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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 17 18 #include <linux/export.h> 19 #include <linux/mm.h> 20 #include <linux/sched/mm.h> 21 #include <linux/vmacache.h> 22 #include <linux/mman.h> 23 #include <linux/swap.h> 24 #include <linux/file.h> 25 #include <linux/highmem.h> 26 #include <linux/pagemap.h> 27 #include <linux/slab.h> 28 #include <linux/vmalloc.h> 29 #include <linux/blkdev.h> 30 #include <linux/backing-dev.h> 31 #include <linux/compiler.h> 32 #include <linux/mount.h> 33 #include <linux/personality.h> 34 #include <linux/security.h> 35 #include <linux/syscalls.h> 36 #include <linux/audit.h> 37 #include <linux/printk.h> 38 39 #include <linux/uaccess.h> 40 #include <asm/tlb.h> 41 #include <asm/tlbflush.h> 42 #include <asm/mmu_context.h> 43 #include "internal.h" 44 45 void *high_memory; 46 EXPORT_SYMBOL(high_memory); 47 struct page *mem_map; 48 unsigned long max_mapnr; 49 EXPORT_SYMBOL(max_mapnr); 50 unsigned long highest_memmap_pfn; 51 int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 52 int heap_stack_gap = 0; 53 54 atomic_long_t mmap_pages_allocated; 55 56 EXPORT_SYMBOL(mem_map); 57 58 /* list of mapped, potentially shareable regions */ 59 static struct kmem_cache *vm_region_jar; 60 struct rb_root nommu_region_tree = RB_ROOT; 61 DECLARE_RWSEM(nommu_region_sem); 62 63 const struct vm_operations_struct generic_file_vm_ops = { 64 }; 65 66 /* 67 * Return the total memory allocated for this pointer, not 68 * just what the caller asked for. 69 * 70 * Doesn't have to be accurate, i.e. may have races. 71 */ 72 unsigned int kobjsize(const void *objp) 73 { 74 struct page *page; 75 76 /* 77 * If the object we have should not have ksize performed on it, 78 * return size of 0 79 */ 80 if (!objp || !virt_addr_valid(objp)) 81 return 0; 82 83 page = virt_to_head_page(objp); 84 85 /* 86 * If the allocator sets PageSlab, we know the pointer came from 87 * kmalloc(). 88 */ 89 if (PageSlab(page)) 90 return ksize(objp); 91 92 /* 93 * If it's not a compound page, see if we have a matching VMA 94 * region. This test is intentionally done in reverse order, 95 * so if there's no VMA, we still fall through and hand back 96 * PAGE_SIZE for 0-order pages. 97 */ 98 if (!PageCompound(page)) { 99 struct vm_area_struct *vma; 100 101 vma = find_vma(current->mm, (unsigned long)objp); 102 if (vma) 103 return vma->vm_end - vma->vm_start; 104 } 105 106 /* 107 * The ksize() function is only guaranteed to work for pointers 108 * returned by kmalloc(). So handle arbitrary pointers here. 109 */ 110 return PAGE_SIZE << compound_order(page); 111 } 112 113 static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 114 unsigned long start, unsigned long nr_pages, 115 unsigned int foll_flags, struct page **pages, 116 struct vm_area_struct **vmas, int *nonblocking) 117 { 118 struct vm_area_struct *vma; 119 unsigned long vm_flags; 120 int i; 121 122 /* calculate required read or write permissions. 123 * If FOLL_FORCE is set, we only require the "MAY" flags. 124 */ 125 vm_flags = (foll_flags & FOLL_WRITE) ? 126 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 127 vm_flags &= (foll_flags & FOLL_FORCE) ? 128 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 129 130 for (i = 0; i < nr_pages; i++) { 131 vma = find_vma(mm, start); 132 if (!vma) 133 goto finish_or_fault; 134 135 /* protect what we can, including chardevs */ 136 if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 137 !(vm_flags & vma->vm_flags)) 138 goto finish_or_fault; 139 140 if (pages) { 141 pages[i] = virt_to_page(start); 142 if (pages[i]) 143 get_page(pages[i]); 144 } 145 if (vmas) 146 vmas[i] = vma; 147 start = (start + PAGE_SIZE) & PAGE_MASK; 148 } 149 150 return i; 151 152 finish_or_fault: 153 return i ? : -EFAULT; 154 } 155 156 /* 157 * get a list of pages in an address range belonging to the specified process 158 * and indicate the VMA that covers each page 159 * - this is potentially dodgy as we may end incrementing the page count of a 160 * slab page or a secondary page from a compound page 161 * - don't permit access to VMAs that don't support it, such as I/O mappings 162 */ 163 long get_user_pages(unsigned long start, unsigned long nr_pages, 164 unsigned int gup_flags, struct page **pages, 165 struct vm_area_struct **vmas) 166 { 167 return __get_user_pages(current, current->mm, start, nr_pages, 168 gup_flags, pages, vmas, NULL); 169 } 170 EXPORT_SYMBOL(get_user_pages); 171 172 long get_user_pages_locked(unsigned long start, unsigned long nr_pages, 173 unsigned int gup_flags, struct page **pages, 174 int *locked) 175 { 176 return get_user_pages(start, nr_pages, gup_flags, pages, NULL); 177 } 178 EXPORT_SYMBOL(get_user_pages_locked); 179 180 static long __get_user_pages_unlocked(struct task_struct *tsk, 181 struct mm_struct *mm, unsigned long start, 182 unsigned long nr_pages, struct page **pages, 183 unsigned int gup_flags) 184 { 185 long ret; 186 down_read(&mm->mmap_sem); 187 ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages, 188 NULL, NULL); 189 up_read(&mm->mmap_sem); 190 return ret; 191 } 192 193 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, 194 struct page **pages, unsigned int gup_flags) 195 { 196 return __get_user_pages_unlocked(current, current->mm, start, nr_pages, 197 pages, gup_flags); 198 } 199 EXPORT_SYMBOL(get_user_pages_unlocked); 200 201 /** 202 * follow_pfn - look up PFN at a user virtual address 203 * @vma: memory mapping 204 * @address: user virtual address 205 * @pfn: location to store found PFN 206 * 207 * Only IO mappings and raw PFN mappings are allowed. 208 * 209 * Returns zero and the pfn at @pfn on success, -ve otherwise. 210 */ 211 int follow_pfn(struct vm_area_struct *vma, unsigned long address, 212 unsigned long *pfn) 213 { 214 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 215 return -EINVAL; 216 217 *pfn = address >> PAGE_SHIFT; 218 return 0; 219 } 220 EXPORT_SYMBOL(follow_pfn); 221 222 LIST_HEAD(vmap_area_list); 223 224 void vfree(const void *addr) 225 { 226 kfree(addr); 227 } 228 EXPORT_SYMBOL(vfree); 229 230 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 231 { 232 /* 233 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 234 * returns only a logical address. 235 */ 236 return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 237 } 238 EXPORT_SYMBOL(__vmalloc); 239 240 void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags) 241 { 242 return __vmalloc(size, flags, PAGE_KERNEL); 243 } 244 245 void *vmalloc_user(unsigned long size) 246 { 247 void *ret; 248 249 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL); 250 if (ret) { 251 struct vm_area_struct *vma; 252 253 down_write(¤t->mm->mmap_sem); 254 vma = find_vma(current->mm, (unsigned long)ret); 255 if (vma) 256 vma->vm_flags |= VM_USERMAP; 257 up_write(¤t->mm->mmap_sem); 258 } 259 260 return ret; 261 } 262 EXPORT_SYMBOL(vmalloc_user); 263 264 struct page *vmalloc_to_page(const void *addr) 265 { 266 return virt_to_page(addr); 267 } 268 EXPORT_SYMBOL(vmalloc_to_page); 269 270 unsigned long vmalloc_to_pfn(const void *addr) 271 { 272 return page_to_pfn(virt_to_page(addr)); 273 } 274 EXPORT_SYMBOL(vmalloc_to_pfn); 275 276 long vread(char *buf, char *addr, unsigned long count) 277 { 278 /* Don't allow overflow */ 279 if ((unsigned long) buf + count < count) 280 count = -(unsigned long) buf; 281 282 memcpy(buf, addr, count); 283 return count; 284 } 285 286 long vwrite(char *buf, char *addr, unsigned long count) 287 { 288 /* Don't allow overflow */ 289 if ((unsigned long) addr + count < count) 290 count = -(unsigned long) addr; 291 292 memcpy(addr, buf, count); 293 return count; 294 } 295 296 /* 297 * vmalloc - allocate virtually contiguous memory 298 * 299 * @size: allocation size 300 * 301 * Allocate enough pages to cover @size from the page level 302 * allocator and map them into contiguous kernel virtual space. 303 * 304 * For tight control over page level allocator and protection flags 305 * use __vmalloc() instead. 306 */ 307 void *vmalloc(unsigned long size) 308 { 309 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 310 } 311 EXPORT_SYMBOL(vmalloc); 312 313 /* 314 * vzalloc - allocate virtually contiguous memory with zero fill 315 * 316 * @size: allocation size 317 * 318 * Allocate enough pages to cover @size from the page level 319 * allocator and map them into contiguous kernel virtual space. 320 * The memory allocated is set to zero. 321 * 322 * For tight control over page level allocator and protection flags 323 * use __vmalloc() instead. 324 */ 325 void *vzalloc(unsigned long size) 326 { 327 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 328 PAGE_KERNEL); 329 } 330 EXPORT_SYMBOL(vzalloc); 331 332 /** 333 * vmalloc_node - allocate memory on a specific node 334 * @size: allocation size 335 * @node: numa node 336 * 337 * Allocate enough pages to cover @size from the page level 338 * allocator and map them into contiguous kernel virtual space. 339 * 340 * For tight control over page level allocator and protection flags 341 * use __vmalloc() instead. 342 */ 343 void *vmalloc_node(unsigned long size, int node) 344 { 345 return vmalloc(size); 346 } 347 EXPORT_SYMBOL(vmalloc_node); 348 349 /** 350 * vzalloc_node - allocate memory on a specific node with zero fill 351 * @size: allocation size 352 * @node: numa node 353 * 354 * Allocate enough pages to cover @size from the page level 355 * allocator and map them into contiguous kernel virtual space. 356 * The memory allocated is set to zero. 357 * 358 * For tight control over page level allocator and protection flags 359 * use __vmalloc() instead. 360 */ 361 void *vzalloc_node(unsigned long size, int node) 362 { 363 return vzalloc(size); 364 } 365 EXPORT_SYMBOL(vzalloc_node); 366 367 /** 368 * vmalloc_exec - allocate virtually contiguous, executable memory 369 * @size: allocation size 370 * 371 * Kernel-internal function to allocate enough pages to cover @size 372 * the page level allocator and map them into contiguous and 373 * executable kernel virtual space. 374 * 375 * For tight control over page level allocator and protection flags 376 * use __vmalloc() instead. 377 */ 378 379 void *vmalloc_exec(unsigned long size) 380 { 381 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 382 } 383 384 /** 385 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 386 * @size: allocation size 387 * 388 * Allocate enough 32bit PA addressable pages to cover @size from the 389 * page level allocator and map them into contiguous kernel virtual space. 390 */ 391 void *vmalloc_32(unsigned long size) 392 { 393 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 394 } 395 EXPORT_SYMBOL(vmalloc_32); 396 397 /** 398 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 399 * @size: allocation size 400 * 401 * The resulting memory area is 32bit addressable and zeroed so it can be 402 * mapped to userspace without leaking data. 403 * 404 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 405 * remap_vmalloc_range() are permissible. 406 */ 407 void *vmalloc_32_user(unsigned long size) 408 { 409 /* 410 * We'll have to sort out the ZONE_DMA bits for 64-bit, 411 * but for now this can simply use vmalloc_user() directly. 412 */ 413 return vmalloc_user(size); 414 } 415 EXPORT_SYMBOL(vmalloc_32_user); 416 417 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 418 { 419 BUG(); 420 return NULL; 421 } 422 EXPORT_SYMBOL(vmap); 423 424 void vunmap(const void *addr) 425 { 426 BUG(); 427 } 428 EXPORT_SYMBOL(vunmap); 429 430 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 431 { 432 BUG(); 433 return NULL; 434 } 435 EXPORT_SYMBOL(vm_map_ram); 436 437 void vm_unmap_ram(const void *mem, unsigned int count) 438 { 439 BUG(); 440 } 441 EXPORT_SYMBOL(vm_unmap_ram); 442 443 void vm_unmap_aliases(void) 444 { 445 } 446 EXPORT_SYMBOL_GPL(vm_unmap_aliases); 447 448 /* 449 * Implement a stub for vmalloc_sync_all() if the architecture chose not to 450 * have one. 451 */ 452 void __weak vmalloc_sync_all(void) 453 { 454 } 455 456 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 457 { 458 BUG(); 459 return NULL; 460 } 461 EXPORT_SYMBOL_GPL(alloc_vm_area); 462 463 void free_vm_area(struct vm_struct *area) 464 { 465 BUG(); 466 } 467 EXPORT_SYMBOL_GPL(free_vm_area); 468 469 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 470 struct page *page) 471 { 472 return -EINVAL; 473 } 474 EXPORT_SYMBOL(vm_insert_page); 475 476 int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 477 unsigned long num) 478 { 479 return -EINVAL; 480 } 481 EXPORT_SYMBOL(vm_map_pages); 482 483 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 484 unsigned long num) 485 { 486 return -EINVAL; 487 } 488 EXPORT_SYMBOL(vm_map_pages_zero); 489 490 /* 491 * sys_brk() for the most part doesn't need the global kernel 492 * lock, except when an application is doing something nasty 493 * like trying to un-brk an area that has already been mapped 494 * to a regular file. in this case, the unmapping will need 495 * to invoke file system routines that need the global lock. 496 */ 497 SYSCALL_DEFINE1(brk, unsigned long, brk) 498 { 499 struct mm_struct *mm = current->mm; 500 501 if (brk < mm->start_brk || brk > mm->context.end_brk) 502 return mm->brk; 503 504 if (mm->brk == brk) 505 return mm->brk; 506 507 /* 508 * Always allow shrinking brk 509 */ 510 if (brk <= mm->brk) { 511 mm->brk = brk; 512 return brk; 513 } 514 515 /* 516 * Ok, looks good - let it rip. 517 */ 518 flush_icache_range(mm->brk, brk); 519 return mm->brk = brk; 520 } 521 522 /* 523 * initialise the percpu counter for VM and region record slabs 524 */ 525 void __init mmap_init(void) 526 { 527 int ret; 528 529 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 530 VM_BUG_ON(ret); 531 vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT); 532 } 533 534 /* 535 * validate the region tree 536 * - the caller must hold the region lock 537 */ 538 #ifdef CONFIG_DEBUG_NOMMU_REGIONS 539 static noinline void validate_nommu_regions(void) 540 { 541 struct vm_region *region, *last; 542 struct rb_node *p, *lastp; 543 544 lastp = rb_first(&nommu_region_tree); 545 if (!lastp) 546 return; 547 548 last = rb_entry(lastp, struct vm_region, vm_rb); 549 BUG_ON(last->vm_end <= last->vm_start); 550 BUG_ON(last->vm_top < last->vm_end); 551 552 while ((p = rb_next(lastp))) { 553 region = rb_entry(p, struct vm_region, vm_rb); 554 last = rb_entry(lastp, struct vm_region, vm_rb); 555 556 BUG_ON(region->vm_end <= region->vm_start); 557 BUG_ON(region->vm_top < region->vm_end); 558 BUG_ON(region->vm_start < last->vm_top); 559 560 lastp = p; 561 } 562 } 563 #else 564 static void validate_nommu_regions(void) 565 { 566 } 567 #endif 568 569 /* 570 * add a region into the global tree 571 */ 572 static void add_nommu_region(struct vm_region *region) 573 { 574 struct vm_region *pregion; 575 struct rb_node **p, *parent; 576 577 validate_nommu_regions(); 578 579 parent = NULL; 580 p = &nommu_region_tree.rb_node; 581 while (*p) { 582 parent = *p; 583 pregion = rb_entry(parent, struct vm_region, vm_rb); 584 if (region->vm_start < pregion->vm_start) 585 p = &(*p)->rb_left; 586 else if (region->vm_start > pregion->vm_start) 587 p = &(*p)->rb_right; 588 else if (pregion == region) 589 return; 590 else 591 BUG(); 592 } 593 594 rb_link_node(®ion->vm_rb, parent, p); 595 rb_insert_color(®ion->vm_rb, &nommu_region_tree); 596 597 validate_nommu_regions(); 598 } 599 600 /* 601 * delete a region from the global tree 602 */ 603 static void delete_nommu_region(struct vm_region *region) 604 { 605 BUG_ON(!nommu_region_tree.rb_node); 606 607 validate_nommu_regions(); 608 rb_erase(®ion->vm_rb, &nommu_region_tree); 609 validate_nommu_regions(); 610 } 611 612 /* 613 * free a contiguous series of pages 614 */ 615 static void free_page_series(unsigned long from, unsigned long to) 616 { 617 for (; from < to; from += PAGE_SIZE) { 618 struct page *page = virt_to_page(from); 619 620 atomic_long_dec(&mmap_pages_allocated); 621 put_page(page); 622 } 623 } 624 625 /* 626 * release a reference to a region 627 * - the caller must hold the region semaphore for writing, which this releases 628 * - the region may not have been added to the tree yet, in which case vm_top 629 * will equal vm_start 630 */ 631 static void __put_nommu_region(struct vm_region *region) 632 __releases(nommu_region_sem) 633 { 634 BUG_ON(!nommu_region_tree.rb_node); 635 636 if (--region->vm_usage == 0) { 637 if (region->vm_top > region->vm_start) 638 delete_nommu_region(region); 639 up_write(&nommu_region_sem); 640 641 if (region->vm_file) 642 fput(region->vm_file); 643 644 /* IO memory and memory shared directly out of the pagecache 645 * from ramfs/tmpfs mustn't be released here */ 646 if (region->vm_flags & VM_MAPPED_COPY) 647 free_page_series(region->vm_start, region->vm_top); 648 kmem_cache_free(vm_region_jar, region); 649 } else { 650 up_write(&nommu_region_sem); 651 } 652 } 653 654 /* 655 * release a reference to a region 656 */ 657 static void put_nommu_region(struct vm_region *region) 658 { 659 down_write(&nommu_region_sem); 660 __put_nommu_region(region); 661 } 662 663 /* 664 * add a VMA into a process's mm_struct in the appropriate place in the list 665 * and tree and add to the address space's page tree also if not an anonymous 666 * page 667 * - should be called with mm->mmap_sem held writelocked 668 */ 669 static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 670 { 671 struct vm_area_struct *pvma, *prev; 672 struct address_space *mapping; 673 struct rb_node **p, *parent, *rb_prev; 674 675 BUG_ON(!vma->vm_region); 676 677 mm->map_count++; 678 vma->vm_mm = mm; 679 680 /* add the VMA to the mapping */ 681 if (vma->vm_file) { 682 mapping = vma->vm_file->f_mapping; 683 684 i_mmap_lock_write(mapping); 685 flush_dcache_mmap_lock(mapping); 686 vma_interval_tree_insert(vma, &mapping->i_mmap); 687 flush_dcache_mmap_unlock(mapping); 688 i_mmap_unlock_write(mapping); 689 } 690 691 /* add the VMA to the tree */ 692 parent = rb_prev = NULL; 693 p = &mm->mm_rb.rb_node; 694 while (*p) { 695 parent = *p; 696 pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 697 698 /* sort by: start addr, end addr, VMA struct addr in that order 699 * (the latter is necessary as we may get identical VMAs) */ 700 if (vma->vm_start < pvma->vm_start) 701 p = &(*p)->rb_left; 702 else if (vma->vm_start > pvma->vm_start) { 703 rb_prev = parent; 704 p = &(*p)->rb_right; 705 } else if (vma->vm_end < pvma->vm_end) 706 p = &(*p)->rb_left; 707 else if (vma->vm_end > pvma->vm_end) { 708 rb_prev = parent; 709 p = &(*p)->rb_right; 710 } else if (vma < pvma) 711 p = &(*p)->rb_left; 712 else if (vma > pvma) { 713 rb_prev = parent; 714 p = &(*p)->rb_right; 715 } else 716 BUG(); 717 } 718 719 rb_link_node(&vma->vm_rb, parent, p); 720 rb_insert_color(&vma->vm_rb, &mm->mm_rb); 721 722 /* add VMA to the VMA list also */ 723 prev = NULL; 724 if (rb_prev) 725 prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 726 727 __vma_link_list(mm, vma, prev, parent); 728 } 729 730 /* 731 * delete a VMA from its owning mm_struct and address space 732 */ 733 static void delete_vma_from_mm(struct vm_area_struct *vma) 734 { 735 int i; 736 struct address_space *mapping; 737 struct mm_struct *mm = vma->vm_mm; 738 struct task_struct *curr = current; 739 740 mm->map_count--; 741 for (i = 0; i < VMACACHE_SIZE; i++) { 742 /* if the vma is cached, invalidate the entire cache */ 743 if (curr->vmacache.vmas[i] == vma) { 744 vmacache_invalidate(mm); 745 break; 746 } 747 } 748 749 /* remove the VMA from the mapping */ 750 if (vma->vm_file) { 751 mapping = vma->vm_file->f_mapping; 752 753 i_mmap_lock_write(mapping); 754 flush_dcache_mmap_lock(mapping); 755 vma_interval_tree_remove(vma, &mapping->i_mmap); 756 flush_dcache_mmap_unlock(mapping); 757 i_mmap_unlock_write(mapping); 758 } 759 760 /* remove from the MM's tree and list */ 761 rb_erase(&vma->vm_rb, &mm->mm_rb); 762 763 if (vma->vm_prev) 764 vma->vm_prev->vm_next = vma->vm_next; 765 else 766 mm->mmap = vma->vm_next; 767 768 if (vma->vm_next) 769 vma->vm_next->vm_prev = vma->vm_prev; 770 } 771 772 /* 773 * destroy a VMA record 774 */ 775 static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 776 { 777 if (vma->vm_ops && vma->vm_ops->close) 778 vma->vm_ops->close(vma); 779 if (vma->vm_file) 780 fput(vma->vm_file); 781 put_nommu_region(vma->vm_region); 782 vm_area_free(vma); 783 } 784 785 /* 786 * look up the first VMA in which addr resides, NULL if none 787 * - should be called with mm->mmap_sem at least held readlocked 788 */ 789 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 790 { 791 struct vm_area_struct *vma; 792 793 /* check the cache first */ 794 vma = vmacache_find(mm, addr); 795 if (likely(vma)) 796 return vma; 797 798 /* trawl the list (there may be multiple mappings in which addr 799 * resides) */ 800 for (vma = mm->mmap; vma; vma = vma->vm_next) { 801 if (vma->vm_start > addr) 802 return NULL; 803 if (vma->vm_end > addr) { 804 vmacache_update(addr, vma); 805 return vma; 806 } 807 } 808 809 return NULL; 810 } 811 EXPORT_SYMBOL(find_vma); 812 813 /* 814 * find a VMA 815 * - we don't extend stack VMAs under NOMMU conditions 816 */ 817 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 818 { 819 return find_vma(mm, addr); 820 } 821 822 /* 823 * expand a stack to a given address 824 * - not supported under NOMMU conditions 825 */ 826 int expand_stack(struct vm_area_struct *vma, unsigned long address) 827 { 828 return -ENOMEM; 829 } 830 831 /* 832 * look up the first VMA exactly that exactly matches addr 833 * - should be called with mm->mmap_sem at least held readlocked 834 */ 835 static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 836 unsigned long addr, 837 unsigned long len) 838 { 839 struct vm_area_struct *vma; 840 unsigned long end = addr + len; 841 842 /* check the cache first */ 843 vma = vmacache_find_exact(mm, addr, end); 844 if (vma) 845 return vma; 846 847 /* trawl the list (there may be multiple mappings in which addr 848 * resides) */ 849 for (vma = mm->mmap; vma; vma = vma->vm_next) { 850 if (vma->vm_start < addr) 851 continue; 852 if (vma->vm_start > addr) 853 return NULL; 854 if (vma->vm_end == end) { 855 vmacache_update(addr, vma); 856 return vma; 857 } 858 } 859 860 return NULL; 861 } 862 863 /* 864 * determine whether a mapping should be permitted and, if so, what sort of 865 * mapping we're capable of supporting 866 */ 867 static int validate_mmap_request(struct file *file, 868 unsigned long addr, 869 unsigned long len, 870 unsigned long prot, 871 unsigned long flags, 872 unsigned long pgoff, 873 unsigned long *_capabilities) 874 { 875 unsigned long capabilities, rlen; 876 int ret; 877 878 /* do the simple checks first */ 879 if (flags & MAP_FIXED) 880 return -EINVAL; 881 882 if ((flags & MAP_TYPE) != MAP_PRIVATE && 883 (flags & MAP_TYPE) != MAP_SHARED) 884 return -EINVAL; 885 886 if (!len) 887 return -EINVAL; 888 889 /* Careful about overflows.. */ 890 rlen = PAGE_ALIGN(len); 891 if (!rlen || rlen > TASK_SIZE) 892 return -ENOMEM; 893 894 /* offset overflow? */ 895 if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 896 return -EOVERFLOW; 897 898 if (file) { 899 /* files must support mmap */ 900 if (!file->f_op->mmap) 901 return -ENODEV; 902 903 /* work out if what we've got could possibly be shared 904 * - we support chardevs that provide their own "memory" 905 * - we support files/blockdevs that are memory backed 906 */ 907 if (file->f_op->mmap_capabilities) { 908 capabilities = file->f_op->mmap_capabilities(file); 909 } else { 910 /* no explicit capabilities set, so assume some 911 * defaults */ 912 switch (file_inode(file)->i_mode & S_IFMT) { 913 case S_IFREG: 914 case S_IFBLK: 915 capabilities = NOMMU_MAP_COPY; 916 break; 917 918 case S_IFCHR: 919 capabilities = 920 NOMMU_MAP_DIRECT | 921 NOMMU_MAP_READ | 922 NOMMU_MAP_WRITE; 923 break; 924 925 default: 926 return -EINVAL; 927 } 928 } 929 930 /* eliminate any capabilities that we can't support on this 931 * device */ 932 if (!file->f_op->get_unmapped_area) 933 capabilities &= ~NOMMU_MAP_DIRECT; 934 if (!(file->f_mode & FMODE_CAN_READ)) 935 capabilities &= ~NOMMU_MAP_COPY; 936 937 /* The file shall have been opened with read permission. */ 938 if (!(file->f_mode & FMODE_READ)) 939 return -EACCES; 940 941 if (flags & MAP_SHARED) { 942 /* do checks for writing, appending and locking */ 943 if ((prot & PROT_WRITE) && 944 !(file->f_mode & FMODE_WRITE)) 945 return -EACCES; 946 947 if (IS_APPEND(file_inode(file)) && 948 (file->f_mode & FMODE_WRITE)) 949 return -EACCES; 950 951 if (locks_verify_locked(file)) 952 return -EAGAIN; 953 954 if (!(capabilities & NOMMU_MAP_DIRECT)) 955 return -ENODEV; 956 957 /* we mustn't privatise shared mappings */ 958 capabilities &= ~NOMMU_MAP_COPY; 959 } else { 960 /* we're going to read the file into private memory we 961 * allocate */ 962 if (!(capabilities & NOMMU_MAP_COPY)) 963 return -ENODEV; 964 965 /* we don't permit a private writable mapping to be 966 * shared with the backing device */ 967 if (prot & PROT_WRITE) 968 capabilities &= ~NOMMU_MAP_DIRECT; 969 } 970 971 if (capabilities & NOMMU_MAP_DIRECT) { 972 if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || 973 ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || 974 ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) 975 ) { 976 capabilities &= ~NOMMU_MAP_DIRECT; 977 if (flags & MAP_SHARED) { 978 pr_warn("MAP_SHARED not completely supported on !MMU\n"); 979 return -EINVAL; 980 } 981 } 982 } 983 984 /* handle executable mappings and implied executable 985 * mappings */ 986 if (path_noexec(&file->f_path)) { 987 if (prot & PROT_EXEC) 988 return -EPERM; 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 & NOMMU_MAP_EXEC) 993 prot |= PROT_EXEC; 994 } 995 } else if ((prot & PROT_READ) && 996 (prot & PROT_EXEC) && 997 !(capabilities & NOMMU_MAP_EXEC) 998 ) { 999 /* backing file is not executable, try to copy */ 1000 capabilities &= ~NOMMU_MAP_DIRECT; 1001 } 1002 } else { 1003 /* anonymous mappings are always memory backed and can be 1004 * privately mapped 1005 */ 1006 capabilities = NOMMU_MAP_COPY; 1007 1008 /* handle PROT_EXEC implication by PROT_READ */ 1009 if ((prot & PROT_READ) && 1010 (current->personality & READ_IMPLIES_EXEC)) 1011 prot |= PROT_EXEC; 1012 } 1013 1014 /* allow the security API to have its say */ 1015 ret = security_mmap_addr(addr); 1016 if (ret < 0) 1017 return ret; 1018 1019 /* looks okay */ 1020 *_capabilities = capabilities; 1021 return 0; 1022 } 1023 1024 /* 1025 * we've determined that we can make the mapping, now translate what we 1026 * now know into VMA flags 1027 */ 1028 static unsigned long determine_vm_flags(struct file *file, 1029 unsigned long prot, 1030 unsigned long flags, 1031 unsigned long capabilities) 1032 { 1033 unsigned long vm_flags; 1034 1035 vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags); 1036 /* vm_flags |= mm->def_flags; */ 1037 1038 if (!(capabilities & NOMMU_MAP_DIRECT)) { 1039 /* attempt to share read-only copies of mapped file chunks */ 1040 vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 1041 if (file && !(prot & PROT_WRITE)) 1042 vm_flags |= VM_MAYSHARE; 1043 } else { 1044 /* overlay a shareable mapping on the backing device or inode 1045 * if possible - used for chardevs, ramfs/tmpfs/shmfs and 1046 * romfs/cramfs */ 1047 vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); 1048 if (flags & MAP_SHARED) 1049 vm_flags |= VM_SHARED; 1050 } 1051 1052 /* refuse to let anyone share private mappings with this process if 1053 * it's being traced - otherwise breakpoints set in it may interfere 1054 * with another untraced process 1055 */ 1056 if ((flags & MAP_PRIVATE) && current->ptrace) 1057 vm_flags &= ~VM_MAYSHARE; 1058 1059 return vm_flags; 1060 } 1061 1062 /* 1063 * set up a shared mapping on a file (the driver or filesystem provides and 1064 * pins the storage) 1065 */ 1066 static int do_mmap_shared_file(struct vm_area_struct *vma) 1067 { 1068 int ret; 1069 1070 ret = call_mmap(vma->vm_file, vma); 1071 if (ret == 0) { 1072 vma->vm_region->vm_top = vma->vm_region->vm_end; 1073 return 0; 1074 } 1075 if (ret != -ENOSYS) 1076 return ret; 1077 1078 /* getting -ENOSYS indicates that direct mmap isn't possible (as 1079 * opposed to tried but failed) so we can only give a suitable error as 1080 * it's not possible to make a private copy if MAP_SHARED was given */ 1081 return -ENODEV; 1082 } 1083 1084 /* 1085 * set up a private mapping or an anonymous shared mapping 1086 */ 1087 static int do_mmap_private(struct vm_area_struct *vma, 1088 struct vm_region *region, 1089 unsigned long len, 1090 unsigned long capabilities) 1091 { 1092 unsigned long total, point; 1093 void *base; 1094 int ret, order; 1095 1096 /* invoke the file's mapping function so that it can keep track of 1097 * shared mappings on devices or memory 1098 * - VM_MAYSHARE will be set if it may attempt to share 1099 */ 1100 if (capabilities & NOMMU_MAP_DIRECT) { 1101 ret = call_mmap(vma->vm_file, vma); 1102 if (ret == 0) { 1103 /* shouldn't return success if we're not sharing */ 1104 BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1105 vma->vm_region->vm_top = vma->vm_region->vm_end; 1106 return 0; 1107 } 1108 if (ret != -ENOSYS) 1109 return ret; 1110 1111 /* getting an ENOSYS error indicates that direct mmap isn't 1112 * possible (as opposed to tried but failed) so we'll try to 1113 * make a private copy of the data and map that instead */ 1114 } 1115 1116 1117 /* allocate some memory to hold the mapping 1118 * - note that this may not return a page-aligned address if the object 1119 * we're allocating is smaller than a page 1120 */ 1121 order = get_order(len); 1122 total = 1 << order; 1123 point = len >> PAGE_SHIFT; 1124 1125 /* we don't want to allocate a power-of-2 sized page set */ 1126 if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) 1127 total = point; 1128 1129 base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL); 1130 if (!base) 1131 goto enomem; 1132 1133 atomic_long_add(total, &mmap_pages_allocated); 1134 1135 region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 1136 region->vm_start = (unsigned long) base; 1137 region->vm_end = region->vm_start + len; 1138 region->vm_top = region->vm_start + (total << PAGE_SHIFT); 1139 1140 vma->vm_start = region->vm_start; 1141 vma->vm_end = region->vm_start + len; 1142 1143 if (vma->vm_file) { 1144 /* read the contents of a file into the copy */ 1145 loff_t fpos; 1146 1147 fpos = vma->vm_pgoff; 1148 fpos <<= PAGE_SHIFT; 1149 1150 ret = kernel_read(vma->vm_file, base, len, &fpos); 1151 if (ret < 0) 1152 goto error_free; 1153 1154 /* clear the last little bit */ 1155 if (ret < len) 1156 memset(base + ret, 0, len - ret); 1157 1158 } else { 1159 vma_set_anonymous(vma); 1160 } 1161 1162 return 0; 1163 1164 error_free: 1165 free_page_series(region->vm_start, region->vm_top); 1166 region->vm_start = vma->vm_start = 0; 1167 region->vm_end = vma->vm_end = 0; 1168 region->vm_top = 0; 1169 return ret; 1170 1171 enomem: 1172 pr_err("Allocation of length %lu from process %d (%s) failed\n", 1173 len, current->pid, current->comm); 1174 show_free_areas(0, NULL); 1175 return -ENOMEM; 1176 } 1177 1178 /* 1179 * handle mapping creation for uClinux 1180 */ 1181 unsigned long do_mmap(struct file *file, 1182 unsigned long addr, 1183 unsigned long len, 1184 unsigned long prot, 1185 unsigned long flags, 1186 vm_flags_t vm_flags, 1187 unsigned long pgoff, 1188 unsigned long *populate, 1189 struct list_head *uf) 1190 { 1191 struct vm_area_struct *vma; 1192 struct vm_region *region; 1193 struct rb_node *rb; 1194 unsigned long capabilities, result; 1195 int ret; 1196 1197 *populate = 0; 1198 1199 /* decide whether we should attempt the mapping, and if so what sort of 1200 * mapping */ 1201 ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 1202 &capabilities); 1203 if (ret < 0) 1204 return ret; 1205 1206 /* we ignore the address hint */ 1207 addr = 0; 1208 len = PAGE_ALIGN(len); 1209 1210 /* we've determined that we can make the mapping, now translate what we 1211 * now know into VMA flags */ 1212 vm_flags |= determine_vm_flags(file, prot, flags, capabilities); 1213 1214 /* we're going to need to record the mapping */ 1215 region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 1216 if (!region) 1217 goto error_getting_region; 1218 1219 vma = vm_area_alloc(current->mm); 1220 if (!vma) 1221 goto error_getting_vma; 1222 1223 region->vm_usage = 1; 1224 region->vm_flags = vm_flags; 1225 region->vm_pgoff = pgoff; 1226 1227 vma->vm_flags = vm_flags; 1228 vma->vm_pgoff = pgoff; 1229 1230 if (file) { 1231 region->vm_file = get_file(file); 1232 vma->vm_file = get_file(file); 1233 } 1234 1235 down_write(&nommu_region_sem); 1236 1237 /* if we want to share, we need to check for regions created by other 1238 * mmap() calls that overlap with our proposed mapping 1239 * - we can only share with a superset match on most regular files 1240 * - shared mappings on character devices and memory backed files are 1241 * permitted to overlap inexactly as far as we are concerned for in 1242 * these cases, sharing is handled in the driver or filesystem rather 1243 * than here 1244 */ 1245 if (vm_flags & VM_MAYSHARE) { 1246 struct vm_region *pregion; 1247 unsigned long pglen, rpglen, pgend, rpgend, start; 1248 1249 pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1250 pgend = pgoff + pglen; 1251 1252 for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 1253 pregion = rb_entry(rb, struct vm_region, vm_rb); 1254 1255 if (!(pregion->vm_flags & VM_MAYSHARE)) 1256 continue; 1257 1258 /* search for overlapping mappings on the same file */ 1259 if (file_inode(pregion->vm_file) != 1260 file_inode(file)) 1261 continue; 1262 1263 if (pregion->vm_pgoff >= pgend) 1264 continue; 1265 1266 rpglen = pregion->vm_end - pregion->vm_start; 1267 rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 1268 rpgend = pregion->vm_pgoff + rpglen; 1269 if (pgoff >= rpgend) 1270 continue; 1271 1272 /* handle inexactly overlapping matches between 1273 * mappings */ 1274 if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 1275 !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 1276 /* new mapping is not a subset of the region */ 1277 if (!(capabilities & NOMMU_MAP_DIRECT)) 1278 goto sharing_violation; 1279 continue; 1280 } 1281 1282 /* we've found a region we can share */ 1283 pregion->vm_usage++; 1284 vma->vm_region = pregion; 1285 start = pregion->vm_start; 1286 start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 1287 vma->vm_start = start; 1288 vma->vm_end = start + len; 1289 1290 if (pregion->vm_flags & VM_MAPPED_COPY) 1291 vma->vm_flags |= VM_MAPPED_COPY; 1292 else { 1293 ret = do_mmap_shared_file(vma); 1294 if (ret < 0) { 1295 vma->vm_region = NULL; 1296 vma->vm_start = 0; 1297 vma->vm_end = 0; 1298 pregion->vm_usage--; 1299 pregion = NULL; 1300 goto error_just_free; 1301 } 1302 } 1303 fput(region->vm_file); 1304 kmem_cache_free(vm_region_jar, region); 1305 region = pregion; 1306 result = start; 1307 goto share; 1308 } 1309 1310 /* obtain the address at which to make a shared mapping 1311 * - this is the hook for quasi-memory character devices to 1312 * tell us the location of a shared mapping 1313 */ 1314 if (capabilities & NOMMU_MAP_DIRECT) { 1315 addr = file->f_op->get_unmapped_area(file, addr, len, 1316 pgoff, flags); 1317 if (IS_ERR_VALUE(addr)) { 1318 ret = addr; 1319 if (ret != -ENOSYS) 1320 goto error_just_free; 1321 1322 /* the driver refused to tell us where to site 1323 * the mapping so we'll have to attempt to copy 1324 * it */ 1325 ret = -ENODEV; 1326 if (!(capabilities & NOMMU_MAP_COPY)) 1327 goto error_just_free; 1328 1329 capabilities &= ~NOMMU_MAP_DIRECT; 1330 } else { 1331 vma->vm_start = region->vm_start = addr; 1332 vma->vm_end = region->vm_end = addr + len; 1333 } 1334 } 1335 } 1336 1337 vma->vm_region = region; 1338 1339 /* set up the mapping 1340 * - the region is filled in if NOMMU_MAP_DIRECT is still set 1341 */ 1342 if (file && vma->vm_flags & VM_SHARED) 1343 ret = do_mmap_shared_file(vma); 1344 else 1345 ret = do_mmap_private(vma, region, len, capabilities); 1346 if (ret < 0) 1347 goto error_just_free; 1348 add_nommu_region(region); 1349 1350 /* clear anonymous mappings that don't ask for uninitialized data */ 1351 if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1352 memset((void *)region->vm_start, 0, 1353 region->vm_end - region->vm_start); 1354 1355 /* okay... we have a mapping; now we have to register it */ 1356 result = vma->vm_start; 1357 1358 current->mm->total_vm += len >> PAGE_SHIFT; 1359 1360 share: 1361 add_vma_to_mm(current->mm, vma); 1362 1363 /* we flush the region from the icache only when the first executable 1364 * mapping of it is made */ 1365 if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1366 flush_icache_range(region->vm_start, region->vm_end); 1367 region->vm_icache_flushed = true; 1368 } 1369 1370 up_write(&nommu_region_sem); 1371 1372 return result; 1373 1374 error_just_free: 1375 up_write(&nommu_region_sem); 1376 error: 1377 if (region->vm_file) 1378 fput(region->vm_file); 1379 kmem_cache_free(vm_region_jar, region); 1380 if (vma->vm_file) 1381 fput(vma->vm_file); 1382 vm_area_free(vma); 1383 return ret; 1384 1385 sharing_violation: 1386 up_write(&nommu_region_sem); 1387 pr_warn("Attempt to share mismatched mappings\n"); 1388 ret = -EINVAL; 1389 goto error; 1390 1391 error_getting_vma: 1392 kmem_cache_free(vm_region_jar, region); 1393 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n", 1394 len, current->pid); 1395 show_free_areas(0, NULL); 1396 return -ENOMEM; 1397 1398 error_getting_region: 1399 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n", 1400 len, current->pid); 1401 show_free_areas(0, NULL); 1402 return -ENOMEM; 1403 } 1404 1405 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1406 unsigned long prot, unsigned long flags, 1407 unsigned long fd, unsigned long pgoff) 1408 { 1409 struct file *file = NULL; 1410 unsigned long retval = -EBADF; 1411 1412 audit_mmap_fd(fd, flags); 1413 if (!(flags & MAP_ANONYMOUS)) { 1414 file = fget(fd); 1415 if (!file) 1416 goto out; 1417 } 1418 1419 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 1420 1421 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1422 1423 if (file) 1424 fput(file); 1425 out: 1426 return retval; 1427 } 1428 1429 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1430 unsigned long, prot, unsigned long, flags, 1431 unsigned long, fd, unsigned long, pgoff) 1432 { 1433 return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1434 } 1435 1436 #ifdef __ARCH_WANT_SYS_OLD_MMAP 1437 struct mmap_arg_struct { 1438 unsigned long addr; 1439 unsigned long len; 1440 unsigned long prot; 1441 unsigned long flags; 1442 unsigned long fd; 1443 unsigned long offset; 1444 }; 1445 1446 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1447 { 1448 struct mmap_arg_struct a; 1449 1450 if (copy_from_user(&a, arg, sizeof(a))) 1451 return -EFAULT; 1452 if (offset_in_page(a.offset)) 1453 return -EINVAL; 1454 1455 return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1456 a.offset >> PAGE_SHIFT); 1457 } 1458 #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1459 1460 /* 1461 * split a vma into two pieces at address 'addr', a new vma is allocated either 1462 * for the first part or the tail. 1463 */ 1464 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 1465 unsigned long addr, int new_below) 1466 { 1467 struct vm_area_struct *new; 1468 struct vm_region *region; 1469 unsigned long npages; 1470 1471 /* we're only permitted to split anonymous regions (these should have 1472 * only a single usage on the region) */ 1473 if (vma->vm_file) 1474 return -ENOMEM; 1475 1476 if (mm->map_count >= sysctl_max_map_count) 1477 return -ENOMEM; 1478 1479 region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 1480 if (!region) 1481 return -ENOMEM; 1482 1483 new = vm_area_dup(vma); 1484 if (!new) { 1485 kmem_cache_free(vm_region_jar, region); 1486 return -ENOMEM; 1487 } 1488 1489 /* most fields are the same, copy all, and then fixup */ 1490 *region = *vma->vm_region; 1491 new->vm_region = region; 1492 1493 npages = (addr - vma->vm_start) >> PAGE_SHIFT; 1494 1495 if (new_below) { 1496 region->vm_top = region->vm_end = new->vm_end = addr; 1497 } else { 1498 region->vm_start = new->vm_start = addr; 1499 region->vm_pgoff = new->vm_pgoff += npages; 1500 } 1501 1502 if (new->vm_ops && new->vm_ops->open) 1503 new->vm_ops->open(new); 1504 1505 delete_vma_from_mm(vma); 1506 down_write(&nommu_region_sem); 1507 delete_nommu_region(vma->vm_region); 1508 if (new_below) { 1509 vma->vm_region->vm_start = vma->vm_start = addr; 1510 vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 1511 } else { 1512 vma->vm_region->vm_end = vma->vm_end = addr; 1513 vma->vm_region->vm_top = addr; 1514 } 1515 add_nommu_region(vma->vm_region); 1516 add_nommu_region(new->vm_region); 1517 up_write(&nommu_region_sem); 1518 add_vma_to_mm(mm, vma); 1519 add_vma_to_mm(mm, new); 1520 return 0; 1521 } 1522 1523 /* 1524 * shrink a VMA by removing the specified chunk from either the beginning or 1525 * the end 1526 */ 1527 static int shrink_vma(struct mm_struct *mm, 1528 struct vm_area_struct *vma, 1529 unsigned long from, unsigned long to) 1530 { 1531 struct vm_region *region; 1532 1533 /* adjust the VMA's pointers, which may reposition it in the MM's tree 1534 * and list */ 1535 delete_vma_from_mm(vma); 1536 if (from > vma->vm_start) 1537 vma->vm_end = from; 1538 else 1539 vma->vm_start = to; 1540 add_vma_to_mm(mm, vma); 1541 1542 /* cut the backing region down to size */ 1543 region = vma->vm_region; 1544 BUG_ON(region->vm_usage != 1); 1545 1546 down_write(&nommu_region_sem); 1547 delete_nommu_region(region); 1548 if (from > region->vm_start) { 1549 to = region->vm_top; 1550 region->vm_top = region->vm_end = from; 1551 } else { 1552 region->vm_start = to; 1553 } 1554 add_nommu_region(region); 1555 up_write(&nommu_region_sem); 1556 1557 free_page_series(from, to); 1558 return 0; 1559 } 1560 1561 /* 1562 * release a mapping 1563 * - under NOMMU conditions the chunk to be unmapped must be backed by a single 1564 * VMA, though it need not cover the whole VMA 1565 */ 1566 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf) 1567 { 1568 struct vm_area_struct *vma; 1569 unsigned long end; 1570 int ret; 1571 1572 len = PAGE_ALIGN(len); 1573 if (len == 0) 1574 return -EINVAL; 1575 1576 end = start + len; 1577 1578 /* find the first potentially overlapping VMA */ 1579 vma = find_vma(mm, start); 1580 if (!vma) { 1581 static int limit; 1582 if (limit < 5) { 1583 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n", 1584 current->pid, current->comm, 1585 start, start + len - 1); 1586 limit++; 1587 } 1588 return -EINVAL; 1589 } 1590 1591 /* we're allowed to split an anonymous VMA but not a file-backed one */ 1592 if (vma->vm_file) { 1593 do { 1594 if (start > vma->vm_start) 1595 return -EINVAL; 1596 if (end == vma->vm_end) 1597 goto erase_whole_vma; 1598 vma = vma->vm_next; 1599 } while (vma); 1600 return -EINVAL; 1601 } else { 1602 /* the chunk must be a subset of the VMA found */ 1603 if (start == vma->vm_start && end == vma->vm_end) 1604 goto erase_whole_vma; 1605 if (start < vma->vm_start || end > vma->vm_end) 1606 return -EINVAL; 1607 if (offset_in_page(start)) 1608 return -EINVAL; 1609 if (end != vma->vm_end && offset_in_page(end)) 1610 return -EINVAL; 1611 if (start != vma->vm_start && end != vma->vm_end) { 1612 ret = split_vma(mm, vma, start, 1); 1613 if (ret < 0) 1614 return ret; 1615 } 1616 return shrink_vma(mm, vma, start, end); 1617 } 1618 1619 erase_whole_vma: 1620 delete_vma_from_mm(vma); 1621 delete_vma(mm, vma); 1622 return 0; 1623 } 1624 EXPORT_SYMBOL(do_munmap); 1625 1626 int vm_munmap(unsigned long addr, size_t len) 1627 { 1628 struct mm_struct *mm = current->mm; 1629 int ret; 1630 1631 down_write(&mm->mmap_sem); 1632 ret = do_munmap(mm, addr, len, NULL); 1633 up_write(&mm->mmap_sem); 1634 return ret; 1635 } 1636 EXPORT_SYMBOL(vm_munmap); 1637 1638 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1639 { 1640 return vm_munmap(addr, len); 1641 } 1642 1643 /* 1644 * release all the mappings made in a process's VM space 1645 */ 1646 void exit_mmap(struct mm_struct *mm) 1647 { 1648 struct vm_area_struct *vma; 1649 1650 if (!mm) 1651 return; 1652 1653 mm->total_vm = 0; 1654 1655 while ((vma = mm->mmap)) { 1656 mm->mmap = vma->vm_next; 1657 delete_vma_from_mm(vma); 1658 delete_vma(mm, vma); 1659 cond_resched(); 1660 } 1661 } 1662 1663 int vm_brk(unsigned long addr, unsigned long len) 1664 { 1665 return -ENOMEM; 1666 } 1667 1668 /* 1669 * expand (or shrink) an existing mapping, potentially moving it at the same 1670 * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 1671 * 1672 * under NOMMU conditions, we only permit changing a mapping's size, and only 1673 * as long as it stays within the region allocated by do_mmap_private() and the 1674 * block is not shareable 1675 * 1676 * MREMAP_FIXED is not supported under NOMMU conditions 1677 */ 1678 static unsigned long do_mremap(unsigned long addr, 1679 unsigned long old_len, unsigned long new_len, 1680 unsigned long flags, unsigned long new_addr) 1681 { 1682 struct vm_area_struct *vma; 1683 1684 /* insanity checks first */ 1685 old_len = PAGE_ALIGN(old_len); 1686 new_len = PAGE_ALIGN(new_len); 1687 if (old_len == 0 || new_len == 0) 1688 return (unsigned long) -EINVAL; 1689 1690 if (offset_in_page(addr)) 1691 return -EINVAL; 1692 1693 if (flags & MREMAP_FIXED && new_addr != addr) 1694 return (unsigned long) -EINVAL; 1695 1696 vma = find_vma_exact(current->mm, addr, old_len); 1697 if (!vma) 1698 return (unsigned long) -EINVAL; 1699 1700 if (vma->vm_end != vma->vm_start + old_len) 1701 return (unsigned long) -EFAULT; 1702 1703 if (vma->vm_flags & VM_MAYSHARE) 1704 return (unsigned long) -EPERM; 1705 1706 if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 1707 return (unsigned long) -ENOMEM; 1708 1709 /* all checks complete - do it */ 1710 vma->vm_end = vma->vm_start + new_len; 1711 return vma->vm_start; 1712 } 1713 1714 SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 1715 unsigned long, new_len, unsigned long, flags, 1716 unsigned long, new_addr) 1717 { 1718 unsigned long ret; 1719 1720 down_write(¤t->mm->mmap_sem); 1721 ret = do_mremap(addr, old_len, new_len, flags, new_addr); 1722 up_write(¤t->mm->mmap_sem); 1723 return ret; 1724 } 1725 1726 struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1727 unsigned int foll_flags) 1728 { 1729 return NULL; 1730 } 1731 1732 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 1733 unsigned long pfn, unsigned long size, pgprot_t prot) 1734 { 1735 if (addr != (pfn << PAGE_SHIFT)) 1736 return -EINVAL; 1737 1738 vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 1739 return 0; 1740 } 1741 EXPORT_SYMBOL(remap_pfn_range); 1742 1743 int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 1744 { 1745 unsigned long pfn = start >> PAGE_SHIFT; 1746 unsigned long vm_len = vma->vm_end - vma->vm_start; 1747 1748 pfn += vma->vm_pgoff; 1749 return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 1750 } 1751 EXPORT_SYMBOL(vm_iomap_memory); 1752 1753 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1754 unsigned long pgoff) 1755 { 1756 unsigned int size = vma->vm_end - vma->vm_start; 1757 1758 if (!(vma->vm_flags & VM_USERMAP)) 1759 return -EINVAL; 1760 1761 vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1762 vma->vm_end = vma->vm_start + size; 1763 1764 return 0; 1765 } 1766 EXPORT_SYMBOL(remap_vmalloc_range); 1767 1768 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 1769 unsigned long len, unsigned long pgoff, unsigned long flags) 1770 { 1771 return -ENOMEM; 1772 } 1773 1774 vm_fault_t filemap_fault(struct vm_fault *vmf) 1775 { 1776 BUG(); 1777 return 0; 1778 } 1779 EXPORT_SYMBOL(filemap_fault); 1780 1781 void filemap_map_pages(struct vm_fault *vmf, 1782 pgoff_t start_pgoff, pgoff_t end_pgoff) 1783 { 1784 BUG(); 1785 } 1786 EXPORT_SYMBOL(filemap_map_pages); 1787 1788 int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 1789 unsigned long addr, void *buf, int len, unsigned int gup_flags) 1790 { 1791 struct vm_area_struct *vma; 1792 int write = gup_flags & FOLL_WRITE; 1793 1794 down_read(&mm->mmap_sem); 1795 1796 /* the access must start within one of the target process's mappings */ 1797 vma = find_vma(mm, addr); 1798 if (vma) { 1799 /* don't overrun this mapping */ 1800 if (addr + len >= vma->vm_end) 1801 len = vma->vm_end - addr; 1802 1803 /* only read or write mappings where it is permitted */ 1804 if (write && vma->vm_flags & VM_MAYWRITE) 1805 copy_to_user_page(vma, NULL, addr, 1806 (void *) addr, buf, len); 1807 else if (!write && vma->vm_flags & VM_MAYREAD) 1808 copy_from_user_page(vma, NULL, addr, 1809 buf, (void *) addr, len); 1810 else 1811 len = 0; 1812 } else { 1813 len = 0; 1814 } 1815 1816 up_read(&mm->mmap_sem); 1817 1818 return len; 1819 } 1820 1821 /** 1822 * access_remote_vm - access another process' address space 1823 * @mm: the mm_struct of the target address space 1824 * @addr: start address to access 1825 * @buf: source or destination buffer 1826 * @len: number of bytes to transfer 1827 * @gup_flags: flags modifying lookup behaviour 1828 * 1829 * The caller must hold a reference on @mm. 1830 */ 1831 int access_remote_vm(struct mm_struct *mm, unsigned long addr, 1832 void *buf, int len, unsigned int gup_flags) 1833 { 1834 return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); 1835 } 1836 1837 /* 1838 * Access another process' address space. 1839 * - source/target buffer must be kernel space 1840 */ 1841 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, 1842 unsigned int gup_flags) 1843 { 1844 struct mm_struct *mm; 1845 1846 if (addr + len < addr) 1847 return 0; 1848 1849 mm = get_task_mm(tsk); 1850 if (!mm) 1851 return 0; 1852 1853 len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); 1854 1855 mmput(mm); 1856 return len; 1857 } 1858 EXPORT_SYMBOL_GPL(access_process_vm); 1859 1860 /** 1861 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 1862 * @inode: The inode to check 1863 * @size: The current filesize of the inode 1864 * @newsize: The proposed filesize of the inode 1865 * 1866 * Check the shared mappings on an inode on behalf of a shrinking truncate to 1867 * make sure that that any outstanding VMAs aren't broken and then shrink the 1868 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 1869 * automatically grant mappings that are too large. 1870 */ 1871 int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 1872 size_t newsize) 1873 { 1874 struct vm_area_struct *vma; 1875 struct vm_region *region; 1876 pgoff_t low, high; 1877 size_t r_size, r_top; 1878 1879 low = newsize >> PAGE_SHIFT; 1880 high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 1881 1882 down_write(&nommu_region_sem); 1883 i_mmap_lock_read(inode->i_mapping); 1884 1885 /* search for VMAs that fall within the dead zone */ 1886 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { 1887 /* found one - only interested if it's shared out of the page 1888 * cache */ 1889 if (vma->vm_flags & VM_SHARED) { 1890 i_mmap_unlock_read(inode->i_mapping); 1891 up_write(&nommu_region_sem); 1892 return -ETXTBSY; /* not quite true, but near enough */ 1893 } 1894 } 1895 1896 /* reduce any regions that overlap the dead zone - if in existence, 1897 * these will be pointed to by VMAs that don't overlap the dead zone 1898 * 1899 * we don't check for any regions that start beyond the EOF as there 1900 * shouldn't be any 1901 */ 1902 vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { 1903 if (!(vma->vm_flags & VM_SHARED)) 1904 continue; 1905 1906 region = vma->vm_region; 1907 r_size = region->vm_top - region->vm_start; 1908 r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 1909 1910 if (r_top > newsize) { 1911 region->vm_top -= r_top - newsize; 1912 if (region->vm_end > region->vm_top) 1913 region->vm_end = region->vm_top; 1914 } 1915 } 1916 1917 i_mmap_unlock_read(inode->i_mapping); 1918 up_write(&nommu_region_sem); 1919 return 0; 1920 } 1921 1922 /* 1923 * Initialise sysctl_user_reserve_kbytes. 1924 * 1925 * This is intended to prevent a user from starting a single memory hogging 1926 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 1927 * mode. 1928 * 1929 * The default value is min(3% of free memory, 128MB) 1930 * 128MB is enough to recover with sshd/login, bash, and top/kill. 1931 */ 1932 static int __meminit init_user_reserve(void) 1933 { 1934 unsigned long free_kbytes; 1935 1936 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 1937 1938 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 1939 return 0; 1940 } 1941 subsys_initcall(init_user_reserve); 1942 1943 /* 1944 * Initialise sysctl_admin_reserve_kbytes. 1945 * 1946 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 1947 * to log in and kill a memory hogging process. 1948 * 1949 * Systems with more than 256MB will reserve 8MB, enough to recover 1950 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 1951 * only reserve 3% of free pages by default. 1952 */ 1953 static int __meminit init_admin_reserve(void) 1954 { 1955 unsigned long free_kbytes; 1956 1957 free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 1958 1959 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 1960 return 0; 1961 } 1962 subsys_initcall(init_admin_reserve); 1963