402{ 403 BUG(); 404 return NULL; 405} 406EXPORT_SYMBOL(vmap); 407 408void vunmap(const void *addr) 409{ 410 BUG(); 411} 412EXPORT_SYMBOL(vunmap); 413 414void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 415{ 416 BUG(); 417 return NULL; 418} 419EXPORT_SYMBOL(vm_map_ram); 420 421void vm_unmap_ram(const void *mem, unsigned int count) 422{ 423 BUG(); 424} 425EXPORT_SYMBOL(vm_unmap_ram); 426 427void vm_unmap_aliases(void) 428{ 429} 430EXPORT_SYMBOL_GPL(vm_unmap_aliases); 431 432/* 433 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 434 * have one. 435 */ 436void __attribute__((weak)) vmalloc_sync_all(void) 437{ 438} 439 440int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 441 struct page *page) 442{ 443 return -EINVAL; 444} 445EXPORT_SYMBOL(vm_insert_page); 446 447/* 448 * sys_brk() for the most part doesn't need the global kernel 449 * lock, except when an application is doing something nasty 450 * like trying to un-brk an area that has already been mapped 451 * to a regular file. in this case, the unmapping will need 452 * to invoke file system routines that need the global lock. 453 */ 454SYSCALL_DEFINE1(brk, unsigned long, brk) 455{ 456 struct mm_struct *mm = current->mm; 457 458 if (brk < mm->start_brk || brk > mm->context.end_brk) 459 return mm->brk; 460 461 if (mm->brk == brk) 462 return mm->brk; 463 464 /* 465 * Always allow shrinking brk 466 */ 467 if (brk <= mm->brk) { 468 mm->brk = brk; 469 return brk; 470 } 471 472 /* 473 * Ok, looks good - let it rip. 474 */ 475 return mm->brk = brk; 476} 477 478/* 479 * initialise the VMA and region record slabs 480 */ 481void __init mmap_init(void) 482{ 483 int ret; 484 485 ret = percpu_counter_init(&vm_committed_as, 0); 486 VM_BUG_ON(ret); 487 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 488} 489 490/* 491 * validate the region tree 492 * - the caller must hold the region lock 493 */ 494#ifdef CONFIG_DEBUG_NOMMU_REGIONS 495static noinline void validate_nommu_regions(void) 496{ 497 struct vm_region *region, *last; 498 struct rb_node *p, *lastp; 499 500 lastp = rb_first(&nommu_region_tree); 501 if (!lastp) 502 return; 503 504 last = rb_entry(lastp, struct vm_region, vm_rb); 505 BUG_ON(unlikely(last->vm_end <= last->vm_start)); 506 BUG_ON(unlikely(last->vm_top < last->vm_end)); 507 508 while ((p = rb_next(lastp))) { 509 region = rb_entry(p, struct vm_region, vm_rb); 510 last = rb_entry(lastp, struct vm_region, vm_rb); 511 512 BUG_ON(unlikely(region->vm_end <= region->vm_start)); 513 BUG_ON(unlikely(region->vm_top < region->vm_end)); 514 BUG_ON(unlikely(region->vm_start < last->vm_top)); 515 516 lastp = p; 517 } 518} 519#else 520static void validate_nommu_regions(void) 521{ 522} 523#endif 524 525/* 526 * add a region into the global tree 527 */ 528static void add_nommu_region(struct vm_region *region) 529{ 530 struct vm_region *pregion; 531 struct rb_node **p, *parent; 532 533 validate_nommu_regions(); 534 535 parent = NULL; 536 p = &nommu_region_tree.rb_node; 537 while (*p) { 538 parent = *p; 539 pregion = rb_entry(parent, struct vm_region, vm_rb); 540 if (region->vm_start < pregion->vm_start) 541 p = &(*p)->rb_left; 542 else if (region->vm_start > pregion->vm_start) 543 p = &(*p)->rb_right; 544 else if (pregion == region) 545 return; 546 else 547 BUG(); 548 } 549 550 rb_link_node(®ion->vm_rb, parent, p); 551 rb_insert_color(®ion->vm_rb, &nommu_region_tree); 552 553 validate_nommu_regions(); 554} 555 556/* 557 * delete a region from the global tree 558 */ 559static void delete_nommu_region(struct vm_region *region) 560{ 561 BUG_ON(!nommu_region_tree.rb_node); 562 563 validate_nommu_regions(); 564 rb_erase(®ion->vm_rb, &nommu_region_tree); 565 validate_nommu_regions(); 566} 567 568/* 569 * free a contiguous series of pages 570 */ 571static void free_page_series(unsigned long from, unsigned long to) 572{ 573 for (; from < to; from += PAGE_SIZE) { 574 struct page *page = virt_to_page(from); 575 576 kdebug("- free %lx", from); 577 atomic_long_dec(&mmap_pages_allocated); 578 if (page_count(page) != 1) 579 kdebug("free page %p: refcount not one: %d", 580 page, page_count(page)); 581 put_page(page); 582 } 583} 584 585/* 586 * release a reference to a region 587 * - the caller must hold the region semaphore for writing, which this releases 588 * - the region may not have been added to the tree yet, in which case vm_top 589 * will equal vm_start 590 */ 591static void __put_nommu_region(struct vm_region *region) 592 __releases(nommu_region_sem) 593{ 594 kenter("%p{%d}", region, atomic_read(®ion->vm_usage)); 595 596 BUG_ON(!nommu_region_tree.rb_node); 597 598 if (atomic_dec_and_test(®ion->vm_usage)) { 599 if (region->vm_top > region->vm_start) 600 delete_nommu_region(region); 601 up_write(&nommu_region_sem); 602 603 if (region->vm_file) 604 fput(region->vm_file); 605 606 /* IO memory and memory shared directly out of the pagecache 607 * from ramfs/tmpfs mustn't be released here */ 608 if (region->vm_flags & VM_MAPPED_COPY) { 609 kdebug("free series"); 610 free_page_series(region->vm_start, region->vm_top); 611 } 612 kmem_cache_free(vm_region_jar, region); 613 } else { 614 up_write(&nommu_region_sem); 615 } 616} 617 618/* 619 * release a reference to a region 620 */ 621static void put_nommu_region(struct vm_region *region) 622{ 623 down_write(&nommu_region_sem); 624 __put_nommu_region(region); 625} 626 627/* 628 * update protection on a vma 629 */ 630static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 631{ 632#ifdef CONFIG_MPU 633 struct mm_struct *mm = vma->vm_mm; 634 long start = vma->vm_start & PAGE_MASK; 635 while (start < vma->vm_end) { 636 protect_page(mm, start, flags); 637 start += PAGE_SIZE; 638 } 639 update_protections(mm); 640#endif 641} 642 643/* 644 * add a VMA into a process's mm_struct in the appropriate place in the list 645 * and tree and add to the address space's page tree also if not an anonymous 646 * page 647 * - should be called with mm->mmap_sem held writelocked 648 */ 649static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 650{ 651 struct vm_area_struct *pvma, **pp; 652 struct address_space *mapping; 653 struct rb_node **p, *parent; 654 655 kenter(",%p", vma); 656 657 BUG_ON(!vma->vm_region); 658 659 mm->map_count++; 660 vma->vm_mm = mm; 661 662 protect_vma(vma, vma->vm_flags); 663 664 /* add the VMA to the mapping */ 665 if (vma->vm_file) { 666 mapping = vma->vm_file->f_mapping; 667 668 flush_dcache_mmap_lock(mapping); 669 vma_prio_tree_insert(vma, &mapping->i_mmap); 670 flush_dcache_mmap_unlock(mapping); 671 } 672 673 /* add the VMA to the tree */ 674 parent = NULL; 675 p = &mm->mm_rb.rb_node; 676 while (*p) { 677 parent = *p; 678 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 679 680 /* sort by: start addr, end addr, VMA struct addr in that order 681 * (the latter is necessary as we may get identical VMAs) */ 682 if (vma->vm_start < pvma->vm_start) 683 p = &(*p)->rb_left; 684 else if (vma->vm_start > pvma->vm_start) 685 p = &(*p)->rb_right; 686 else if (vma->vm_end < pvma->vm_end) 687 p = &(*p)->rb_left; 688 else if (vma->vm_end > pvma->vm_end) 689 p = &(*p)->rb_right; 690 else if (vma < pvma) 691 p = &(*p)->rb_left; 692 else if (vma > pvma) 693 p = &(*p)->rb_right; 694 else 695 BUG(); 696 } 697 698 rb_link_node(&vma->vm_rb, parent, p); 699 rb_insert_color(&vma->vm_rb, &mm->mm_rb); 700 701 /* add VMA to the VMA list also */ 702 for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) { 703 if (pvma->vm_start > vma->vm_start) 704 break; 705 if (pvma->vm_start < vma->vm_start) 706 continue; 707 if (pvma->vm_end < vma->vm_end) 708 break; 709 } 710 711 vma->vm_next = *pp; 712 *pp = vma; 713} 714 715/* 716 * delete a VMA from its owning mm_struct and address space 717 */ 718static void delete_vma_from_mm(struct vm_area_struct *vma) 719{ 720 struct vm_area_struct **pp; 721 struct address_space *mapping; 722 struct mm_struct *mm = vma->vm_mm; 723 724 kenter("%p", vma); 725 726 protect_vma(vma, 0); 727 728 mm->map_count--; 729 if (mm->mmap_cache == vma) 730 mm->mmap_cache = NULL; 731 732 /* remove the VMA from the mapping */ 733 if (vma->vm_file) { 734 mapping = vma->vm_file->f_mapping; 735 736 flush_dcache_mmap_lock(mapping); 737 vma_prio_tree_remove(vma, &mapping->i_mmap); 738 flush_dcache_mmap_unlock(mapping); 739 } 740 741 /* remove from the MM's tree and list */ 742 rb_erase(&vma->vm_rb, &mm->mm_rb); 743 for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) { 744 if (*pp == vma) { 745 *pp = vma->vm_next; 746 break; 747 } 748 } 749 750 vma->vm_mm = NULL; 751} 752 753/* 754 * destroy a VMA record 755 */ 756static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 757{ 758 kenter("%p", vma); 759 if (vma->vm_ops && vma->vm_ops->close) 760 vma->vm_ops->close(vma); 761 if (vma->vm_file) { 762 fput(vma->vm_file); 763 if (vma->vm_flags & VM_EXECUTABLE) 764 removed_exe_file_vma(mm); 765 } 766 put_nommu_region(vma->vm_region); 767 kmem_cache_free(vm_area_cachep, vma); 768} 769 770/* 771 * look up the first VMA in which addr resides, NULL if none 772 * - should be called with mm->mmap_sem at least held readlocked 773 */ 774struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 775{ 776 struct vm_area_struct *vma; 777 struct rb_node *n = mm->mm_rb.rb_node; 778 779 /* check the cache first */ 780 vma = mm->mmap_cache; 781 if (vma && vma->vm_start <= addr && vma->vm_end > addr) 782 return vma; 783 784 /* trawl the tree (there may be multiple mappings in which addr 785 * resides) */ 786 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { 787 vma = rb_entry(n, struct vm_area_struct, vm_rb); 788 if (vma->vm_start > addr) 789 return NULL; 790 if (vma->vm_end > addr) { 791 mm->mmap_cache = vma; 792 return vma; 793 } 794 } 795 796 return NULL; 797} 798EXPORT_SYMBOL(find_vma); 799 800/* 801 * find a VMA 802 * - we don't extend stack VMAs under NOMMU conditions 803 */ 804struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 805{ 806 return find_vma(mm, addr); 807} 808 809/* 810 * expand a stack to a given address 811 * - not supported under NOMMU conditions 812 */ 813int expand_stack(struct vm_area_struct *vma, unsigned long address) 814{ 815 return -ENOMEM; 816} 817 818/* 819 * look up the first VMA exactly that exactly matches addr 820 * - should be called with mm->mmap_sem at least held readlocked 821 */ 822static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 823 unsigned long addr, 824 unsigned long len) 825{ 826 struct vm_area_struct *vma; 827 struct rb_node *n = mm->mm_rb.rb_node; 828 unsigned long end = addr + len; 829 830 /* check the cache first */ 831 vma = mm->mmap_cache; 832 if (vma && vma->vm_start == addr && vma->vm_end == end) 833 return vma; 834 835 /* trawl the tree (there may be multiple mappings in which addr 836 * resides) */ 837 for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { 838 vma = rb_entry(n, struct vm_area_struct, vm_rb); 839 if (vma->vm_start < addr) 840 continue; 841 if (vma->vm_start > addr) 842 return NULL; 843 if (vma->vm_end == end) { 844 mm->mmap_cache = vma; 845 return vma; 846 } 847 } 848 849 return NULL; 850} 851 852/* 853 * determine whether a mapping should be permitted and, if so, what sort of 854 * mapping we're capable of supporting 855 */ 856static int validate_mmap_request(struct file *file, 857 unsigned long addr, 858 unsigned long len, 859 unsigned long prot, 860 unsigned long flags, 861 unsigned long pgoff, 862 unsigned long *_capabilities) 863{ 864 unsigned long capabilities, rlen; 865 unsigned long reqprot = prot; 866 int ret; 867 868 /* do the simple checks first */ 869 if (flags & MAP_FIXED || addr) { 870 printk(KERN_DEBUG 871 "%d: Can't do fixed-address/overlay mmap of RAM\n", 872 current->pid); 873 return -EINVAL; 874 } 875 876 if ((flags & MAP_TYPE) != MAP_PRIVATE && 877 (flags & MAP_TYPE) != MAP_SHARED) 878 return -EINVAL; 879 880 if (!len) 881 return -EINVAL; 882 883 /* Careful about overflows.. */ 884 rlen = PAGE_ALIGN(len); 885 if (!rlen || rlen > TASK_SIZE) 886 return -ENOMEM; 887 888 /* offset overflow? */ 889 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 890 return -EOVERFLOW; 891 892 if (file) { 893 /* validate file mapping requests */ 894 struct address_space *mapping; 895 896 /* files must support mmap */ 897 if (!file->f_op || !file->f_op->mmap) 898 return -ENODEV; 899 900 /* work out if what we've got could possibly be shared 901 * - we support chardevs that provide their own "memory" 902 * - we support files/blockdevs that are memory backed 903 */ 904 mapping = file->f_mapping; 905 if (!mapping) 906 mapping = file->f_path.dentry->d_inode->i_mapping; 907 908 capabilities = 0; 909 if (mapping && mapping->backing_dev_info) 910 capabilities = mapping->backing_dev_info->capabilities; 911 912 if (!capabilities) { 913 /* no explicit capabilities set, so assume some 914 * defaults */ 915 switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { 916 case S_IFREG: 917 case S_IFBLK: 918 capabilities = BDI_CAP_MAP_COPY; 919 break; 920 921 case S_IFCHR: 922 capabilities = 923 BDI_CAP_MAP_DIRECT | 924 BDI_CAP_READ_MAP | 925 BDI_CAP_WRITE_MAP; 926 break; 927 928 default: 929 return -EINVAL; 930 } 931 } 932 933 /* eliminate any capabilities that we can't support on this 934 * device */ 935 if (!file->f_op->get_unmapped_area) 936 capabilities &= ~BDI_CAP_MAP_DIRECT; 937 if (!file->f_op->read) 938 capabilities &= ~BDI_CAP_MAP_COPY; 939 940 /* The file shall have been opened with read permission. */ 941 if (!(file->f_mode & FMODE_READ)) 942 return -EACCES; 943 944 if (flags & MAP_SHARED) { 945 /* do checks for writing, appending and locking */ 946 if ((prot & PROT_WRITE) && 947 !(file->f_mode & FMODE_WRITE)) 948 return -EACCES; 949 950 if (IS_APPEND(file->f_path.dentry->d_inode) && 951 (file->f_mode & FMODE_WRITE)) 952 return -EACCES; 953 954 if (locks_verify_locked(file->f_path.dentry->d_inode)) 955 return -EAGAIN; 956 957 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 958 return -ENODEV; 959 960 if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 961 ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 962 ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 963 ) { 964 printk("MAP_SHARED not completely supported on !MMU\n"); 965 return -EINVAL; 966 } 967 968 /* we mustn't privatise shared mappings */ 969 capabilities &= ~BDI_CAP_MAP_COPY; 970 } 971 else { 972 /* we're going to read the file into private memory we 973 * allocate */ 974 if (!(capabilities & BDI_CAP_MAP_COPY)) 975 return -ENODEV; 976 977 /* we don't permit a private writable mapping to be 978 * shared with the backing device */ 979 if (prot & PROT_WRITE) 980 capabilities &= ~BDI_CAP_MAP_DIRECT; 981 } 982 983 /* handle executable mappings and implied executable 984 * mappings */ 985 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 986 if (prot & PROT_EXEC) 987 return -EPERM; 988 } 989 else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 990 /* handle implication of PROT_EXEC by PROT_READ */ 991 if (current->personality & READ_IMPLIES_EXEC) { 992 if (capabilities & BDI_CAP_EXEC_MAP) 993 prot |= PROT_EXEC; 994 } 995 } 996 else if ((prot & PROT_READ) && 997 (prot & PROT_EXEC) && 998 !(capabilities & BDI_CAP_EXEC_MAP) 999 ) { 1000 /* backing file is not executable, try to copy */ 1001 capabilities &= ~BDI_CAP_MAP_DIRECT; 1002 } 1003 } 1004 else { 1005 /* anonymous mappings are always memory backed and can be 1006 * privately mapped 1007 */ 1008 capabilities = BDI_CAP_MAP_COPY; 1009 1010 /* handle PROT_EXEC implication by PROT_READ */ 1011 if ((prot & PROT_READ) && 1012 (current->personality & READ_IMPLIES_EXEC)) 1013 prot |= PROT_EXEC; 1014 } 1015 1016 /* allow the security API to have its say */ 1017 ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); 1018 if (ret < 0) 1019 return ret; 1020 1021 /* looks okay */ 1022 *_capabilities = capabilities; 1023 return 0; 1024} 1025 1026/* 1027 * we've determined that we can make the mapping, now translate what we 1028 * now know into VMA flags 1029 */ 1030static unsigned long determine_vm_flags(struct file *file, 1031 unsigned long prot, 1032 unsigned long flags, 1033 unsigned long capabilities) 1034{ 1035 unsigned long vm_flags; 1036 1037 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 1038 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 1039 /* vm_flags |= mm->def_flags; */ 1040 1041 if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 1042 /* attempt to share read-only copies of mapped file chunks */ 1043 if (file && !(prot & PROT_WRITE)) 1044 vm_flags |= VM_MAYSHARE; 1045 } 1046 else { 1047 /* overlay a shareable mapping on the backing device or inode 1048 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 1049 * romfs/cramfs */ 1050 if (flags & MAP_SHARED) 1051 vm_flags |= VM_MAYSHARE | VM_SHARED; 1052 else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0) 1053 vm_flags |= VM_MAYSHARE; 1054 } 1055 1056 /* refuse to let anyone share private mappings with this process if 1057 * it's being traced - otherwise breakpoints set in it may interfere 1058 * with another untraced process 1059 */ 1060 if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) 1061 vm_flags &= ~VM_MAYSHARE; 1062 1063 return vm_flags; 1064} 1065 1066/* 1067 * set up a shared mapping on a file (the driver or filesystem provides and 1068 * pins the storage) 1069 */ 1070static int do_mmap_shared_file(struct vm_area_struct *vma) 1071{ 1072 int ret; 1073 1074 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1075 if (ret == 0) { 1076 vma->vm_region->vm_top = vma->vm_region->vm_end; 1077 return ret; 1078 } 1079 if (ret != -ENOSYS) 1080 return ret; 1081 1082 /* getting an ENOSYS error indicates that direct mmap isn't 1083 * possible (as opposed to tried but failed) so we'll fall 1084 * through to making a private copy of the data and mapping 1085 * that if we can */ 1086 return -ENODEV; 1087} 1088 1089/* 1090 * set up a private mapping or an anonymous shared mapping 1091 */ 1092static int do_mmap_private(struct vm_area_struct *vma, 1093 struct vm_region *region, 1094 unsigned long len) 1095{ 1096 struct page *pages; 1097 unsigned long total, point, n, rlen; 1098 void *base; 1099 int ret, order; 1100 1101 /* invoke the file's mapping function so that it can keep track of 1102 * shared mappings on devices or memory 1103 * - VM_MAYSHARE will be set if it may attempt to share 1104 */ 1105 if (vma->vm_file) { 1106 ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1107 if (ret == 0) { 1108 /* shouldn't return success if we're not sharing */ 1109 BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1110 vma->vm_region->vm_top = vma->vm_region->vm_end; 1111 return ret; 1112 } 1113 if (ret != -ENOSYS) 1114 return ret; 1115 1116 /* getting an ENOSYS error indicates that direct mmap isn't 1117 * possible (as opposed to tried but failed) so we'll try to 1118 * make a private copy of the data and map that instead */ 1119 } 1120 1121 rlen = PAGE_ALIGN(len); 1122 1123 /* allocate some memory to hold the mapping 1124 * - note that this may not return a page-aligned address if the object 1125 * we're allocating is smaller than a page 1126 */ 1127 order = get_order(rlen); 1128 kdebug("alloc order %d for %lx", order, len); 1129 1130 pages = alloc_pages(GFP_KERNEL, order); 1131 if (!pages) 1132 goto enomem; 1133 1134 total = 1 << order; 1135 atomic_long_add(total, &mmap_pages_allocated); 1136 1137 point = rlen >> PAGE_SHIFT; 1138 1139 /* we allocated a power-of-2 sized page set, so we may want to trim off 1140 * the excess */ 1141 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 1142 while (total > point) { 1143 order = ilog2(total - point); 1144 n = 1 << order; 1145 kdebug("shave %lu/%lu @%lu", n, total - point, total); 1146 atomic_long_sub(n, &mmap_pages_allocated); 1147 total -= n; 1148 set_page_refcounted(pages + total); 1149 __free_pages(pages + total, order); 1150 } 1151 } 1152 1153 for (point = 1; point < total; point++) 1154 set_page_refcounted(&pages[point]); 1155 1156 base = page_address(pages); 1157 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 1158 region->vm_start = (unsigned long) base; 1159 region->vm_end = region->vm_start + rlen; 1160 region->vm_top = region->vm_start + (total << PAGE_SHIFT); 1161 1162 vma->vm_start = region->vm_start; 1163 vma->vm_end = region->vm_start + len; 1164 1165 if (vma->vm_file) { 1166 /* read the contents of a file into the copy */ 1167 mm_segment_t old_fs; 1168 loff_t fpos; 1169 1170 fpos = vma->vm_pgoff; 1171 fpos <<= PAGE_SHIFT; 1172 1173 old_fs = get_fs(); 1174 set_fs(KERNEL_DS); 1175 ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); 1176 set_fs(old_fs); 1177 1178 if (ret < 0) 1179 goto error_free; 1180 1181 /* clear the last little bit */ 1182 if (ret < rlen) 1183 memset(base + ret, 0, rlen - ret); 1184 1185 } else { 1186 /* if it's an anonymous mapping, then just clear it */ 1187 memset(base, 0, rlen); 1188 } 1189 1190 return 0; 1191 1192error_free: 1193 free_page_series(region->vm_start, region->vm_end); 1194 region->vm_start = vma->vm_start = 0; 1195 region->vm_end = vma->vm_end = 0; 1196 region->vm_top = 0; 1197 return ret; 1198 1199enomem: 1200 printk("Allocation of length %lu from process %d (%s) failed\n", 1201 len, current->pid, current->comm); 1202 show_free_areas(); 1203 return -ENOMEM; 1204} 1205 1206/* 1207 * handle mapping creation for uClinux 1208 */ 1209unsigned long do_mmap_pgoff(struct file *file, 1210 unsigned long addr, 1211 unsigned long len, 1212 unsigned long prot, 1213 unsigned long flags, 1214 unsigned long pgoff) 1215{ 1216 struct vm_area_struct *vma; 1217 struct vm_region *region; 1218 struct rb_node *rb; 1219 unsigned long capabilities, vm_flags, result; 1220 int ret; 1221 1222 kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 1223 1224 if (!(flags & MAP_FIXED)) 1225 addr = round_hint_to_min(addr); 1226 1227 /* decide whether we should attempt the mapping, and if so what sort of 1228 * mapping */ 1229 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 1230 &capabilities); 1231 if (ret < 0) { 1232 kleave(" = %d [val]", ret); 1233 return ret; 1234 } 1235 1236 /* we've determined that we can make the mapping, now translate what we 1237 * now know into VMA flags */ 1238 vm_flags = determine_vm_flags(file, prot, flags, capabilities); 1239 1240 /* we're going to need to record the mapping */ 1241 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 1242 if (!region) 1243 goto error_getting_region; 1244 1245 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 1246 if (!vma) 1247 goto error_getting_vma; 1248 1249 atomic_set(®ion->vm_usage, 1); 1250 region->vm_flags = vm_flags; 1251 region->vm_pgoff = pgoff; 1252 1253 INIT_LIST_HEAD(&vma->anon_vma_node); 1254 vma->vm_flags = vm_flags; 1255 vma->vm_pgoff = pgoff; 1256 1257 if (file) { 1258 region->vm_file = file; 1259 get_file(file); 1260 vma->vm_file = file; 1261 get_file(file); 1262 if (vm_flags & VM_EXECUTABLE) { 1263 added_exe_file_vma(current->mm); 1264 vma->vm_mm = current->mm; 1265 } 1266 } 1267 1268 down_write(&nommu_region_sem); 1269 1270 /* if we want to share, we need to check for regions created by other 1271 * mmap() calls that overlap with our proposed mapping 1272 * - we can only share with a superset match on most regular files 1273 * - shared mappings on character devices and memory backed files are 1274 * permitted to overlap inexactly as far as we are concerned for in 1275 * these cases, sharing is handled in the driver or filesystem rather 1276 * than here 1277 */ 1278 if (vm_flags & VM_MAYSHARE) { 1279 struct vm_region *pregion; 1280 unsigned long pglen, rpglen, pgend, rpgend, start; 1281 1282 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1283 pgend = pgoff + pglen; 1284 1285 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 1286 pregion = rb_entry(rb, struct vm_region, vm_rb); 1287 1288 if (!(pregion->vm_flags & VM_MAYSHARE)) 1289 continue; 1290 1291 /* search for overlapping mappings on the same file */ 1292 if (pregion->vm_file->f_path.dentry->d_inode != 1293 file->f_path.dentry->d_inode) 1294 continue; 1295 1296 if (pregion->vm_pgoff >= pgend) 1297 continue; 1298 1299 rpglen = pregion->vm_end - pregion->vm_start; 1300 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 1301 rpgend = pregion->vm_pgoff + rpglen; 1302 if (pgoff >= rpgend) 1303 continue; 1304 1305 /* handle inexactly overlapping matches between 1306 * mappings */ 1307 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 1308 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 1309 /* new mapping is not a subset of the region */ 1310 if (!(capabilities & BDI_CAP_MAP_DIRECT)) 1311 goto sharing_violation; 1312 continue; 1313 } 1314 1315 /* we've found a region we can share */ 1316 atomic_inc(&pregion->vm_usage); 1317 vma->vm_region = pregion; 1318 start = pregion->vm_start; 1319 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 1320 vma->vm_start = start; 1321 vma->vm_end = start + len; 1322 1323 if (pregion->vm_flags & VM_MAPPED_COPY) { 1324 kdebug("share copy"); 1325 vma->vm_flags |= VM_MAPPED_COPY; 1326 } else { 1327 kdebug("share mmap"); 1328 ret = do_mmap_shared_file(vma); 1329 if (ret < 0) { 1330 vma->vm_region = NULL; 1331 vma->vm_start = 0; 1332 vma->vm_end = 0; 1333 atomic_dec(&pregion->vm_usage); 1334 pregion = NULL; 1335 goto error_just_free; 1336 } 1337 } 1338 fput(region->vm_file); 1339 kmem_cache_free(vm_region_jar, region); 1340 region = pregion; 1341 result = start; 1342 goto share; 1343 } 1344 1345 /* obtain the address at which to make a shared mapping 1346 * - this is the hook for quasi-memory character devices to 1347 * tell us the location of a shared mapping 1348 */ 1349 if (file && file->f_op->get_unmapped_area) { 1350 addr = file->f_op->get_unmapped_area(file, addr, len, 1351 pgoff, flags); 1352 if (IS_ERR((void *) addr)) { 1353 ret = addr; 1354 if (ret != (unsigned long) -ENOSYS) 1355 goto error_just_free; 1356 1357 /* the driver refused to tell us where to site 1358 * the mapping so we'll have to attempt to copy 1359 * it */ 1360 ret = (unsigned long) -ENODEV; 1361 if (!(capabilities & BDI_CAP_MAP_COPY)) 1362 goto error_just_free; 1363 1364 capabilities &= ~BDI_CAP_MAP_DIRECT; 1365 } else { 1366 vma->vm_start = region->vm_start = addr; 1367 vma->vm_end = region->vm_end = addr + len; 1368 } 1369 } 1370 } 1371 1372 vma->vm_region = region; 1373 add_nommu_region(region); 1374 1375 /* set up the mapping */ 1376 if (file && vma->vm_flags & VM_SHARED) 1377 ret = do_mmap_shared_file(vma); 1378 else 1379 ret = do_mmap_private(vma, region, len); 1380 if (ret < 0) 1381 goto error_put_region; 1382 1383 /* okay... we have a mapping; now we have to register it */ 1384 result = vma->vm_start; 1385 1386 current->mm->total_vm += len >> PAGE_SHIFT; 1387 1388share: 1389 add_vma_to_mm(current->mm, vma); 1390 1391 up_write(&nommu_region_sem); 1392 1393 if (prot & PROT_EXEC) 1394 flush_icache_range(result, result + len); 1395 1396 kleave(" = %lx", result); 1397 return result; 1398 1399error_put_region: 1400 __put_nommu_region(region); 1401 if (vma) { 1402 if (vma->vm_file) { 1403 fput(vma->vm_file); 1404 if (vma->vm_flags & VM_EXECUTABLE) 1405 removed_exe_file_vma(vma->vm_mm); 1406 } 1407 kmem_cache_free(vm_area_cachep, vma); 1408 } 1409 kleave(" = %d [pr]", ret); 1410 return ret; 1411 1412error_just_free: 1413 up_write(&nommu_region_sem); 1414error: 1415 fput(region->vm_file); 1416 kmem_cache_free(vm_region_jar, region); 1417 fput(vma->vm_file); 1418 if (vma->vm_flags & VM_EXECUTABLE) 1419 removed_exe_file_vma(vma->vm_mm); 1420 kmem_cache_free(vm_area_cachep, vma); 1421 kleave(" = %d", ret); 1422 return ret; 1423 1424sharing_violation: 1425 up_write(&nommu_region_sem); 1426 printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 1427 ret = -EINVAL; 1428 goto error; 1429 1430error_getting_vma: 1431 kmem_cache_free(vm_region_jar, region); 1432 printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 1433 " from process %d failed\n", 1434 len, current->pid); 1435 show_free_areas(); 1436 return -ENOMEM; 1437 1438error_getting_region: 1439 printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 1440 " from process %d failed\n", 1441 len, current->pid); 1442 show_free_areas(); 1443 return -ENOMEM; 1444} 1445EXPORT_SYMBOL(do_mmap_pgoff); 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 that have a single 1461 * owner */ 1462 if (vma->vm_file || 1463 atomic_read(&vma->vm_region->vm_usage) != 1) 1464 return -ENOMEM; 1465 1466 if (mm->map_count >= sysctl_max_map_count) 1467 return -ENOMEM; 1468 1469 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 1470 if (!region) 1471 return -ENOMEM; 1472 1473 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 1474 if (!new) { 1475 kmem_cache_free(vm_region_jar, region); 1476 return -ENOMEM; 1477 } 1478 1479 /* most fields are the same, copy all, and then fixup */ 1480 *new = *vma; 1481 *region = *vma->vm_region; 1482 new->vm_region = region; 1483 1484 npages = (addr - vma->vm_start) >> PAGE_SHIFT; 1485 1486 if (new_below) { 1487 region->vm_top = region->vm_end = new->vm_end = addr; 1488 } else { 1489 region->vm_start = new->vm_start = addr; 1490 region->vm_pgoff = new->vm_pgoff += npages; 1491 } 1492 1493 if (new->vm_ops && new->vm_ops->open) 1494 new->vm_ops->open(new); 1495 1496 delete_vma_from_mm(vma); 1497 down_write(&nommu_region_sem); 1498 delete_nommu_region(vma->vm_region); 1499 if (new_below) { 1500 vma->vm_region->vm_start = vma->vm_start = addr; 1501 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 1502 } else { 1503 vma->vm_region->vm_end = vma->vm_end = addr; 1504 vma->vm_region->vm_top = addr; 1505 } 1506 add_nommu_region(vma->vm_region); 1507 add_nommu_region(new->vm_region); 1508 up_write(&nommu_region_sem); 1509 add_vma_to_mm(mm, vma); 1510 add_vma_to_mm(mm, new); 1511 return 0; 1512} 1513 1514/* 1515 * shrink a VMA by removing the specified chunk from either the beginning or 1516 * the end 1517 */ 1518static int shrink_vma(struct mm_struct *mm, 1519 struct vm_area_struct *vma, 1520 unsigned long from, unsigned long to) 1521{ 1522 struct vm_region *region; 1523 1524 kenter(""); 1525 1526 /* adjust the VMA's pointers, which may reposition it in the MM's tree 1527 * and list */ 1528 delete_vma_from_mm(vma); 1529 if (from > vma->vm_start) 1530 vma->vm_end = from; 1531 else 1532 vma->vm_start = to; 1533 add_vma_to_mm(mm, vma); 1534 1535 /* cut the backing region down to size */ 1536 region = vma->vm_region; 1537 BUG_ON(atomic_read(®ion->vm_usage) != 1); 1538 1539 down_write(&nommu_region_sem); 1540 delete_nommu_region(region); 1541 if (from > region->vm_start) { 1542 to = region->vm_top; 1543 region->vm_top = region->vm_end = from; 1544 } else { 1545 region->vm_start = to; 1546 } 1547 add_nommu_region(region); 1548 up_write(&nommu_region_sem); 1549 1550 free_page_series(from, to); 1551 return 0; 1552} 1553 1554/* 1555 * release a mapping 1556 * - under NOMMU conditions the chunk to be unmapped must be backed by a single 1557 * VMA, though it need not cover the whole VMA 1558 */ 1559int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 1560{ 1561 struct vm_area_struct *vma; 1562 struct rb_node *rb; 1563 unsigned long end = start + len; 1564 int ret; 1565 1566 kenter(",%lx,%zx", start, len); 1567 1568 if (len == 0) 1569 return -EINVAL; 1570 1571 /* find the first potentially overlapping VMA */ 1572 vma = find_vma(mm, start); 1573 if (!vma) { 1574 static int limit = 0; 1575 if (limit < 5) { 1576 printk(KERN_WARNING 1577 "munmap of memory not mmapped by process %d" 1578 " (%s): 0x%lx-0x%lx\n", 1579 current->pid, current->comm, 1580 start, start + len - 1); 1581 limit++; 1582 } 1583 return -EINVAL; 1584 } 1585 1586 /* we're allowed to split an anonymous VMA but not a file-backed one */ 1587 if (vma->vm_file) { 1588 do { 1589 if (start > vma->vm_start) { 1590 kleave(" = -EINVAL [miss]"); 1591 return -EINVAL; 1592 } 1593 if (end == vma->vm_end) 1594 goto erase_whole_vma; 1595 rb = rb_next(&vma->vm_rb); 1596 vma = rb_entry(rb, struct vm_area_struct, vm_rb); 1597 } while (rb); 1598 kleave(" = -EINVAL [split file]"); 1599 return -EINVAL; 1600 } else { 1601 /* the chunk must be a subset of the VMA found */ 1602 if (start == vma->vm_start && end == vma->vm_end) 1603 goto erase_whole_vma; 1604 if (start < vma->vm_start || end > vma->vm_end) { 1605 kleave(" = -EINVAL [superset]"); 1606 return -EINVAL; 1607 } 1608 if (start & ~PAGE_MASK) { 1609 kleave(" = -EINVAL [unaligned start]"); 1610 return -EINVAL; 1611 } 1612 if (end != vma->vm_end && end & ~PAGE_MASK) { 1613 kleave(" = -EINVAL [unaligned split]"); 1614 return -EINVAL; 1615 } 1616 if (start != vma->vm_start && end != vma->vm_end) { 1617 ret = split_vma(mm, vma, start, 1); 1618 if (ret < 0) { 1619 kleave(" = %d [split]", ret); 1620 return ret; 1621 } 1622 } 1623 return shrink_vma(mm, vma, start, end); 1624 } 1625 1626erase_whole_vma: 1627 delete_vma_from_mm(vma); 1628 delete_vma(mm, vma); 1629 kleave(" = 0"); 1630 return 0; 1631} 1632EXPORT_SYMBOL(do_munmap); 1633 1634SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1635{ 1636 int ret; 1637 struct mm_struct *mm = current->mm; 1638 1639 down_write(&mm->mmap_sem); 1640 ret = do_munmap(mm, addr, len); 1641 up_write(&mm->mmap_sem); 1642 return ret; 1643} 1644 1645/* 1646 * release all the mappings made in a process's VM space 1647 */ 1648void exit_mmap(struct mm_struct *mm) 1649{ 1650 struct vm_area_struct *vma; 1651 1652 if (!mm) 1653 return; 1654 1655 kenter(""); 1656 1657 mm->total_vm = 0; 1658 1659 while ((vma = mm->mmap)) { 1660 mm->mmap = vma->vm_next; 1661 delete_vma_from_mm(vma); 1662 delete_vma(mm, vma); 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 * ask for an unmapped area at which to create a mapping on a file 1782 */ 1783unsigned long get_unmapped_area(struct file *file, unsigned long addr, 1784 unsigned long len, unsigned long pgoff, 1785 unsigned long flags) 1786{ 1787 unsigned long (*get_area)(struct file *, unsigned long, unsigned long, 1788 unsigned long, unsigned long); 1789 1790 get_area = current->mm->get_unmapped_area; 1791 if (file && file->f_op && file->f_op->get_unmapped_area) 1792 get_area = file->f_op->get_unmapped_area; 1793 1794 if (!get_area) 1795 return -ENOSYS; 1796 1797 return get_area(file, addr, len, pgoff, flags); 1798} 1799EXPORT_SYMBOL(get_unmapped_area); 1800 1801/* 1802 * Check that a process has enough memory to allocate a new virtual 1803 * mapping. 0 means there is enough memory for the allocation to 1804 * succeed and -ENOMEM implies there is not. 1805 * 1806 * We currently support three overcommit policies, which are set via the 1807 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1808 * 1809 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1810 * Additional code 2002 Jul 20 by Robert Love. 1811 * 1812 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1813 * 1814 * Note this is a helper function intended to be used by LSMs which 1815 * wish to use this logic. 1816 */ 1817int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1818{ 1819 unsigned long free, allowed; 1820 1821 vm_acct_memory(pages); 1822 1823 /* 1824 * Sometimes we want to use more memory than we have 1825 */ 1826 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1827 return 0; 1828 1829 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1830 unsigned long n; 1831 1832 free = global_page_state(NR_FILE_PAGES); 1833 free += nr_swap_pages; 1834 1835 /* 1836 * Any slabs which are created with the 1837 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1838 * which are reclaimable, under pressure. The dentry 1839 * cache and most inode caches should fall into this 1840 */ 1841 free += global_page_state(NR_SLAB_RECLAIMABLE); 1842 1843 /* 1844 * Leave the last 3% for root 1845 */ 1846 if (!cap_sys_admin) 1847 free -= free / 32; 1848 1849 if (free > pages) 1850 return 0; 1851 1852 /* 1853 * nr_free_pages() is very expensive on large systems, 1854 * only call if we're about to fail. 1855 */ 1856 n = nr_free_pages(); 1857 1858 /* 1859 * Leave reserved pages. The pages are not for anonymous pages. 1860 */ 1861 if (n <= totalreserve_pages) 1862 goto error; 1863 else 1864 n -= totalreserve_pages; 1865 1866 /* 1867 * Leave the last 3% for root 1868 */ 1869 if (!cap_sys_admin) 1870 n -= n / 32; 1871 free += n; 1872 1873 if (free > pages) 1874 return 0; 1875 1876 goto error; 1877 } 1878 1879 allowed = totalram_pages * sysctl_overcommit_ratio / 100; 1880 /* 1881 * Leave the last 3% for root 1882 */ 1883 if (!cap_sys_admin) 1884 allowed -= allowed / 32; 1885 allowed += total_swap_pages; 1886 1887 /* Don't let a single process grow too big: 1888 leave 3% of the size of this process for other processes */ 1889 if (mm) 1890 allowed -= mm->total_vm / 32; 1891 1892 if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1893 return 0; 1894 1895error: 1896 vm_unacct_memory(pages); 1897 1898 return -ENOMEM; 1899} 1900 1901int in_gate_area_no_task(unsigned long addr) 1902{ 1903 return 0; 1904} 1905 1906int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1907{ 1908 BUG(); 1909 return 0; 1910} 1911EXPORT_SYMBOL(filemap_fault); 1912 1913/* 1914 * Access another process' address space. 1915 * - source/target buffer must be kernel space 1916 */ 1917int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 1918{ 1919 struct vm_area_struct *vma; 1920 struct mm_struct *mm; 1921 1922 if (addr + len < addr) 1923 return 0; 1924 1925 mm = get_task_mm(tsk); 1926 if (!mm) 1927 return 0; 1928 1929 down_read(&mm->mmap_sem); 1930 1931 /* the access must start within one of the target process's mappings */ 1932 vma = find_vma(mm, addr); 1933 if (vma) { 1934 /* don't overrun this mapping */ 1935 if (addr + len >= vma->vm_end) 1936 len = vma->vm_end - addr; 1937 1938 /* only read or write mappings where it is permitted */ 1939 if (write && vma->vm_flags & VM_MAYWRITE) 1940 len -= copy_to_user((void *) addr, buf, len); 1941 else if (!write && vma->vm_flags & VM_MAYREAD) 1942 len -= copy_from_user(buf, (void *) addr, len); 1943 else 1944 len = 0; 1945 } else { 1946 len = 0; 1947 } 1948 1949 up_read(&mm->mmap_sem); 1950 mmput(mm); 1951 return len; 1952} 1953